executorimpl.c 202.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
 *
 * This program is free software: you can use, redistribute, and/or modify
 * it under the terms of the GNU Affero General Public License, version 3
 * or later ("AGPL"), as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.
 *
 * You should have received a copy of the GNU Affero General Public License
 * along with this program. If not, see <http://www.gnu.org/licenses/>.
 */
15

H
Haojun Liao 已提交
16 17
#include "filter.h"
#include "function.h"
18 19
#include "functionMgt.h"
#include "os.h"
H
Haojun Liao 已提交
20
#include "querynodes.h"
21
#include "tfill.h"
dengyihao's avatar
dengyihao 已提交
22
#include "tname.h"
23

H
Haojun Liao 已提交
24
#include "tdatablock.h"
25
#include "tglobal.h"
H
Haojun Liao 已提交
26
#include "tmsg.h"
H
Haojun Liao 已提交
27
#include "tsort.h"
28
#include "ttime.h"
H
Haojun Liao 已提交
29

30
#include "executorimpl.h"
31
#include "query.h"
32 33
#include "tcompare.h"
#include "tcompression.h"
H
Haojun Liao 已提交
34
#include "thash.h"
35
#include "ttypes.h"
dengyihao's avatar
dengyihao 已提交
36
#include "vnode.h"
37

H
Haojun Liao 已提交
38
#define IS_MAIN_SCAN(runtime)          ((runtime)->scanFlag == MAIN_SCAN)
39 40
#define IS_REVERSE_SCAN(runtime)       ((runtime)->scanFlag == REVERSE_SCAN)
#define IS_REPEAT_SCAN(runtime)        ((runtime)->scanFlag == REPEAT_SCAN)
L
Liu Jicong 已提交
41
#define SET_MAIN_SCAN_FLAG(runtime)    ((runtime)->scanFlag = MAIN_SCAN)
42 43
#define SET_REVERSE_SCAN_FLAG(runtime) ((runtime)->scanFlag = REVERSE_SCAN)

L
Liu Jicong 已提交
44 45
#define SDATA_BLOCK_INITIALIZER \
  (SDataBlockInfo) { {0}, 0 }
46 47 48 49

#define GET_FORWARD_DIRECTION_FACTOR(ord) (((ord) == TSDB_ORDER_ASC) ? QUERY_ASC_FORWARD_STEP : QUERY_DESC_FORWARD_STEP)

enum {
L
Liu Jicong 已提交
50 51
  TS_JOIN_TS_EQUAL = 0,
  TS_JOIN_TS_NOT_EQUALS = 1,
52 53 54 55 56
  TS_JOIN_TAG_NOT_EQUALS = 2,
};

#if 0
static UNUSED_FUNC void *u_malloc (size_t __size) {
wafwerar's avatar
wafwerar 已提交
57
  uint32_t v = taosRand();
58 59 60 61

  if (v % 1000 <= 0) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
62
    return taosMemoryMalloc(__size);
63 64 65 66
  }
}

static UNUSED_FUNC void* u_calloc(size_t num, size_t __size) {
wafwerar's avatar
wafwerar 已提交
67
  uint32_t v = taosRand();
68 69 70
  if (v % 1000 <= 0) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
71
    return taosMemoryCalloc(num, __size);
72 73 74 75
  }
}

static UNUSED_FUNC void* u_realloc(void* p, size_t __size) {
wafwerar's avatar
wafwerar 已提交
76
  uint32_t v = taosRand();
77 78 79
  if (v % 5 <= 1) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
80
    return taosMemoryRealloc(p, __size);
81 82 83 84 85 86 87 88 89 90 91 92
  }
}

#define calloc  u_calloc
#define malloc  u_malloc
#define realloc u_realloc
#endif

#define CLEAR_QUERY_STATUS(q, st)   ((q)->status &= (~(st)))
#define GET_NUM_OF_TABLEGROUP(q)    taosArrayGetSize((q)->tableqinfoGroupInfo.pGroupList)
#define QUERY_IS_INTERVAL_QUERY(_q) ((_q)->interval.interval > 0)

L
Liu Jicong 已提交
93 94 95
int32_t getMaximumIdleDurationSec() { return tsShellActivityTimer * 2; }

static int32_t getExprFunctionId(SExprInfo* pExprInfo) {
96
  assert(pExprInfo != NULL && pExprInfo->pExpr != NULL && pExprInfo->pExpr->nodeType == TEXPR_UNARYEXPR_NODE);
97
  return 0;
98 99 100
}

static void doSetTagValueToResultBuf(char* output, const char* val, int16_t type, int16_t bytes);
L
Liu Jicong 已提交
101
static bool functionNeedToExecute(SqlFunctionCtx* pCtx);
102

103
static void setBlockStatisInfo(SqlFunctionCtx* pCtx, SExprInfo* pExpr, SSDataBlock* pSDataBlock);
104

L
Liu Jicong 已提交
105
static void destroyTableQueryInfoImpl(STableQueryInfo* pTableQueryInfo);
106 107 108

static SColumnInfo* extractColumnFilterInfo(SExprInfo* pExpr, int32_t numOfOutput, int32_t* numOfFilterCols);

L
Liu Jicong 已提交
109 110
static int32_t setTimestampListJoinInfo(STaskRuntimeEnv* pRuntimeEnv, SVariant* pTag, STableQueryInfo* pTableQueryInfo);
static void    releaseQueryBuf(size_t numOfTables);
111

H
Haojun Liao 已提交
112
static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr);
113 114 115 116 117 118

static void destroySFillOperatorInfo(void* param, int32_t numOfOutput);
static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput);
static void destroyTagScanOperatorInfo(void* param, int32_t numOfOutput);
static void destroyOrderOperatorInfo(void* param, int32_t numOfOutput);
static void destroyAggOperatorInfo(void* param, int32_t numOfOutput);
X
Xiaoyu Wang 已提交
119

H
Haojun Liao 已提交
120
static void destroyIntervalOperatorInfo(void* param, int32_t numOfOutput);
H
Haojun Liao 已提交
121 122
static void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput);

123
static void destroyOperatorInfo(SOperatorInfo* pOperator);
124
static void destroySysTableScannerOperatorInfo(void* param, int32_t numOfOutput);
125

126
void doSetOperatorCompleted(SOperatorInfo* pOperator) {
127
  pOperator->status = OP_EXEC_DONE;
H
Haojun Liao 已提交
128
  if (pOperator->pTaskInfo != NULL) {
129
    setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
130 131
  }
}
132

H
Haojun Liao 已提交
133
int32_t operatorDummyOpenFn(SOperatorInfo* pOperator) {
134
  OPTR_SET_OPENED(pOperator);
135
  pOperator->cost.openCost = 0;
H
Haojun Liao 已提交
136
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
137 138
}

139
SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t streamFn,
L
Liu Jicong 已提交
140 141
                                   __optr_fn_t cleanup, __optr_close_fn_t closeFn, __optr_encode_fn_t encode,
                                   __optr_decode_fn_t decode, __optr_get_explain_fn_t explain) {
142 143 144 145 146 147 148 149 150 151 152 153 154 155
  SOperatorFpSet fpSet = {
      ._openFn = openFn,
      .getNextFn = nextFn,
      .getStreamResFn = streamFn,
      .cleanupFn = cleanup,
      .closeFn = closeFn,
      .encodeResultRow = encode,
      .decodeResultRow = decode,
      .getExplainFn = explain,
  };

  return fpSet;
}

H
Haojun Liao 已提交
156
void operatorDummyCloseFn(void* param, int32_t numOfCols) {}
H
Haojun Liao 已提交
157

dengyihao's avatar
dengyihao 已提交
158 159 160
static int32_t doCopyToSDataBlock(SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf,
                                  SGroupResInfo* pGroupResInfo, int32_t orderType, int32_t* rowCellOffset,
                                  SqlFunctionCtx* pCtx);
H
Haojun Liao 已提交
161

162
static void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size);
H
Haojun Liao 已提交
163
static void setResultBufSize(STaskAttr* pQueryAttr, SResultInfo* pResultInfo);
dengyihao's avatar
dengyihao 已提交
164 165
static void doSetTableGroupOutputBuf(SAggOperatorInfo* pAggInfo, int32_t numOfOutput, uint64_t groupId,
                                     SExecTaskInfo* pTaskInfo);
166

H
Haojun Liao 已提交
167
SArray* getOrderCheckColumns(STaskAttr* pQuery);
168 169

typedef struct SRowCompSupporter {
L
Liu Jicong 已提交
170 171 172
  STaskRuntimeEnv* pRuntimeEnv;
  int16_t          dataOffset;
  __compar_fn_t    comFunc;
173 174
} SRowCompSupporter;

L
Liu Jicong 已提交
175 176 177
static int compareRowData(const void* a, const void* b, const void* userData) {
  const SResultRow* pRow1 = (const SResultRow*)a;
  const SResultRow* pRow2 = (const SResultRow*)b;
178

L
Liu Jicong 已提交
179 180
  SRowCompSupporter* supporter = (SRowCompSupporter*)userData;
  STaskRuntimeEnv*   pRuntimeEnv = supporter->pRuntimeEnv;
181

L
Liu Jicong 已提交
182 183
  SFilePage* page1 = getBufPage(pRuntimeEnv->pResultBuf, pRow1->pageId);
  SFilePage* page2 = getBufPage(pRuntimeEnv->pResultBuf, pRow2->pageId);
184 185

  int16_t offset = supporter->dataOffset;
L
Liu Jicong 已提交
186 187
  char*   in1 = getPosInResultPage(pRuntimeEnv->pQueryAttr, page1, pRow1->offset, offset);
  char*   in2 = getPosInResultPage(pRuntimeEnv->pQueryAttr, page2, pRow2->offset, offset);
188 189 190 191

  return (in1 != NULL && in2 != NULL) ? supporter->comFunc(in1, in2) : 0;
}

L
Liu Jicong 已提交
192
// setup the output buffer for each operator
193
SSDataBlock* createResDataBlock(SDataBlockDescNode* pNode) {
H
Haojun Liao 已提交
194
  int32_t numOfCols = LIST_LENGTH(pNode->pSlots);
H
Haojun Liao 已提交
195

wafwerar's avatar
wafwerar 已提交
196
  SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
H
Haojun Liao 已提交
197 198
  pBlock->pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData));

H
Haojun Liao 已提交
199
  pBlock->info.blockId = pNode->dataBlockId;
dengyihao's avatar
dengyihao 已提交
200
  pBlock->info.rowSize = pNode->totalRowSize;  // todo ??
H
Haojun Liao 已提交
201

L
Liu Jicong 已提交
202
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
203
    SColumnInfoData idata = {{0}};
L
Liu Jicong 已提交
204
    SSlotDescNode*  pDescNode = nodesListGetNode(pNode->pSlots, i);
H
Haojun Liao 已提交
205 206
    if (!pDescNode->output) {
      continue;
H
Haojun Liao 已提交
207
    }
H
Haojun Liao 已提交
208

dengyihao's avatar
dengyihao 已提交
209
    idata.info.type = pDescNode->dataType.type;
L
Liu Jicong 已提交
210 211
    idata.info.bytes = pDescNode->dataType.bytes;
    idata.info.scale = pDescNode->dataType.scale;
H
Haojun Liao 已提交
212
    idata.info.slotId = pDescNode->slotId;
H
Haojun Liao 已提交
213 214
    idata.info.precision = pDescNode->dataType.precision;

H
Hongze Cheng 已提交
215 216 217 218
    if (IS_VAR_DATA_TYPE(idata.info.type)) {
      pBlock->info.hasVarCol = true;
    }

H
Haojun Liao 已提交
219
    taosArrayPush(pBlock->pDataBlock, &idata);
H
Haojun Liao 已提交
220
  }
H
Haojun Liao 已提交
221

222
  pBlock->info.numOfCols = taosArrayGetSize(pBlock->pDataBlock);
H
Haojun Liao 已提交
223
  return pBlock;
H
Haojun Liao 已提交
224 225
}

L
Liu Jicong 已提交
226
static bool hasNull(SColumn* pColumn, SColumnDataAgg* pStatis) {
dengyihao's avatar
dengyihao 已提交
227 228
  if (TSDB_COL_IS_TAG(pColumn->flag) || TSDB_COL_IS_UD_COL(pColumn->flag) ||
      pColumn->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
229 230 231 232 233 234 235 236 237 238
    return false;
  }

  if (pStatis != NULL && pStatis->numOfNull == 0) {
    return false;
  }

  return true;
}

H
Haojun Liao 已提交
239
static void prepareResultListBuffer(SResultRowInfo* pResultRowInfo, jmp_buf env) {
240 241
  int64_t newCapacity = 0;

242 243 244 245 246 247 248 249 250 251 252
  // more than the capacity, reallocate the resources
  if (pResultRowInfo->size < pResultRowInfo->capacity) {
    return;
  }

  if (pResultRowInfo->capacity > 10000) {
    newCapacity = (int64_t)(pResultRowInfo->capacity * 1.25);
  } else {
    newCapacity = (int64_t)(pResultRowInfo->capacity * 1.5);
  }

253
  if (newCapacity <= pResultRowInfo->capacity) {
H
Haojun Liao 已提交
254 255 256
    newCapacity += 4;
  }

257 258 259 260 261 262
  char* p = taosMemoryRealloc(pResultRowInfo->pPosition, newCapacity * sizeof(SResultRowPosition));
  if (p == NULL) {
    longjmp(env, TSDB_CODE_OUT_OF_MEMORY);
  }

  pResultRowInfo->pPosition = (SResultRowPosition*)p;
263 264

  int32_t inc = (int32_t)newCapacity - pResultRowInfo->capacity;
wmmhello's avatar
wmmhello 已提交
265
  memset(&pResultRowInfo->pPosition[pResultRowInfo->capacity], 0, sizeof(SResultRowPosition) * inc);
266 267 268
  pResultRowInfo->capacity = (int32_t)newCapacity;
}

L
Liu Jicong 已提交
269 270
static bool chkResultRowFromKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, char* pData,
                                int16_t bytes, bool masterscan, uint64_t uid) {
271 272 273
  bool existed = false;
  SET_RES_WINDOW_KEY(pRuntimeEnv->keyBuf, pData, bytes, uid);

L
Liu Jicong 已提交
274 275
  SResultRow** p1 =
      (SResultRow**)taosHashGet(pRuntimeEnv->pResultRowHashTable, pRuntimeEnv->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
276 277 278 279 280 281 282 283 284 285 286

  // in case of repeat scan/reverse scan, no new time window added.
  if (QUERY_IS_INTERVAL_QUERY(pRuntimeEnv->pQueryAttr)) {
    if (!masterscan) {  // the *p1 may be NULL in case of sliding+offset exists.
      return p1 != NULL;
    }

    if (p1 != NULL) {
      if (pResultRowInfo->size == 0) {
        existed = false;
      } else if (pResultRowInfo->size == 1) {
dengyihao's avatar
dengyihao 已提交
287
        //        existed = (pResultRowInfo->pResult[0] == (*p1));
288 289
      } else {  // check if current pResultRowInfo contains the existed pResultRow
        SET_RES_EXT_WINDOW_KEY(pRuntimeEnv->keyBuf, pData, bytes, uid, pResultRowInfo);
L
Liu Jicong 已提交
290 291
        int64_t* index =
            taosHashGet(pRuntimeEnv->pResultRowListSet, pRuntimeEnv->keyBuf, GET_RES_EXT_WINDOW_KEY_LEN(bytes));
292 293 294 295 296 297 298 299 300 301 302 303 304 305
        if (index != NULL) {
          existed = true;
        } else {
          existed = false;
        }
      }
    }

    return existed;
  }

  return p1 != NULL;
}

306
SResultRow* getNewResultRow_rv(SDiskbasedBuf* pResultBuf, int64_t tableGroupId, int32_t interBufSize) {
L
Liu Jicong 已提交
307
  SFilePage* pData = NULL;
308 309 310 311 312 313 314 315 316 317 318 319 320

  // in the first scan, new space needed for results
  int32_t pageId = -1;
  SIDList list = getDataBufPagesIdList(pResultBuf, tableGroupId);

  if (taosArrayGetSize(list) == 0) {
    pData = getNewBufPage(pResultBuf, tableGroupId, &pageId);
    pData->num = sizeof(SFilePage);
  } else {
    SPageInfo* pi = getLastPageInfo(list);
    pData = getBufPage(pResultBuf, getPageId(pi));
    pageId = getPageId(pi);

wmmhello's avatar
wmmhello 已提交
321
    if (pData->num + interBufSize > getBufPageSize(pResultBuf)) {
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
      // release current page first, and prepare the next one
      releaseBufPageInfo(pResultBuf, pi);

      pData = getNewBufPage(pResultBuf, tableGroupId, &pageId);
      if (pData != NULL) {
        pData->num = sizeof(SFilePage);
      }
    }
  }

  if (pData == NULL) {
    return NULL;
  }

  // set the number of rows in current disk page
  SResultRow* pResultRow = (SResultRow*)((char*)pData + pData->num);
  pResultRow->pageId = pageId;
  pResultRow->offset = (int32_t)pData->num;

wmmhello's avatar
wmmhello 已提交
341
  pData->num += interBufSize;
342 343 344 345

  return pResultRow;
}

346 347 348 349 350 351 352
/**
 * the struct of key in hash table
 * +----------+---------------+
 * | group id |   key data    |
 * | 8 bytes  | actual length |
 * +----------+---------------+
 */
353 354 355
SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pResultRowInfo, char* pData,
                                   int16_t bytes, bool masterscan, uint64_t groupId, SExecTaskInfo* pTaskInfo,
                                   bool isIntervalQuery, SAggSupporter* pSup) {
356
  SET_RES_WINDOW_KEY(pSup->keyBuf, pData, bytes, groupId);
H
Haojun Liao 已提交
357

dengyihao's avatar
dengyihao 已提交
358 359
  SResultRowPosition* p1 =
      (SResultRowPosition*)taosHashGet(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
H
Haojun Liao 已提交
360

361 362
  SResultRow* pResult = NULL;

H
Haojun Liao 已提交
363 364
  // in case of repeat scan/reverse scan, no new time window added.
  if (isIntervalQuery) {
365 366
    if (masterscan && p1 != NULL) {  // the *p1 may be NULL in case of sliding+offset exists.
      pResult = getResultRowByPos(pResultBuf, p1);
H
Haojun Liao 已提交
367 368
    }
  } else {
dengyihao's avatar
dengyihao 已提交
369 370
    // In case of group by column query, the required SResultRow object must be existInCurrentResusltRowInfo in the
    // pResultRowInfo object.
H
Haojun Liao 已提交
371
    if (p1 != NULL) {
372
      pResult = getResultRowByPos(pResultBuf, p1);
H
Haojun Liao 已提交
373 374 375
    }
  }

L
Liu Jicong 已提交
376
  // 1. close current opened time window
377
  if (pResultRowInfo->cur.pageId != -1 && ((pResult == NULL) || (pResult->pageId != pResultRowInfo->cur.pageId &&
L
Liu Jicong 已提交
378
                                                                 pResult->offset != pResultRowInfo->cur.offset))) {
379 380 381 382 383 384 385 386 387 388 389 390 391
    // todo extract function
    SResultRowPosition pos = pResultRowInfo->cur;
    SFilePage*         pPage = getBufPage(pResultBuf, pos.pageId);
    SResultRow*        pRow = (SResultRow*)((char*)pPage + pos.offset);
    closeResultRow(pRow);
    releaseBufPage(pResultBuf, pPage);
  }

  // allocate a new buffer page
  prepareResultListBuffer(pResultRowInfo, pTaskInfo->env);
  if (pResult == NULL) {
    pResult = getNewResultRow_rv(pResultBuf, groupId, pSup->resultRowSize);
    initResultRow(pResult);
H
Haojun Liao 已提交
392

393 394
    // add a new result set for a new group
    SResultRowPosition pos = {.pageId = pResult->pageId, .offset = pResult->offset};
L
Liu Jicong 已提交
395 396
    taosHashPut(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes), &pos,
                sizeof(SResultRowPosition));
H
Haojun Liao 已提交
397 398
  }

399 400 401 402 403
  // 2. set the new time window to be the new active time window
  pResultRowInfo->pPosition[pResultRowInfo->size++] =
      (SResultRowPosition){.pageId = pResult->pageId, .offset = pResult->offset};
  pResultRowInfo->cur = (SResultRowPosition){.pageId = pResult->pageId, .offset = pResult->offset};

H
Haojun Liao 已提交
404 405 406 407 408
  // too many time window in query
  if (pResultRowInfo->size > MAX_INTERVAL_TIME_WINDOW) {
    longjmp(pTaskInfo->env, TSDB_CODE_QRY_TOO_MANY_TIMEWINDOW);
  }

H
Haojun Liao 已提交
409
  return pResult;
H
Haojun Liao 已提交
410 411
}

412
// get the correct time window according to the handled timestamp
L
Liu Jicong 已提交
413
static STimeWindow getCurrentActiveTimeWindow(SResultRowInfo* pResultRowInfo, int64_t ts, STaskAttr* pQueryAttr) {
414
  STimeWindow w = {0};
H
Haojun Liao 已提交
415
#if 0
L
Liu Jicong 已提交
416 417
  if (pResultRowInfo->curPos == -1) {  // the first window, from the previous stored value
                                       //    getInitialStartTimeWindow(pQueryAttr, ts, &w);
418 419

    if (pQueryAttr->interval.intervalUnit == 'n' || pQueryAttr->interval.intervalUnit == 'y') {
L
Liu Jicong 已提交
420 421 422
      w.ekey =
          taosTimeAdd(w.skey, pQueryAttr->interval.interval, pQueryAttr->interval.intervalUnit, pQueryAttr->precision) -
          1;
423 424 425 426
    } else {
      w.ekey = w.skey + pQueryAttr->interval.interval - 1;
    }
  } else {
H
Haojun Liao 已提交
427
    w = pRow->win;
428 429 430 431 432 433 434 435 436
  }

  /*
   * query border check, skey should not be bounded by the query time range, since the value skey will
   * be used as the time window index value. So we only change ekey of time window accordingly.
   */
  if (w.ekey > pQueryAttr->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) {
    w.ekey = pQueryAttr->window.ekey;
  }
H
Haojun Liao 已提交
437
#endif
438 439 440 441 442

  return w;
}

// a new buffer page for each table. Needs to opt this design
L
Liu Jicong 已提交
443
static int32_t addNewWindowResultBuf(SResultRow* pWindowRes, SDiskbasedBuf* pResultBuf, int32_t tid, uint32_t size) {
444 445 446 447
  if (pWindowRes->pageId != -1) {
    return 0;
  }

L
Liu Jicong 已提交
448
  SFilePage* pData = NULL;
449 450 451 452 453 454

  // in the first scan, new space needed for results
  int32_t pageId = -1;
  SIDList list = getDataBufPagesIdList(pResultBuf, tid);

  if (taosArrayGetSize(list) == 0) {
H
Haojun Liao 已提交
455
    pData = getNewBufPage(pResultBuf, tid, &pageId);
456
    pData->num = sizeof(SFilePage);
457 458
  } else {
    SPageInfo* pi = getLastPageInfo(list);
459
    pData = getBufPage(pResultBuf, getPageId(pi));
460
    pageId = getPageId(pi);
461

462
    if (pData->num + size > getBufPageSize(pResultBuf)) {
463
      // release current page first, and prepare the next one
464
      releaseBufPageInfo(pResultBuf, pi);
465

H
Haojun Liao 已提交
466
      pData = getNewBufPage(pResultBuf, tid, &pageId);
467
      if (pData != NULL) {
468
        pData->num = sizeof(SFilePage);
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
      }
    }
  }

  if (pData == NULL) {
    return -1;
  }

  // set the number of rows in current disk page
  if (pWindowRes->pageId == -1) {  // not allocated yet, allocate new buffer
    pWindowRes->pageId = pageId;
    pWindowRes->offset = (int32_t)pData->num;

    pData->num += size;
    assert(pWindowRes->pageId >= 0);
  }

  return 0;
}

L
Liu Jicong 已提交
489 490 491
static bool chkWindowOutputBufByKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, STimeWindow* win,
                                    bool masterscan, SResultRow** pResult, int64_t groupId, SqlFunctionCtx* pCtx,
                                    int32_t numOfOutput, int32_t* rowCellInfoOffset) {
492
  assert(win->skey <= win->ekey);
L
Liu Jicong 已提交
493
  return chkResultRowFromKey(pRuntimeEnv, pResultRowInfo, (char*)&win->skey, TSDB_KEYSIZE, masterscan, groupId);
494 495
}

dengyihao's avatar
dengyihao 已提交
496 497
static void doUpdateResultRowIndex(SResultRowInfo* pResultRowInfo, TSKEY lastKey, bool ascQuery,
                                   bool timeWindowInterpo) {
498
  int64_t skey = TSKEY_INITIAL_VAL;
H
Haojun Liao 已提交
499
#if 0
500 501
  int32_t i = 0;
  for (i = pResultRowInfo->size - 1; i >= 0; --i) {
L
Liu Jicong 已提交
502
    SResultRow* pResult = pResultRowInfo->pResult[i];
503 504 505 506 507 508
    if (pResult->closed) {
      break;
    }

    // new closed result rows
    if (timeWindowInterpo) {
L
Liu Jicong 已提交
509 510 511
      if (pResult->endInterp &&
          ((pResult->win.skey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery))) {
        if (i > 0) {  // the first time window, the startInterp is false.
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
          assert(pResult->startInterp);
        }

        closeResultRow(pResultRowInfo, i);
      } else {
        skey = pResult->win.skey;
      }
    } else {
      if ((pResult->win.ekey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery)) {
        closeResultRow(pResultRowInfo, i);
      } else {
        skey = pResult->win.skey;
      }
    }
  }

  // all result rows are closed, set the last one to be the skey
  if (skey == TSKEY_INITIAL_VAL) {
    if (pResultRowInfo->size == 0) {
L
Liu Jicong 已提交
531
      //      assert(pResultRowInfo->current == NULL);
532 533 534 535 536 537 538
      assert(pResultRowInfo->curPos == -1);
      pResultRowInfo->curPos = -1;
    } else {
      pResultRowInfo->curPos = pResultRowInfo->size - 1;
    }
  } else {
    for (i = pResultRowInfo->size - 1; i >= 0; --i) {
L
Liu Jicong 已提交
539
      SResultRow* pResult = pResultRowInfo->pResult[i];
540 541 542 543 544 545 546 547 548 549 550
      if (pResult->closed) {
        break;
      }
    }

    if (i == pResultRowInfo->size - 1) {
      pResultRowInfo->curPos = i;
    } else {
      pResultRowInfo->curPos = i + 1;  // current not closed result object
    }
  }
H
Haojun Liao 已提交
551
#endif
552
}
553
//
dengyihao's avatar
dengyihao 已提交
554
// static void updateResultRowInfoActiveIndex(SResultRowInfo* pResultRowInfo, const STimeWindow* pWin, TSKEY lastKey,
555 556 557 558 559 560 561 562 563
//                                           bool ascQuery, bool interp) {
//  if ((lastKey > pWin->ekey && ascQuery) || (lastKey < pWin->ekey && (!ascQuery))) {
//    closeAllResultRows(pResultRowInfo);
//    pResultRowInfo->curPos = pResultRowInfo->size - 1;
//  } else {
//    int32_t step = ascQuery ? 1 : -1;
//    doUpdateResultRowIndex(pResultRowInfo, lastKey - step, ascQuery, interp);
//  }
//}
564

565
//  query_range_start, query_range_end, window_duration, window_start, window_end
566
void initExecTimeWindowInfo(SColumnInfoData* pColData, STimeWindow* pQueryWindow) {
567 568 569
  pColData->info.type = TSDB_DATA_TYPE_TIMESTAMP;
  pColData->info.bytes = sizeof(int64_t);

570
  colInfoDataEnsureCapacity(pColData, 0, 5);
571 572 573 574 575 576 577 578 579
  colDataAppendInt64(pColData, 0, &pQueryWindow->skey);
  colDataAppendInt64(pColData, 1, &pQueryWindow->ekey);

  int64_t interval = 0;
  colDataAppendInt64(pColData, 2, &interval);  // this value may be variable in case of 'n' and 'y'.
  colDataAppendInt64(pColData, 3, &pQueryWindow->skey);
  colDataAppendInt64(pColData, 4, &pQueryWindow->ekey);
}

dengyihao's avatar
dengyihao 已提交
580 581
void doApplyFunctions(SqlFunctionCtx* pCtx, STimeWindow* pWin, SColumnInfoData* pTimeWindowData, int32_t offset,
                      int32_t forwardStep, TSKEY* tsCol, int32_t numOfTotal, int32_t numOfOutput, int32_t order) {
582 583 584
  for (int32_t k = 0; k < numOfOutput; ++k) {
    pCtx[k].startTs = pWin->skey;

H
Haojun Liao 已提交
585
    // keep it temporarily
dengyihao's avatar
dengyihao 已提交
586 587
    bool    hasAgg = pCtx[k].input.colDataAggIsSet;
    int32_t numOfRows = pCtx[k].input.numOfRows;
H
Haojun Liao 已提交
588
    int32_t startOffset = pCtx[k].input.startRowIndex;
589

H
Haojun Liao 已提交
590
    int32_t pos = (order == TSDB_ORDER_ASC) ? offset : offset - (forwardStep - 1);
591 592
    pCtx[k].input.startRowIndex = pos;
    pCtx[k].input.numOfRows = forwardStep;
593 594

    if (tsCol != NULL) {
H
Haojun Liao 已提交
595
      pCtx[k].ptsList = tsCol;
596 597 598 599 600 601 602 603
    }

    // not a whole block involved in query processing, statistics data can not be used
    // NOTE: the original value of isSet have been changed here
    if (pCtx[k].isAggSet && forwardStep < numOfTotal) {
      pCtx[k].isAggSet = false;
    }

604 605
    if (fmIsWindowPseudoColumnFunc(pCtx[k].functionId)) {
      SResultRowEntryInfo* pEntryInfo = GET_RES_INFO(&pCtx[k]);
dengyihao's avatar
dengyihao 已提交
606
      char*                p = GET_ROWCELL_INTERBUF(pEntryInfo);
607

608
      SColumnInfoData idata = {0};
dengyihao's avatar
dengyihao 已提交
609
      idata.info.type = TSDB_DATA_TYPE_BIGINT;
610
      idata.info.bytes = tDataTypes[TSDB_DATA_TYPE_BIGINT].bytes;
dengyihao's avatar
dengyihao 已提交
611
      idata.pData = p;
612 613 614 615

      SScalarParam out = {.columnData = &idata};
      SScalarParam tw = {.numOfRows = 5, .columnData = pTimeWindowData};
      pCtx[k].sfp.process(&tw, 1, &out);
616 617 618 619
      pEntryInfo->numOfRes = 1;
      continue;
    }

H
Haojun Liao 已提交
620
    if (functionNeedToExecute(&pCtx[k])) {
621
      pCtx[k].fpSet.process(&pCtx[k]);
622 623 624
    }

    // restore it
625 626 627
    pCtx[k].input.colDataAggIsSet = hasAgg;
    pCtx[k].input.startRowIndex = startOffset;
    pCtx[k].input.numOfRows = numOfRows;
628 629 630
  }
}

L
Liu Jicong 已提交
631
static FORCE_INLINE TSKEY reviseWindowEkey(STaskAttr* pQueryAttr, STimeWindow* pWindow) {
dengyihao's avatar
dengyihao 已提交
632
  TSKEY   ekey = -1;
633 634
  int32_t order = TSDB_ORDER_ASC;
  if (order == TSDB_ORDER_ASC) {
635 636 637 638 639 640 641 642 643 644 645 646 647 648
    ekey = pWindow->ekey;
    if (ekey > pQueryAttr->window.ekey) {
      ekey = pQueryAttr->window.ekey;
    }
  } else {
    ekey = pWindow->skey;
    if (ekey < pQueryAttr->window.ekey) {
      ekey = pQueryAttr->window.ekey;
    }
  }

  return ekey;
}

dengyihao's avatar
dengyihao 已提交
649 650
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                                   bool createDummyCol);
651

dengyihao's avatar
dengyihao 已提交
652 653
static void doSetInputDataBlockInfo(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock,
                                    int32_t order) {
654 655
  for (int32_t i = 0; i < pOperator->numOfOutput; ++i) {
    pCtx[i].order = order;
L
Liu Jicong 已提交
656
    pCtx[i].size = pBlock->info.rows;
657
    setBlockStatisInfo(&pCtx[i], &pOperator->pExpr[i], pBlock);
658 659 660
  }
}

dengyihao's avatar
dengyihao 已提交
661 662
void setInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                       bool createDummyCol) {
663
  if (pBlock->pBlockAgg != NULL) {
H
Haojun Liao 已提交
664
    doSetInputDataBlockInfo(pOperator, pCtx, pBlock, order);
665
  } else {
666
    doSetInputDataBlock(pOperator, pCtx, pBlock, order, createDummyCol);
H
Haojun Liao 已提交
667
  }
668 669
}

L
Liu Jicong 已提交
670 671
static int32_t doCreateConstantValColumnInfo(SInputColumnInfoData* pInput, SFunctParam* pFuncParam, int32_t paramIndex,
                                             int32_t numOfRows) {
672 673 674 675 676 677 678 679
  SColumnInfoData* pColInfo = NULL;
  if (pInput->pData[paramIndex] == NULL) {
    pColInfo = taosMemoryCalloc(1, sizeof(SColumnInfoData));
    if (pColInfo == NULL) {
      return TSDB_CODE_OUT_OF_MEMORY;
    }

    // Set the correct column info (data type and bytes)
680 681
    pColInfo->info.type = pFuncParam->param.nType;
    pColInfo->info.bytes = pFuncParam->param.nLen;
682 683

    pInput->pData[paramIndex] = pColInfo;
684 685
  } else {
    pColInfo = pInput->pData[paramIndex];
686 687
  }

688
  colInfoDataEnsureCapacity(pColInfo, 0, numOfRows);
689

690
  int8_t type = pFuncParam->param.nType;
691 692
  if (type == TSDB_DATA_TYPE_BIGINT || type == TSDB_DATA_TYPE_UBIGINT) {
    int64_t v = pFuncParam->param.i;
dengyihao's avatar
dengyihao 已提交
693
    for (int32_t i = 0; i < numOfRows; ++i) {
694 695 696 697
      colDataAppendInt64(pColInfo, i, &v);
    }
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
dengyihao's avatar
dengyihao 已提交
698
    for (int32_t i = 0; i < numOfRows; ++i) {
699 700
      colDataAppendDouble(pColInfo, i, &v);
    }
701
  } else if (type == TSDB_DATA_TYPE_VARCHAR) {
L
Liu Jicong 已提交
702
    char* tmp = taosMemoryMalloc(pFuncParam->param.nLen + VARSTR_HEADER_SIZE);
703
    STR_WITH_SIZE_TO_VARSTR(tmp, pFuncParam->param.pz, pFuncParam->param.nLen);
L
Liu Jicong 已提交
704
    for (int32_t i = 0; i < numOfRows; ++i) {
705 706
      colDataAppend(pColInfo, i, tmp, false);
    }
707 708 709 710 711
  }

  return TSDB_CODE_SUCCESS;
}

dengyihao's avatar
dengyihao 已提交
712 713
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                                   bool createDummyCol) {
714 715
  int32_t code = TSDB_CODE_SUCCESS;

716
  for (int32_t i = 0; i < pOperator->numOfOutput; ++i) {
L
Liu Jicong 已提交
717 718 719
    pCtx[i].order = order;
    pCtx[i].size = pBlock->info.rows;
    pCtx[i].pSrcBlock = pBlock;
H
Haojun Liao 已提交
720 721
    pCtx[i].currentStage = MAIN_SCAN;

722
    SInputColumnInfoData* pInput = &pCtx[i].input;
723
    pInput->uid = pBlock->info.uid;
C
Cary Xu 已提交
724
    pInput->colDataAggIsSet = false;
725

726 727
    SExprInfo* pOneExpr = &pOperator->pExpr[i];
    for (int32_t j = 0; j < pOneExpr->base.numOfParams; ++j) {
dengyihao's avatar
dengyihao 已提交
728
      SFunctParam* pFuncParam = &pOneExpr->base.pParam[j];
G
Ganlin Zhao 已提交
729 730
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
dengyihao's avatar
dengyihao 已提交
731
        pInput->pData[j] = taosArrayGet(pBlock->pDataBlock, slotId);
732 733 734
        pInput->totalRows = pBlock->info.rows;
        pInput->numOfRows = pBlock->info.rows;
        pInput->startRowIndex = 0;
735

736 737 738 739
        // the last parameter is the timestamp column
        if (fmIsTimelineFunc(pCtx[i].functionId) && (j == pOneExpr->base.numOfParams - 1)) {
          pInput->pPTS = pInput->pData[j];
        }
740 741
        ASSERT(pInput->pData[j] != NULL);
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
742 743 744
        // todo avoid case: top(k, 12), 12 is the value parameter.
        // sum(11), 11 is also the value parameter.
        if (createDummyCol && pOneExpr->base.numOfParams == 1) {
745 746 747 748
          pInput->totalRows = pBlock->info.rows;
          pInput->numOfRows = pBlock->info.rows;
          pInput->startRowIndex = 0;

749
          code = doCreateConstantValColumnInfo(pInput, pFuncParam, j, pBlock->info.rows);
750 751 752
          if (code != TSDB_CODE_SUCCESS) {
            return code;
          }
753
        }
G
Ganlin Zhao 已提交
754 755
      }
    }
H
Haojun Liao 已提交
756

H
Haojun Liao 已提交
757
    //    setBlockStatisInfo(&pCtx[i], pBlock, pOperator->pExpr[i].base.pColumns);
H
Haojun Liao 已提交
758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795
    //      uint32_t flag = pOperator->pExpr[i].base.pParam[0].pCol->flag;
    //      if (TSDB_COL_IS_NORMAL_COL(flag) /*|| (pCtx[i].functionId == FUNCTION_BLKINFO) ||
    //          (TSDB_COL_IS_TAG(flag) && pOperator->pRuntimeEnv->scanFlag == MERGE_STAGE)*/) {

    //        SColumn* pCol = pOperator->pExpr[i].base.pParam[0].pCol;
    //        if (pCtx[i].columnIndex == -1) {
    //          for(int32_t j = 0; j < pBlock->info.numOfCols; ++j) {
    //            SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, j);
    //            if (pColData->info.colId == pCol->colId) {
    //              pCtx[i].columnIndex = j;
    //              break;
    //            }
    //          }
    //        }

    //        uint32_t status = aAggs[pCtx[i].functionId].status;
    //        if ((status & (FUNCSTATE_SELECTIVITY | FUNCSTATE_NEED_TS)) != 0) {
    //          SColumnInfoData* tsInfo = taosArrayGet(pBlock->pDataBlock, 0);
    // In case of the top/bottom query again the nest query result, which has no timestamp column
    // don't set the ptsList attribute.
    //          if (tsInfo->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
    //            pCtx[i].ptsList = (int64_t*) tsInfo->pData;
    //          } else {
    //            pCtx[i].ptsList = NULL;
    //          }
    //        }
    //      } else if (TSDB_COL_IS_UD_COL(pCol->flag) && (pOperator->pRuntimeEnv->scanFlag == MERGE_STAGE)) {
    //        SColIndex*       pColIndex = &pOperator->pExpr[i].base.colInfo;
    //        SColumnInfoData* p = taosArrayGet(pBlock->pDataBlock, pColIndex->colIndex);
    //
    //        pCtx[i].pInput = p->pData;
    //        assert(p->info.colId == pColIndex->info.colId && pCtx[i].inputType == p->info.type);
    //        for(int32_t j = 0; j < pBlock->info.rows; ++j) {
    //          char* dst = p->pData + j * p->info.bytes;
    //          taosVariantDump(&pOperator->pExpr[i].base.param[1], dst, p->info.type, true);
    //        }
    //      }
  }
796 797

  return code;
H
Haojun Liao 已提交
798 799 800
}

static void doAggregateImpl(SOperatorInfo* pOperator, TSKEY startTs, SqlFunctionCtx* pCtx) {
801
  for (int32_t k = 0; k < pOperator->numOfOutput; ++k) {
H
Haojun Liao 已提交
802
    if (functionNeedToExecute(&pCtx[k])) {
L
Liu Jicong 已提交
803
      pCtx[k].startTs = startTs;  // this can be set during create the struct
H
Haojun Liao 已提交
804
      pCtx[k].fpSet.process(&pCtx[k]);
805 806 807 808
    }
  }
}

H
Haojun Liao 已提交
809
static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, SArray* pPseudoList) {
dengyihao's avatar
dengyihao 已提交
810
  size_t num = (pPseudoList != NULL) ? taosArrayGetSize(pPseudoList) : 0;
H
Haojun Liao 已提交
811 812 813 814 815
  for (int32_t i = 0; i < num; ++i) {
    pCtx[i].pOutput = taosArrayGet(pResult->pDataBlock, i);
  }
}

816
int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
X
Xiaoyu Wang 已提交
817
                              int32_t numOfOutput, SArray* pPseudoList) {
H
Haojun Liao 已提交
818
  setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
H
Haojun Liao 已提交
819
  pResult->info.groupId = pSrcBlock->info.groupId;
H
Haojun Liao 已提交
820

dengyihao's avatar
dengyihao 已提交
821 822
  // if the source equals to the destination, it is to create a new column as the result of scalar function or some
  // operators.
823 824
  bool createNewColModel = (pResult == pSrcBlock);

825 826
  int32_t numOfRows = 0;

827
  for (int32_t k = 0; k < numOfOutput; ++k) {
dengyihao's avatar
dengyihao 已提交
828
    int32_t         outputSlotId = pExpr[k].base.resSchema.slotId;
829 830
    SqlFunctionCtx* pfCtx = &pCtx[k];

L
Liu Jicong 已提交
831
    if (pExpr[k].pExpr->nodeType == QUERY_NODE_COLUMN) {  // it is a project query
832
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
833
      if (pResult->info.rows > 0 && !createNewColModel) {
834
        colDataMergeCol(pColInfoData, pResult->info.rows, &pResult->info.capacity, pfCtx->input.pData[0], pfCtx->input.numOfRows);
835 836 837
      } else {
        colDataAssign(pColInfoData, pfCtx->input.pData[0], pfCtx->input.numOfRows);
      }
838

839
      numOfRows = pfCtx->input.numOfRows;
840
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE) {
841
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
842

dengyihao's avatar
dengyihao 已提交
843
      int32_t offset = createNewColModel ? 0 : pResult->info.rows;
844
      for (int32_t i = 0; i < pSrcBlock->info.rows; ++i) {
dengyihao's avatar
dengyihao 已提交
845 846 847
        colDataAppend(pColInfoData, i + offset,
                      taosVariantGet(&pExpr[k].base.pParam[0].param, pExpr[k].base.pParam[0].param.nType),
                      TSDB_DATA_TYPE_NULL == pExpr[k].base.pParam[0].param.nType);
848
      }
849 850

      numOfRows = pSrcBlock->info.rows;
H
Haojun Liao 已提交
851
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_OPERATOR) {
852 853 854
      SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
      taosArrayPush(pBlockList, &pSrcBlock);

855
      SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
856
      SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
857

858
      SScalarParam dest = {.columnData = &idata};
859 860
      scalarCalculate(pExpr[k].pExpr->_optrRoot.pRootNode, pBlockList, &dest);

dengyihao's avatar
dengyihao 已提交
861
      int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
862
      colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
863 864

      numOfRows = dest.numOfRows;
865 866
      taosArrayDestroy(pBlockList);
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_FUNCTION) {
867
      ASSERT(!fmIsAggFunc(pfCtx->functionId));
868

869
      if (fmIsPseudoColumnFunc(pfCtx->functionId)) {
H
Haojun Liao 已提交
870
        // do nothing
871
      } else if (fmIsNonstandardSQLFunc(pfCtx->functionId)) {
dengyihao's avatar
dengyihao 已提交
872
        SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[k]);
873
        pfCtx->fpSet.init(&pCtx[k], pResInfo);
874

875
        pfCtx->pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
dengyihao's avatar
dengyihao 已提交
876
        pfCtx->offset = createNewColModel ? 0 : pResult->info.rows;  // set the start offset
H
Haojun Liao 已提交
877

878
        // set the timestamp(_rowts) output buffer
879 880
        if (taosArrayGetSize(pPseudoList) > 0) {
          int32_t* outputColIndex = taosArrayGet(pPseudoList, 0);
881
          pfCtx->pTsOutput = (SColumnInfoData*)pCtx[*outputColIndex].pOutput;
882
        }
H
Haojun Liao 已提交
883

884
        numOfRows = pfCtx->fpSet.process(pfCtx);
H
Haojun Liao 已提交
885 886 887
      } else {
        SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
        taosArrayPush(pBlockList, &pSrcBlock);
G
Ganlin Zhao 已提交
888

889
        SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
890
        SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
H
Haojun Liao 已提交
891

892
        SScalarParam dest = {.columnData = &idata};
X
Xiaoyu Wang 已提交
893
        int32_t      code = scalarCalculate((SNode*)pExpr[k].pExpr->_function.pFunctNode, pBlockList, &dest);
894 895 896 897
        if (code != TSDB_CODE_SUCCESS) {
          taosArrayDestroy(pBlockList);
          return code;
        }
898

dengyihao's avatar
dengyihao 已提交
899
        int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
900
        colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
901 902

        numOfRows = dest.numOfRows;
H
Haojun Liao 已提交
903 904
        taosArrayDestroy(pBlockList);
      }
905
    } else {
906
      ASSERT(0);
907 908
    }
  }
909

910 911 912
  if (!createNewColModel) {
    pResult->info.rows += numOfRows;
  }
913 914

  return TSDB_CODE_SUCCESS;
915 916
}

917 918 919 920 921 922 923 924 925 926 927 928
static void setResultRowKey(SResultRow* pResultRow, char* pData, int16_t type) {
  if (IS_VAR_DATA_TYPE(type)) {
    // todo disable this
    //    if (pResultRow->key == NULL) {
    //      pResultRow->key = taosMemoryMalloc(varDataTLen(pData));
    //      varDataCopy(pResultRow->key, pData);
    //    } else {
    //      ASSERT(memcmp(pResultRow->key, pData, varDataTLen(pData)) == 0);
    //    }
  } else {
    int64_t v = -1;
    GET_TYPED_DATA(v, int64_t, type, pData);
929

930 931
    pResultRow->win.skey = v;
    pResultRow->win.ekey = v;
932 933 934
  }
}

935 936 937 938 939
int32_t setGroupResultOutputBuf(SOptrBasicInfo* binfo, int32_t numOfCols, char* pData, int16_t type, int16_t bytes,
                                int32_t groupId, SDiskbasedBuf* pBuf, SExecTaskInfo* pTaskInfo,
                                SAggSupporter* pAggSup) {
  SResultRowInfo* pResultRowInfo = &binfo->resultRowInfo;
  SqlFunctionCtx* pCtx = binfo->pCtx;
940

941 942 943
  SResultRow* pResultRow =
      doSetResultOutBufByKey(pBuf, pResultRowInfo, (char*)pData, bytes, true, groupId, pTaskInfo, false, pAggSup);
  assert(pResultRow != NULL);
944

945
  setResultRowKey(pResultRow, pData, type);
946
  setResultRowInitCtx(pResultRow, pCtx, numOfCols, binfo->rowCellInfoOffset);
947
  return TSDB_CODE_SUCCESS;
948 949
}

950 951
static bool functionNeedToExecute(SqlFunctionCtx* pCtx) {
  struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
952

953 954 955 956 957
  // in case of timestamp column, always generated results.
  int32_t functionId = pCtx->functionId;
  if (functionId == -1) {
    return false;
  }
958

959
  if (isRowEntryCompleted(pResInfo)) {
960 961 962
    return false;
  }

963 964 965 966 967 968 969
  if (functionId == FUNCTION_FIRST_DST || functionId == FUNCTION_FIRST) {
    //    return QUERY_IS_ASC_QUERY(pQueryAttr);
  }

  // denote the order type
  if ((functionId == FUNCTION_LAST_DST || functionId == FUNCTION_LAST)) {
    //    return pCtx->param[0].i == pQueryAttr->order.order;
970 971
  }

972 973 974 975 976
  // in the reverse table scan, only the following functions need to be executed
  //  if (IS_REVERSE_SCAN(pRuntimeEnv) ||
  //      (pRuntimeEnv->scanFlag == REPEAT_SCAN && functionId != FUNCTION_STDDEV && functionId != FUNCTION_PERCT)) {
  //    return false;
  //  }
977 978 979 980

  return true;
}

981 982 983 984 985 986 987
static int32_t doCreateConstantValColumnAggInfo(SInputColumnInfoData* pInput, SFunctParam* pFuncParam, int32_t type,
                                                int32_t paramIndex, int32_t numOfRows) {
  if (pInput->pData[paramIndex] == NULL) {
    pInput->pData[paramIndex] = taosMemoryCalloc(1, sizeof(SColumnInfoData));
    if (pInput->pData[paramIndex] == NULL) {
      return TSDB_CODE_OUT_OF_MEMORY;
    }
988

989 990 991
    // Set the correct column info (data type and bytes)
    pInput->pData[paramIndex]->info.type = type;
    pInput->pData[paramIndex]->info.bytes = tDataTypes[type].bytes;
992
  }
H
Haojun Liao 已提交
993

994 995 996 997 998 999
  SColumnDataAgg* da = NULL;
  if (pInput->pColumnDataAgg[paramIndex] == NULL) {
    da = taosMemoryCalloc(1, sizeof(SColumnDataAgg));
    pInput->pColumnDataAgg[paramIndex] = da;
    if (da == NULL) {
      return TSDB_CODE_OUT_OF_MEMORY;
1000 1001
    }
  } else {
1002
    da = pInput->pColumnDataAgg[paramIndex];
1003 1004
  }

1005
  ASSERT(!IS_VAR_DATA_TYPE(type));
1006

1007 1008 1009 1010 1011 1012
  if (type == TSDB_DATA_TYPE_BIGINT) {
    int64_t v = pFuncParam->param.i;
    *da = (SColumnDataAgg){.numOfNull = 0, .min = v, .max = v, .maxIndex = 0, .minIndex = 0, .sum = v * numOfRows};
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
    *da = (SColumnDataAgg){.numOfNull = 0, .maxIndex = 0, .minIndex = 0};
1013

1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
    *(double*)&da->min = v;
    *(double*)&da->max = v;
    *(double*)&da->sum = v * numOfRows;
  } else if (type == TSDB_DATA_TYPE_BOOL) {  // todo validate this data type
    bool v = pFuncParam->param.i;

    *da = (SColumnDataAgg){.numOfNull = 0, .maxIndex = 0, .minIndex = 0};
    *(bool*)&da->min = 0;
    *(bool*)&da->max = v;
    *(bool*)&da->sum = v * numOfRows;
  } else if (type == TSDB_DATA_TYPE_TIMESTAMP) {
    // do nothing
1026
  } else {
1027
    ASSERT(0);
1028 1029
  }

1030 1031
  return TSDB_CODE_SUCCESS;
}
1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042

void setBlockStatisInfo(SqlFunctionCtx* pCtx, SExprInfo* pExprInfo, SSDataBlock* pBlock) {
  int32_t numOfRows = pBlock->info.rows;

  SInputColumnInfoData* pInput = &pCtx->input;
  pInput->numOfRows = numOfRows;
  pInput->totalRows = numOfRows;

  if (pBlock->pBlockAgg != NULL) {
    pInput->colDataAggIsSet = true;

1043 1044
    for (int32_t j = 0; j < pExprInfo->base.numOfParams; ++j) {
      SFunctParam* pFuncParam = &pExprInfo->base.pParam[j];
1045

1046 1047
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
1048 1049 1050 1051
        pInput->pColumnDataAgg[j] = pBlock->pBlockAgg[slotId];
        if (pInput->pColumnDataAgg[j] == NULL) {
          pInput->colDataAggIsSet = false;
        }
1052 1053 1054 1055

        // Here we set the column info data since the data type for each column data is required, but
        // the data in the corresponding SColumnInfoData will not be used.
        pInput->pData[j] = taosArrayGet(pBlock->pDataBlock, slotId);
1056 1057
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
        doCreateConstantValColumnAggInfo(pInput, pFuncParam, pFuncParam->param.nType, j, pBlock->info.rows);
1058 1059
      }
    }
1060
  } else {
1061
    pInput->colDataAggIsSet = false;
1062 1063 1064
  }

  // set the statistics data for primary time stamp column
1065 1066 1067 1068 1069
  //  if (pCtx->functionId == FUNCTION_SPREAD && pColumn->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
  //    pCtx->isAggSet = true;
  //    pCtx->agg.min = pBlock->info.window.skey;
  //    pCtx->agg.max = pBlock->info.window.ekey;
  //  }
1070 1071 1072
}

// set the output buffer for the selectivity + tag query
L
Liu Jicong 已提交
1073
static int32_t setCtxTagColumnInfo(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1074 1075 1076
  int32_t num = 0;
  int16_t tagLen = 0;

H
Haojun Liao 已提交
1077
  SqlFunctionCtx*  p = NULL;
wafwerar's avatar
wafwerar 已提交
1078
  SqlFunctionCtx** pTagCtx = taosMemoryCalloc(numOfOutput, POINTER_BYTES);
1079 1080 1081 1082 1083 1084 1085 1086
  if (pTagCtx == NULL) {
    return TSDB_CODE_QRY_OUT_OF_MEMORY;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
    int32_t functionId = pCtx[i].functionId;

    if (functionId == FUNCTION_TAG_DUMMY || functionId == FUNCTION_TS_DUMMY) {
H
Haojun Liao 已提交
1087
      tagLen += pCtx[i].resDataInfo.bytes;
1088
      pTagCtx[num++] = &pCtx[i];
L
Liu Jicong 已提交
1089
    } else if (1 /*(aAggs[functionId].status & FUNCSTATE_SELECTIVITY) != 0*/) {
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
      p = &pCtx[i];
    } else if (functionId == FUNCTION_TS || functionId == FUNCTION_TAG) {
      // tag function may be the group by tag column
      // ts may be the required primary timestamp column
      continue;
    } else {
      // the column may be the normal column, group by normal_column, the functionId is FUNCTION_PRJ
    }
  }
  if (p != NULL) {
1100 1101
    p->subsidiaries.pCtx = pTagCtx;
    p->subsidiaries.num = num;
1102
  } else {
wafwerar's avatar
wafwerar 已提交
1103
    taosMemoryFreeClear(pTagCtx);
1104 1105 1106 1107 1108
  }

  return TSDB_CODE_SUCCESS;
}

1109
SqlFunctionCtx* createSqlFunctionCtx(SExprInfo* pExprInfo, int32_t numOfOutput, int32_t** rowCellInfoOffset) {
L
Liu Jicong 已提交
1110
  SqlFunctionCtx* pFuncCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfOutput, sizeof(SqlFunctionCtx));
H
Haojun Liao 已提交
1111 1112 1113 1114
  if (pFuncCtx == NULL) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
1115
  *rowCellInfoOffset = taosMemoryCalloc(numOfOutput, sizeof(int32_t));
H
Haojun Liao 已提交
1116
  if (*rowCellInfoOffset == 0) {
wafwerar's avatar
wafwerar 已提交
1117
    taosMemoryFreeClear(pFuncCtx);
H
Haojun Liao 已提交
1118 1119 1120 1121
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
1122
    SExprInfo* pExpr = &pExprInfo[i];
H
Haojun Liao 已提交
1123

L
Liu Jicong 已提交
1124
    SExprBasicInfo* pFunct = &pExpr->base;
H
Haojun Liao 已提交
1125
    SqlFunctionCtx* pCtx = &pFuncCtx[i];
H
Haojun Liao 已提交
1126

1127
    pCtx->functionId = -1;
1128
    pCtx->curBufPage = -1;
L
Liu Jicong 已提交
1129
    pCtx->pExpr = pExpr;
1130

H
Haojun Liao 已提交
1131
    if (pExpr->pExpr->nodeType == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
1132
      SFuncExecEnv env = {0};
H
Haojun Liao 已提交
1133 1134
      pCtx->functionId = pExpr->pExpr->_function.pFunctNode->funcId;

H
Haojun Liao 已提交
1135
      if (fmIsAggFunc(pCtx->functionId) || fmIsNonstandardSQLFunc(pCtx->functionId)) {
1136 1137 1138 1139
        bool isUdaf = fmIsUserDefinedFunc(pCtx->functionId);
        if (!isUdaf) {
          fmGetFuncExecFuncs(pCtx->functionId, &pCtx->fpSet);
        } else {
1140
          char* udfName = pExpr->pExpr->_function.pFunctNode->functionName;
1141 1142 1143
          strncpy(pCtx->udfName, udfName, strlen(udfName));
          fmGetUdafExecFuncs(pCtx->functionId, &pCtx->fpSet);
        }
1144 1145 1146
        pCtx->fpSet.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
      } else {
        fmGetScalarFuncExecFuncs(pCtx->functionId, &pCtx->sfp);
1147 1148 1149
        if (pCtx->sfp.getEnv != NULL) {
          pCtx->sfp.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
        }
1150
      }
H
Haojun Liao 已提交
1151
      pCtx->resDataInfo.interBufSize = env.calcMemSize;
1152 1153 1154 1155
    } else if (pExpr->pExpr->nodeType == QUERY_NODE_COLUMN || pExpr->pExpr->nodeType == QUERY_NODE_OPERATOR ||
               pExpr->pExpr->nodeType == QUERY_NODE_VALUE) {
      // for simple column, the intermediate buffer needs to hold one element.
      pCtx->resDataInfo.interBufSize = pFunct->resSchema.bytes;
H
Haojun Liao 已提交
1156
    }
H
Haojun Liao 已提交
1157

H
Haojun Liao 已提交
1158
    pCtx->input.numOfInputCols = pFunct->numOfParams;
wafwerar's avatar
wafwerar 已提交
1159 1160
    pCtx->input.pData = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
    pCtx->input.pColumnDataAgg = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
H
Haojun Liao 已提交
1161

1162
    pCtx->pTsOutput = NULL;
L
Liu Jicong 已提交
1163
    pCtx->resDataInfo.bytes = pFunct->resSchema.bytes;
1164
    pCtx->resDataInfo.type = pFunct->resSchema.type;
L
Liu Jicong 已提交
1165
    pCtx->order = TSDB_ORDER_ASC;
1166
    pCtx->start.key = INT64_MIN;
L
Liu Jicong 已提交
1167
    pCtx->end.key = INT64_MIN;
1168
    pCtx->numOfParams = pExpr->base.numOfParams;
H
Haojun Liao 已提交
1169

1170
    pCtx->param = pFunct->pParam;
dengyihao's avatar
dengyihao 已提交
1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
    //    for (int32_t j = 0; j < pCtx->numOfParams; ++j) {
    //      // set the order information for top/bottom query
    //      int32_t functionId = pCtx->functionId;
    //      if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF) {
    //        int32_t f = getExprFunctionId(&pExpr[0]);
    //        assert(f == FUNCTION_TS || f == FUNCTION_TS_DUMMY);
    //
    //        //      pCtx->param[2].i = pQueryAttr->order.order;
    //        //      pCtx->param[2].nType = TSDB_DATA_TYPE_BIGINT;
    //        //      pCtx->param[3].i = functionId;
    //        //      pCtx->param[3].nType = TSDB_DATA_TYPE_BIGINT;
    //
    //        //      pCtx->param[1].i = pQueryAttr->order.col.info.colId;
    //      } else if (functionId == FUNCTION_INTERP) {
    //        //      pCtx->param[2].i = (int8_t)pQueryAttr->fillType;
    //        //      if (pQueryAttr->fillVal != NULL) {
    //        //        if (isNull((const char *)&pQueryAttr->fillVal[i], pCtx->inputType)) {
    //        //          pCtx->param[1].nType = TSDB_DATA_TYPE_NULL;
    //        //        } else {  // todo refactor, taosVariantCreateFromBinary should handle the NULL value
    //        //          if (pCtx->inputType != TSDB_DATA_TYPE_BINARY && pCtx->inputType != TSDB_DATA_TYPE_NCHAR) {
    //        //            taosVariantCreateFromBinary(&pCtx->param[1], (char *)&pQueryAttr->fillVal[i],
    //        pCtx->inputBytes, pCtx->inputType);
    //        //          }
    //        //        }
    //        //      }
    //      } else if (functionId == FUNCTION_TWA) {
    //        //      pCtx->param[1].i = pQueryAttr->window.skey;
    //        //      pCtx->param[1].nType = TSDB_DATA_TYPE_BIGINT;
    //        //      pCtx->param[2].i = pQueryAttr->window.ekey;
    //        //      pCtx->param[2].nType = TSDB_DATA_TYPE_BIGINT;
    //      } else if (functionId == FUNCTION_ARITHM) {
    //        //      pCtx->param[1].pz = (char*) getScalarFuncSupport(pRuntimeEnv->scalarSup, i);
    //      }
    //    }
H
Haojun Liao 已提交
1205 1206
  }

L
Liu Jicong 已提交
1207 1208 1209
  for (int32_t i = 1; i < numOfOutput; ++i) {
    (*rowCellInfoOffset)[i] =
        (int32_t)((*rowCellInfoOffset)[i - 1] + sizeof(SResultRowEntryInfo) + pFuncCtx[i - 1].resDataInfo.interBufSize);
H
Haojun Liao 已提交
1210
  }
H
Haojun Liao 已提交
1211 1212 1213 1214 1215

  setCtxTagColumnInfo(pFuncCtx, numOfOutput);
  return pFuncCtx;
}

1216
static void* destroySqlFunctionCtx(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1217 1218 1219 1220 1221 1222
  if (pCtx == NULL) {
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
    for (int32_t j = 0; j < pCtx[i].numOfParams; ++j) {
1223
      taosVariantDestroy(&pCtx[i].param[j].param);
1224 1225 1226
    }

    taosVariantDestroy(&pCtx[i].tag);
1227
    taosMemoryFreeClear(pCtx[i].subsidiaries.pCtx);
1228 1229
  }

wafwerar's avatar
wafwerar 已提交
1230
  taosMemoryFreeClear(pCtx);
1231 1232 1233
  return NULL;
}

L
Liu Jicong 已提交
1234
bool isTaskKilled(SExecTaskInfo* pTaskInfo) {
1235 1236
  // query has been executed more than tsShellActivityTimer, and the retrieve has not arrived
  // abort current query execution.
L
Liu Jicong 已提交
1237 1238
  if (pTaskInfo->owner != 0 &&
      ((taosGetTimestampSec() - pTaskInfo->cost.start / 1000) > 10 * getMaximumIdleDurationSec())
1239 1240
      /*(!needBuildResAfterQueryComplete(pTaskInfo))*/) {
    assert(pTaskInfo->cost.start != 0);
L
Liu Jicong 已提交
1241 1242 1243
    //    qDebug("QInfo:%" PRIu64 " retrieve not arrive beyond %d ms, abort current query execution, start:%" PRId64
    //           ", current:%d", pQInfo->qId, 1, pQInfo->startExecTs, taosGetTimestampSec());
    //    return true;
1244 1245 1246 1247 1248
  }

  return false;
}

L
Liu Jicong 已提交
1249
void setTaskKilled(SExecTaskInfo* pTaskInfo) { pTaskInfo->code = TSDB_CODE_TSC_QUERY_CANCELLED; }
1250

L
Liu Jicong 已提交
1251
static bool isCachedLastQuery(STaskAttr* pQueryAttr) {
1252 1253 1254 1255 1256 1257 1258 1259 1260
  for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
    int32_t functionId = getExprFunctionId(&pQueryAttr->pExpr1[i]);
    if (functionId == FUNCTION_LAST || functionId == FUNCTION_LAST_DST) {
      continue;
    }

    return false;
  }

1261 1262
  int32_t order = TSDB_ORDER_ASC;
  if (order != TSDB_ORDER_DESC || !TSWINDOW_IS_EQUAL(pQueryAttr->window, TSWINDOW_DESC_INITIALIZER)) {
1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281
    return false;
  }

  if (pQueryAttr->groupbyColumn) {
    return false;
  }

  if (pQueryAttr->interval.interval > 0) {
    return false;
  }

  if (pQueryAttr->numOfFilterCols > 0 || pQueryAttr->havingNum > 0) {
    return false;
  }

  return true;
}

/////////////////////////////////////////////////////////////////////////////////////////////
L
Liu Jicong 已提交
1282
// todo refactor : return window
1283
void getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int64_t key, STimeWindow* win) {
H
Haojun Liao 已提交
1284
  win->skey = taosTimeTruncate(key, pInterval, precision);
1285 1286

  /*
H
Haojun Liao 已提交
1287
   * if the realSkey > INT64_MAX - pInterval->interval, the query duration between
1288 1289
   * realSkey and realEkey must be less than one interval.Therefore, no need to adjust the query ranges.
   */
1290 1291
  win->ekey = taosTimeAdd(win->skey, pInterval->interval, pInterval->intervalUnit, precision) - 1;
  if (win->ekey < win->skey) {
1292 1293 1294 1295
    win->ekey = INT64_MAX;
  }
}

L
Liu Jicong 已提交
1296
static int32_t updateBlockLoadStatus(STaskAttr* pQuery, int32_t status) {
1297 1298 1299
  bool hasFirstLastFunc = false;
  bool hasOtherFunc = false;

1300
  if (status == BLK_DATA_DATA_LOAD || status == BLK_DATA_FILTEROUT) {
1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
    return status;
  }

  for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
    int32_t functionId = getExprFunctionId(&pQuery->pExpr1[i]);

    if (functionId == FUNCTION_TS || functionId == FUNCTION_TS_DUMMY || functionId == FUNCTION_TAG ||
        functionId == FUNCTION_TAG_DUMMY) {
      continue;
    }

    if (functionId == FUNCTION_FIRST_DST || functionId == FUNCTION_LAST_DST) {
      hasFirstLastFunc = true;
    } else {
      hasOtherFunc = true;
    }
  }

1319
  if (hasFirstLastFunc && status == BLK_DATA_NOT_LOAD) {
L
Liu Jicong 已提交
1320
    if (!hasOtherFunc) {
1321
      return BLK_DATA_FILTEROUT;
1322
    } else {
1323
      return BLK_DATA_DATA_LOAD;
1324 1325 1326 1327 1328 1329
    }
  }

  return status;
}

L
Liu Jicong 已提交
1330 1331
// static void updateDataCheckOrder(SQInfo *pQInfo, SQueryTableReq* pQueryMsg, bool stableQuery) {
//   STaskAttr* pQueryAttr = pQInfo->runtimeEnv.pQueryAttr;
H
Haojun Liao 已提交
1332
//
L
Liu Jicong 已提交
1333 1334 1335 1336
//   // in case of point-interpolation query, use asc order scan
//   char msg[] = "QInfo:0x%"PRIx64" scan order changed for %s query, old:%d, new:%d, qrange exchanged, old qrange:%"
//   PRId64
//                "-%" PRId64 ", new qrange:%" PRId64 "-%" PRId64;
H
Haojun Liao 已提交
1337
//
L
Liu Jicong 已提交
1338 1339 1340 1341 1342
//   // todo handle the case the the order irrelevant query type mixed up with order critical query type
//   // descending order query for last_row query
//   if (isFirstLastRowQuery(pQueryAttr)) {
//     //qDebug("QInfo:0x%"PRIx64" scan order changed for last_row query, old:%d, new:%d", pQInfo->qId,
//     pQueryAttr->order.order, TSDB_ORDER_ASC);
H
Haojun Liao 已提交
1343
//
L
Liu Jicong 已提交
1344 1345
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     if (pQueryAttr->window.skey > pQueryAttr->window.ekey) {
wafwerar's avatar
wafwerar 已提交
1346
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1347
//     }
H
Haojun Liao 已提交
1348
//
L
Liu Jicong 已提交
1349 1350 1351
//     pQueryAttr->needReverseScan = false;
//     return;
//   }
H
Haojun Liao 已提交
1352
//
L
Liu Jicong 已提交
1353 1354 1355
//   if (pQueryAttr->groupbyColumn && pQueryAttr->order.order == TSDB_ORDER_DESC) {
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     if (pQueryAttr->window.skey > pQueryAttr->window.ekey) {
wafwerar's avatar
wafwerar 已提交
1356
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1357
//     }
H
Haojun Liao 已提交
1358
//
L
Liu Jicong 已提交
1359 1360 1361 1362
//     pQueryAttr->needReverseScan = false;
//     doUpdateLastKey(pQueryAttr);
//     return;
//   }
H
Haojun Liao 已提交
1363
//
L
Liu Jicong 已提交
1364 1365 1366 1367 1368 1369
//   if (pQueryAttr->pointInterpQuery && pQueryAttr->interval.interval == 0) {
//     if (!QUERY_IS_ASC_QUERY(pQueryAttr)) {
//       //qDebug(msg, pQInfo->qId, "interp", pQueryAttr->order.order, TSDB_ORDER_ASC, pQueryAttr->window.skey,
//       pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey); TSWAP(pQueryAttr->window.skey,
//       pQueryAttr->window.ekey, TSKEY);
//     }
H
Haojun Liao 已提交
1370
//
L
Liu Jicong 已提交
1371 1372 1373
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     return;
//   }
H
Haojun Liao 已提交
1374
//
L
Liu Jicong 已提交
1375 1376 1377 1378
//   if (pQueryAttr->interval.interval == 0) {
//     if (onlyFirstQuery(pQueryAttr)) {
//       if (!QUERY_IS_ASC_QUERY(pQueryAttr)) {
//         //qDebug(msg, pQInfo->qId, "only-first", pQueryAttr->order.order, TSDB_ORDER_ASC, pQueryAttr->window.skey,
H
Haojun Liao 已提交
1379 1380
////               pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey);
//
wafwerar's avatar
wafwerar 已提交
1381
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
//        doUpdateLastKey(pQueryAttr);
//      }
//
//      pQueryAttr->order.order = TSDB_ORDER_ASC;
//      pQueryAttr->needReverseScan = false;
//    } else if (onlyLastQuery(pQueryAttr) && notContainSessionOrStateWindow(pQueryAttr)) {
//      if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//        //qDebug(msg, pQInfo->qId, "only-last", pQueryAttr->order.order, TSDB_ORDER_DESC, pQueryAttr->window.skey,
////               pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey);
//
wafwerar's avatar
wafwerar 已提交
1392
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404
//        doUpdateLastKey(pQueryAttr);
//      }
//
//      pQueryAttr->order.order = TSDB_ORDER_DESC;
//      pQueryAttr->needReverseScan = false;
//    }
//
//  } else {  // interval query
//    if (stableQuery) {
//      if (onlyFirstQuery(pQueryAttr)) {
//        if (!QUERY_IS_ASC_QUERY(pQueryAttr)) {
//          //qDebug(msg, pQInfo->qId, "only-first stable", pQueryAttr->order.order, TSDB_ORDER_ASC,
L
Liu Jicong 已提交
1405 1406
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1407
//
wafwerar's avatar
wafwerar 已提交
1408
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1409 1410 1411 1412 1413 1414 1415 1416
//          doUpdateLastKey(pQueryAttr);
//        }
//
//        pQueryAttr->order.order = TSDB_ORDER_ASC;
//        pQueryAttr->needReverseScan = false;
//      } else if (onlyLastQuery(pQueryAttr)) {
//        if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//          //qDebug(msg, pQInfo->qId, "only-last stable", pQueryAttr->order.order, TSDB_ORDER_DESC,
L
Liu Jicong 已提交
1417 1418
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1419
//
wafwerar's avatar
wafwerar 已提交
1420
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1421 1422 1423 1424 1425 1426 1427 1428 1429
//          doUpdateLastKey(pQueryAttr);
//        }
//
//        pQueryAttr->order.order = TSDB_ORDER_DESC;
//        pQueryAttr->needReverseScan = false;
//      }
//    }
//  }
//}
1430

H
Haojun Liao 已提交
1431 1432
static void getIntermediateBufInfo(STaskRuntimeEnv* pRuntimeEnv, int32_t* ps, int32_t* rowsize) {
  STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
L
Liu Jicong 已提交
1433
  int32_t    MIN_ROWS_PER_PAGE = 4;
1434

L
Liu Jicong 已提交
1435 1436
  *rowsize = (int32_t)(pQueryAttr->resultRowSize *
                       getRowNumForMultioutput(pQueryAttr, pQueryAttr->topBotQuery, pQueryAttr->stableQuery));
1437 1438 1439 1440
  int32_t overhead = sizeof(SFilePage);

  // one page contains at least two rows
  *ps = DEFAULT_INTERN_BUF_PAGE_SIZE;
L
Liu Jicong 已提交
1441
  while (((*rowsize) * MIN_ROWS_PER_PAGE) > (*ps) - overhead) {
1442 1443 1444 1445 1446 1447
    *ps = ((*ps) << 1u);
  }
}

#define IS_PREFILTER_TYPE(_t) ((_t) != TSDB_DATA_TYPE_BINARY && (_t) != TSDB_DATA_TYPE_NCHAR)

L
Liu Jicong 已提交
1448 1449 1450
// static FORCE_INLINE bool doFilterByBlockStatistics(STaskRuntimeEnv* pRuntimeEnv, SDataStatis *pDataStatis,
// SqlFunctionCtx *pCtx, int32_t numOfRows) {
//   STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
1451
//
L
Liu Jicong 已提交
1452 1453 1454
//   if (pDataStatis == NULL || pQueryAttr->pFilters == NULL) {
//     return true;
//   }
1455
//
L
Liu Jicong 已提交
1456 1457
//   return filterRangeExecute(pQueryAttr->pFilters, pDataStatis, pQueryAttr->numOfCols, numOfRows);
// }
1458

H
Haojun Liao 已提交
1459
static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
1460 1461
  STimeWindow w = {0};

dengyihao's avatar
dengyihao 已提交
1462 1463
  TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
  TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
1464

1465
  if (true) {
L
Liu Jicong 已提交
1466
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
1467 1468 1469 1470 1471 1472
    assert(w.ekey >= pBlockInfo->window.skey);

    if (w.ekey < pBlockInfo->window.ekey) {
      return true;
    }

L
Liu Jicong 已提交
1473 1474
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1475 1476 1477 1478 1479 1480 1481 1482 1483 1484
      if (w.skey > pBlockInfo->window.ekey) {
        break;
      }

      assert(w.ekey > pBlockInfo->window.ekey);
      if (w.skey <= pBlockInfo->window.ekey && w.skey > pBlockInfo->window.skey) {
        return true;
      }
    }
  } else {
L
Liu Jicong 已提交
1485
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
1486 1487 1488 1489 1490 1491
    assert(w.skey <= pBlockInfo->window.ekey);

    if (w.skey > pBlockInfo->window.skey) {
      return true;
    }

L
Liu Jicong 已提交
1492 1493
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
      if (w.ekey < pBlockInfo->window.skey) {
        break;
      }

      assert(w.skey < pBlockInfo->window.skey);
      if (w.ekey < pBlockInfo->window.ekey && w.ekey >= pBlockInfo->window.skey) {
        return true;
      }
    }
  }

  return false;
}

void doCompactSDataBlock(SSDataBlock* pBlock, int32_t numOfRows, int8_t* p) {
  int32_t len = 0;
  int32_t start = 0;
  for (int32_t j = 0; j < numOfRows; ++j) {
    if (p[j] == 1) {
      len++;
    } else {
      if (len > 0) {
        int32_t cstart = j - len;
        for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
          SColumnInfoData* pColumnInfoData = taosArrayGet(pBlock->pDataBlock, i);

          int16_t bytes = pColumnInfoData->info.bytes;
          memmove(((char*)pColumnInfoData->pData) + start * bytes, pColumnInfoData->pData + cstart * bytes,
                  len * bytes);
        }

        start += len;
        len = 0;
      }
    }
  }

  if (len > 0) {
    int32_t cstart = numOfRows - len;
    for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
      SColumnInfoData* pColumnInfoData = taosArrayGet(pBlock->pDataBlock, i);

      int16_t bytes = pColumnInfoData->info.bytes;
      memmove(pColumnInfoData->pData + start * bytes, pColumnInfoData->pData + cstart * bytes, len * bytes);
    }

    start += len;
    len = 0;
  }

  pBlock->info.rows = start;
  pBlock->pBlockAgg = NULL;  // clean the block statistics info

  if (start > 0) {
    SColumnInfoData* pColumnInfoData = taosArrayGet(pBlock->pDataBlock, 0);
    if (pColumnInfoData->info.type == TSDB_DATA_TYPE_TIMESTAMP &&
        pColumnInfoData->info.colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
      pBlock->info.window.skey = *(int64_t*)pColumnInfoData->pData;
      pBlock->info.window.ekey = *(int64_t*)(pColumnInfoData->pData + TSDB_KEYSIZE * (start - 1));
    }
  }
}

static SColumnInfo* doGetTagColumnInfoById(SColumnInfo* pTagColList, int32_t numOfTags, int16_t colId);
L
Liu Jicong 已提交
1558
static void         doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes);
1559 1560

static uint32_t doFilterByBlockTimeWindow(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock) {
H
Haojun Liao 已提交
1561
  SqlFunctionCtx* pCtx = pTableScanInfo->pCtx;
1562
  uint32_t        status = BLK_DATA_NOT_LOAD;
1563 1564 1565 1566

  int32_t numOfOutput = pTableScanInfo->numOfOutput;
  for (int32_t i = 0; i < numOfOutput; ++i) {
    int32_t functionId = pCtx[i].functionId;
H
Haojun Liao 已提交
1567
    int32_t colId = pTableScanInfo->pExpr[i].base.pParam[0].pCol->colId;
1568 1569 1570

    // group by + first/last should not apply the first/last block filter
    if (functionId < 0) {
1571
      status |= BLK_DATA_DATA_LOAD;
1572 1573
      return status;
    } else {
L
Liu Jicong 已提交
1574
      //      status |= aAggs[functionId].dataReqFunc(&pTableScanInfo->pCtx[i], &pBlock->info.window, colId);
1575
      //      if ((status & BLK_DATA_DATA_LOAD) == BLK_DATA_DATA_LOAD) {
L
Liu Jicong 已提交
1576 1577
      //        return status;
      //      }
1578 1579 1580 1581 1582 1583
    }
  }

  return status;
}

L
Liu Jicong 已提交
1584 1585
int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                              uint32_t* status) {
1586
  *status = BLK_DATA_NOT_LOAD;
1587

H
Haojun Liao 已提交
1588
  pBlock->pDataBlock = NULL;
L
Liu Jicong 已提交
1589
  pBlock->pBlockAgg = NULL;
H
Haojun Liao 已提交
1590

L
Liu Jicong 已提交
1591 1592
  //  int64_t groupId = pRuntimeEnv->current->groupIndex;
  //  bool    ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
1593

H
Haojun Liao 已提交
1594
  STaskCostInfo* pCost = &pTaskInfo->cost;
1595

1596 1597
//  pCost->totalBlocks += 1;
//  pCost->totalRows += pBlock->info.rows;
H
Haojun Liao 已提交
1598
#if 0
1599 1600 1601
  // Calculate all time windows that are overlapping or contain current data block.
  // If current data block is contained by all possible time window, do not load current data block.
  if (/*pQueryAttr->pFilters || */pQueryAttr->groupbyColumn || pQueryAttr->sw.gap > 0 ||
H
Haojun Liao 已提交
1602
      (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && overlapWithTimeWindow(pTaskInfo, &pBlock->info))) {
1603
    (*status) = BLK_DATA_DATA_LOAD;
1604 1605 1606
  }

  // check if this data block is required to load
1607
  if ((*status) != BLK_DATA_DATA_LOAD) {
1608 1609 1610 1611 1612 1613 1614
    bool needFilter = true;

    // the pCtx[i] result is belonged to previous time window since the outputBuf has not been set yet,
    // the filter result may be incorrect. So in case of interval query, we need to set the correct time output buffer
    if (QUERY_IS_INTERVAL_QUERY(pQueryAttr)) {
      SResultRow* pResult = NULL;

H
Haojun Liao 已提交
1615
      bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638
      TSKEY k = ascQuery? pBlock->info.window.skey : pBlock->info.window.ekey;

      STimeWindow win = getActiveTimeWindow(pTableScanInfo->pResultRowInfo, k, pQueryAttr);
      if (pQueryAttr->pointInterpQuery) {
        needFilter = chkWindowOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, &win, masterScan, &pResult, groupId,
                                    pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
                                    pTableScanInfo->rowCellInfoOffset);
      } else {
        if (setResultOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pBlock->info.uid, &win, masterScan, &pResult, groupId,
                                    pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
                                    pTableScanInfo->rowCellInfoOffset) != TSDB_CODE_SUCCESS) {
          longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
        }
      }
    } else if (pQueryAttr->stableQuery && (!pQueryAttr->tsCompQuery) && (!pQueryAttr->diffQuery)) { // stable aggregate, not interval aggregate or normal column aggregate
      doSetTableGroupOutputBuf(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pTableScanInfo->pCtx,
                               pTableScanInfo->rowCellInfoOffset, pTableScanInfo->numOfOutput,
                               pRuntimeEnv->current->groupIndex);
    }

    if (needFilter) {
      (*status) = doFilterByBlockTimeWindow(pTableScanInfo, pBlock);
    } else {
1639
      (*status) = BLK_DATA_DATA_LOAD;
1640 1641 1642 1643
    }
  }

  SDataBlockInfo* pBlockInfo = &pBlock->info;
H
Haojun Liao 已提交
1644
//  *status = updateBlockLoadStatus(pRuntimeEnv->pQueryAttr, *status);
1645

1646
  if ((*status) == BLK_DATA_NOT_LOAD || (*status) == BLK_DATA_FILTEROUT) {
1647 1648
    //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//           pBlockInfo->window.ekey, pBlockInfo->rows);
1649
    pCost->skipBlocks += 1;
1650
  } else if ((*status) == BLK_DATA_SMA_LOAD) {
1651 1652
    // this function never returns error?
    pCost->loadBlockStatis += 1;
1653
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1654 1655

    if (pBlock->pBlockAgg == NULL) {  // data block statistics does not exist, load data block
1656
//      pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1657 1658 1659
      pCost->totalCheckedRows += pBlock->info.rows;
    }
  } else {
1660
    assert((*status) == BLK_DATA_DATA_LOAD);
1661 1662 1663

    // load the data block statistics to perform further filter
    pCost->loadBlockStatis += 1;
1664
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1665 1666 1667 1668 1669 1670

    if (pQueryAttr->topBotQuery && pBlock->pBlockAgg != NULL) {
      { // set previous window
        if (QUERY_IS_INTERVAL_QUERY(pQueryAttr)) {
          SResultRow* pResult = NULL;

H
Haojun Liao 已提交
1671
          bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688
          TSKEY k = ascQuery? pBlock->info.window.skey : pBlock->info.window.ekey;

          STimeWindow win = getActiveTimeWindow(pTableScanInfo->pResultRowInfo, k, pQueryAttr);
          if (setResultOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pBlock->info.uid, &win, masterScan, &pResult, groupId,
                                      pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
                                      pTableScanInfo->rowCellInfoOffset) != TSDB_CODE_SUCCESS) {
            longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
          }
        }
      }
      bool load = false;
      for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
        int32_t functionId = pTableScanInfo->pCtx[i].functionId;
        if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM) {
//          load = topbot_datablock_filter(&pTableScanInfo->pCtx[i], (char*)&(pBlock->pBlockAgg[i].min),
//                                         (char*)&(pBlock->pBlockAgg[i].max));
          if (!load) { // current block has been discard due to filter applied
1689
            pCost->skipBlocks += 1;
1690 1691
            //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId,
//                   pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
1692
            (*status) = BLK_DATA_FILTEROUT;
1693 1694 1695 1696 1697 1698 1699 1700
            return TSDB_CODE_SUCCESS;
          }
        }
      }
    }

    // current block has been discard due to filter applied
//    if (!doFilterByBlockStatistics(pRuntimeEnv, pBlock->pBlockAgg, pTableScanInfo->pCtx, pBlockInfo->rows)) {
1701
//      pCost->skipBlocks += 1;
1702 1703
//      qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//             pBlockInfo->window.ekey, pBlockInfo->rows);
1704
//      (*status) = BLK_DATA_FILTEROUT;
1705 1706 1707 1708 1709
//      return TSDB_CODE_SUCCESS;
//    }

    pCost->totalCheckedRows += pBlockInfo->rows;
    pCost->loadBlocks += 1;
1710
//    pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1711 1712 1713 1714 1715 1716 1717
//    if (pBlock->pDataBlock == NULL) {
//      return terrno;
//    }

//    if (pQueryAttr->pFilters != NULL) {
//      filterSetColFieldData(pQueryAttr->pFilters, pBlock->info.numOfCols, pBlock->pDataBlock);
//    }
1718

1719 1720 1721 1722
//    if (pQueryAttr->pFilters != NULL || pRuntimeEnv->pTsBuf != NULL) {
//      filterColRowsInDataBlock(pRuntimeEnv, pBlock, ascQuery);
//    }
  }
H
Haojun Liao 已提交
1723
#endif
1724 1725 1726 1727
  return TSDB_CODE_SUCCESS;
}

/*
H
Haojun Liao 已提交
1728
 * set tag value in SqlFunctionCtx
1729 1730
 * e.g.,tag information into input buffer
 */
L
Liu Jicong 已提交
1731
static void doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes) {
1732 1733 1734
  taosVariantDestroy(tag);

  char* val = NULL;
L
Liu Jicong 已提交
1735 1736 1737 1738 1739 1740
  //  if (tagColId == TSDB_TBNAME_COLUMN_INDEX) {
  //    val = tsdbGetTableName(pTable);
  //    assert(val != NULL);
  //  } else {
  //    val = tsdbGetTableTagVal(pTable, tagColId, type, bytes);
  //  }
1741 1742 1743 1744 1745 1746 1747 1748

  if (val == NULL || isNull(val, type)) {
    tag->nType = TSDB_DATA_TYPE_NULL;
    return;
  }

  if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR) {
    int32_t maxLen = bytes - VARSTR_HEADER_SIZE;
L
Liu Jicong 已提交
1749
    int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
1750
    taosVariantCreateFromBinary(tag, varDataVal(val), len, type);
L
Liu Jicong 已提交
1751
    // taosVariantCreateFromBinary(tag, varDataVal(val), varDataLen(val), type);
1752 1753 1754 1755 1756 1757 1758 1759
  } else {
    taosVariantCreateFromBinary(tag, val, bytes, type);
  }
}

static SColumnInfo* doGetTagColumnInfoById(SColumnInfo* pTagColList, int32_t numOfTags, int16_t colId) {
  assert(pTagColList != NULL && numOfTags > 0);

L
Liu Jicong 已提交
1760
  for (int32_t i = 0; i < numOfTags; ++i) {
1761 1762 1763 1764 1765 1766 1767 1768
    if (pTagColList[i].colId == colId) {
      return &pTagColList[i];
    }
  }

  return NULL;
}

L
Liu Jicong 已提交
1769 1770
void setTagValue(SOperatorInfo* pOperatorInfo, void* pTable, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
  SExprInfo* pExpr = pOperatorInfo->pExpr;
1771
  SExprInfo* pExprInfo = &pExpr[0];
L
Liu Jicong 已提交
1772
  int32_t    functionId = getExprFunctionId(pExprInfo);
1773
#if 0
1774 1775 1776
  if (pQueryAttr->numOfOutput == 1 && functionId == FUNCTION_TS_COMP && pQueryAttr->stableQuery) {
    assert(pExprInfo->base.numOfParams == 1);

L
Liu Jicong 已提交
1777
    //    int16_t      tagColId = (int16_t)pExprInfo->base.param[0].i;
H
Haojun Liao 已提交
1778
    int16_t      tagColId = -1;
1779 1780 1781
    SColumnInfo* pColInfo = doGetTagColumnInfoById(pQueryAttr->tagColList, pQueryAttr->numOfTags, tagColId);

    doSetTagValueInParam(pTable, tagColId, &pCtx[0].tag, pColInfo->type, pColInfo->bytes);
H
Haojun Liao 已提交
1782

1783 1784 1785 1786 1787 1788 1789 1790 1791
  } else {
    // set tag value, by which the results are aggregated.
    int32_t offset = 0;
    memset(pRuntimeEnv->tagVal, 0, pQueryAttr->tagLen);

    for (int32_t idx = 0; idx < numOfOutput; ++idx) {
      SExprInfo* pLocalExprInfo = &pExpr[idx];

      // ts_comp column required the tag value for join filter
H
Haojun Liao 已提交
1792
      if (!TSDB_COL_IS_TAG(pLocalExprInfo->base.pParam[0].pCol->flag)) {
1793 1794 1795 1796
        continue;
      }

      // todo use tag column index to optimize performance
L
Liu Jicong 已提交
1797 1798
      doSetTagValueInParam(pTable, pLocalExprInfo->base.pParam[0].pCol->colId, &pCtx[idx].tag,
                           pLocalExprInfo->base.resSchema.type, pLocalExprInfo->base.resSchema.bytes);
1799

L
Liu Jicong 已提交
1800 1801 1802
      if (IS_NUMERIC_TYPE(pLocalExprInfo->base.resSchema.type) ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_BOOL ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_TIMESTAMP) {
1803 1804 1805 1806
        memcpy(pRuntimeEnv->tagVal + offset, &pCtx[idx].tag.i, pLocalExprInfo->base.resSchema.bytes);
      } else {
        if (pCtx[idx].tag.pz != NULL) {
          memcpy(pRuntimeEnv->tagVal + offset, pCtx[idx].tag.pz, pCtx[idx].tag.nLen);
L
Liu Jicong 已提交
1807
        }
1808 1809 1810 1811 1812 1813 1814 1815 1816 1817
      }

      offset += pLocalExprInfo->base.resSchema.bytes;
    }
  }

  // set the tsBuf start position before check each data block
  if (pRuntimeEnv->pTsBuf != NULL) {
    setCtxTagForJoin(pRuntimeEnv, &pCtx[0], pExprInfo, pTable);
  }
1818
#endif
1819 1820
}

H
Haojun Liao 已提交
1821
void copyToSDataBlock(SSDataBlock* pBlock, int32_t* offset, SGroupResInfo* pGroupResInfo, SDiskbasedBuf* pResBuf) {
1822 1823 1824 1825 1826 1827 1828
  pBlock->info.rows = 0;

  int32_t code = TSDB_CODE_SUCCESS;
  while (pGroupResInfo->currentGroup < pGroupResInfo->totalGroup) {
    // all results in current group have been returned to client, try next group
    if ((pGroupResInfo->pRows == NULL) || taosArrayGetSize(pGroupResInfo->pRows) == 0) {
      assert(pGroupResInfo->index == 0);
L
Liu Jicong 已提交
1829 1830 1831
      //      if ((code = mergeIntoGroupResult(&pGroupResInfo, pRuntimeEnv, offset)) != TSDB_CODE_SUCCESS) {
      return;
      //      }
1832 1833
    }

L
Liu Jicong 已提交
1834
    //    doCopyToSDataBlock(pResBuf, pGroupResInfo, TSDB_ORDER_ASC, pBlock, );
1835 1836 1837 1838 1839 1840 1841 1842 1843 1844

    // current data are all dumped to result buffer, clear it
    if (!hasRemainDataInCurrentGroup(pGroupResInfo)) {
      cleanupGroupResInfo(pGroupResInfo);
      if (!incNextGroup(pGroupResInfo)) {
        break;
      }
    }

    // enough results in data buffer, return
L
Liu Jicong 已提交
1845 1846 1847
    //    if (pBlock->info.rows >= threshold) {
    //      break;
    //    }
1848 1849 1850
  }
}

L
Liu Jicong 已提交
1851
static void updateTableQueryInfoForReverseScan(STableQueryInfo* pTableQueryInfo) {
1852 1853 1854 1855
  if (pTableQueryInfo == NULL) {
    return;
  }

wafwerar's avatar
wafwerar 已提交
1856
  //  TSWAP(pTableQueryInfo->win.skey, pTableQueryInfo->win.ekey);
L
Liu Jicong 已提交
1857
  //  pTableQueryInfo->lastKey = pTableQueryInfo->win.skey;
1858

L
Liu Jicong 已提交
1859 1860
  //  SWITCH_ORDER(pTableQueryInfo->cur.order);
  //  pTableQueryInfo->cur.vgroupIndex = -1;
1861 1862

  // set the index to be the end slot of result rows array
dengyihao's avatar
dengyihao 已提交
1863 1864 1865 1866 1867 1868
  //  SResultRowInfo* pResultRowInfo = &pTableQueryInfo->resInfo;
  //  if (pResultRowInfo->size > 0) {
  //    pResultRowInfo->curPos = pResultRowInfo->size - 1;
  //  } else {
  //    pResultRowInfo->curPos = -1;
  //  }
1869 1870
}

H
Haojun Liao 已提交
1871
void initResultRow(SResultRow* pResultRow) {
1872 1873 1874 1875 1876 1877
  pResultRow->pEntryInfo = (struct SResultRowEntryInfo*)((char*)pResultRow + sizeof(SResultRow));
}

/*
 * The start of each column SResultRowEntryInfo is denote by RowCellInfoOffset.
 * Note that in case of top/bottom query, the whole multiple rows of result is treated as only one row of results.
H
Haojun Liao 已提交
1878 1879 1880
 * +------------+-----------------result column 1------------+------------------result column 2-----------+
 * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
 * +------------+--------------------------------------------+--------------------------------------------+
1881 1882
 *           offset[0]                                  offset[1]                                   offset[2]
 */
1883
// TODO refactor: some function move away
H
Haojun Liao 已提交
1884
void setFunctionResultOutput(SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage, SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
1885 1886 1887
  SqlFunctionCtx* pCtx = pInfo->pCtx;
  SSDataBlock*    pDataBlock = pInfo->pRes;
  int32_t*        rowCellInfoOffset = pInfo->rowCellInfoOffset;
H
Haojun Liao 已提交
1888

H
Haojun Liao 已提交
1889
  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
H
Haojun Liao 已提交
1890
  initResultRowInfo(pResultRowInfo, 16);
H
Haojun Liao 已提交
1891

L
Liu Jicong 已提交
1892 1893
  int64_t     tid = 0;
  int64_t     groupId = 0;
1894 1895
  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
                                            pTaskInfo, false, pSup);
H
Haojun Liao 已提交
1896 1897 1898 1899 1900

  for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
    struct SResultRowEntryInfo* pEntry = getResultCell(pRow, i, rowCellInfoOffset);
    cleanupResultRowEntry(pEntry);

L
Liu Jicong 已提交
1901
    pCtx[i].resultInfo = pEntry;
H
Haojun Liao 已提交
1902 1903 1904
    pCtx[i].currentStage = stage;

    // set the timestamp output buffer for top/bottom/diff query
L
Liu Jicong 已提交
1905 1906
    //    int32_t fid = pCtx[i].functionId;
    //    if (fid == FUNCTION_TOP || fid == FUNCTION_BOTTOM || fid == FUNCTION_DIFF || fid == FUNCTION_DERIVATIVE) {
H
Haojun Liao 已提交
1907
    //      if (i > 0) pCtx[i].pTsOutput = pCtx[i-1].pOutput;
L
Liu Jicong 已提交
1908
    //    }
H
Haojun Liao 已提交
1909 1910 1911 1912 1913
  }

  initCtxOutputBuffer(pCtx, pDataBlock->info.numOfCols);
}

L
Liu Jicong 已提交
1914
void updateOutputBuf(SOptrBasicInfo* pBInfo, int32_t* bufCapacity, int32_t numOfInputRows) {
1915 1916
  SSDataBlock* pDataBlock = pBInfo->pRes;

L
Liu Jicong 已提交
1917
  int32_t newSize = pDataBlock->info.rows + numOfInputRows + 5;  // extra output buffer
1918
  if ((*bufCapacity) < newSize) {
L
Liu Jicong 已提交
1919 1920
    for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
      SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1921

wafwerar's avatar
wafwerar 已提交
1922
      char* p = taosMemoryRealloc(pColInfo->pData, newSize * pColInfo->info.bytes);
1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935
      if (p != NULL) {
        pColInfo->pData = p;

        // it starts from the tail of the previously generated results.
        pBInfo->pCtx[i].pOutput = pColInfo->pData;
        (*bufCapacity) = newSize;
      } else {
        // longjmp
      }
    }
  }

  for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
L
Liu Jicong 已提交
1936
    SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1937 1938 1939 1940 1941
    pBInfo->pCtx[i].pOutput = pColInfo->pData + pColInfo->info.bytes * pDataBlock->info.rows;

    // set the correct pointer after the memory buffer reallocated.
    int32_t functionId = pBInfo->pCtx[i].functionId;

L
Liu Jicong 已提交
1942 1943
    if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF ||
        functionId == FUNCTION_DERIVATIVE) {
dengyihao's avatar
dengyihao 已提交
1944
      //      if (i > 0) pBInfo->pCtx[i].pTsOutput = pBInfo->pCtx[i - 1].pOutput;
1945 1946 1947 1948
    }
  }
}

H
Haojun Liao 已提交
1949
void copyTsColoum(SSDataBlock* pRes, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1950 1951
  bool    needCopyTs = false;
  int32_t tsNum = 0;
L
Liu Jicong 已提交
1952
  char*   src = NULL;
1953 1954 1955 1956
  for (int32_t i = 0; i < numOfOutput; i++) {
    int32_t functionId = pCtx[i].functionId;
    if (functionId == FUNCTION_DIFF || functionId == FUNCTION_DERIVATIVE) {
      needCopyTs = true;
L
Liu Jicong 已提交
1957 1958
      if (i > 0 && pCtx[i - 1].functionId == FUNCTION_TS_DUMMY) {
        SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i - 1);  // find ts data
1959 1960
        src = pColRes->pData;
      }
L
Liu Jicong 已提交
1961
    } else if (functionId == FUNCTION_TS_DUMMY) {
1962 1963 1964 1965 1966 1967 1968 1969 1970 1971
      tsNum++;
    }
  }

  if (!needCopyTs) return;
  if (tsNum < 2) return;
  if (src == NULL) return;

  for (int32_t i = 0; i < numOfOutput; i++) {
    int32_t functionId = pCtx[i].functionId;
L
Liu Jicong 已提交
1972
    if (functionId == FUNCTION_TS_DUMMY) {
1973 1974 1975 1976 1977 1978
      SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i);
      memcpy(pColRes->pData, src, pColRes->info.bytes * pRes->info.rows);
    }
  }
}

H
Haojun Liao 已提交
1979
void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
1980 1981
  for (int32_t j = 0; j < size; ++j) {
    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
dengyihao's avatar
dengyihao 已提交
1982 1983
    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
        fmIsScalarFunc(pCtx[j].functionId)) {
1984 1985 1986
      continue;
    }

H
Haojun Liao 已提交
1987
    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
1988 1989 1990
  }
}

L
Liu Jicong 已提交
1991
void setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status) {
1992
  if (status == TASK_NOT_COMPLETED) {
H
Haojun Liao 已提交
1993
    pTaskInfo->status = status;
1994 1995
  } else {
    // QUERY_NOT_COMPLETED is not compatible with any other status, so clear its position first
1996
    CLEAR_QUERY_STATUS(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
1997
    pTaskInfo->status |= status;
1998 1999 2000
  }
}

2001
// todo merged with the build group result.
L
Liu Jicong 已提交
2002 2003
void finalizeMultiTupleQueryResult(SqlFunctionCtx* pCtx, int32_t numOfOutput, SDiskbasedBuf* pBuf,
                                   SResultRowInfo* pResultRowInfo, int32_t* rowCellInfoOffset) {
2004 2005 2006
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
    SResultRowPosition* pPos = &pResultRowInfo->pPosition[i];

L
Liu Jicong 已提交
2007
    SFilePage*  bufPage = getBufPage(pBuf, pPos->pageId);
2008
    SResultRow* pRow = (SResultRow*)((char*)bufPage + pPos->offset);
H
Haojun Liao 已提交
2009 2010

    // TODO ignore the close status anyway.
dengyihao's avatar
dengyihao 已提交
2011 2012 2013
    //    if (!isResultRowClosed(pRow)) {
    //      continue;
    //    }
2014

2015
    for (int32_t j = 0; j < numOfOutput; ++j) {
2016 2017
      pCtx[j].resultInfo = getResultCell(pRow, j, rowCellInfoOffset);

H
Haojun Liao 已提交
2018
      struct SResultRowEntryInfo* pResInfo = pCtx[j].resultInfo;
H
Haojun Liao 已提交
2019
      if (!isRowEntryInitialized(pResInfo)) {
2020 2021 2022
        continue;
      }

H
Haojun Liao 已提交
2023
      if (pCtx[j].fpSet.process) {  // TODO set the dummy function, to avoid the check for null ptr.
dengyihao's avatar
dengyihao 已提交
2024
                                    //        pCtx[j].fpSet.finalize(&pCtx[j]);
H
Haojun Liao 已提交
2025
      }
2026 2027 2028 2029

      if (pRow->numOfRows < pResInfo->numOfRes) {
        pRow->numOfRows = pResInfo->numOfRes;
      }
2030
    }
2031 2032 2033

    releaseBufPage(pBuf, bufPage);
  }
2034 2035
}

2036
STableQueryInfo* createTableQueryInfo(void* buf, STimeWindow win) {
L
Liu Jicong 已提交
2037
  STableQueryInfo* pTableQueryInfo = buf;
2038 2039 2040 2041
  pTableQueryInfo->lastKey = win.skey;
  return pTableQueryInfo;
}

L
Liu Jicong 已提交
2042
void destroyTableQueryInfoImpl(STableQueryInfo* pTableQueryInfo) {
2043 2044 2045 2046
  if (pTableQueryInfo == NULL) {
    return;
  }

L
Liu Jicong 已提交
2047
  //  taosVariantDestroy(&pTableQueryInfo->tag);
dengyihao's avatar
dengyihao 已提交
2048
  //  cleanupResultRowInfo(&pTableQueryInfo->resInfo);
2049 2050
}

2051
void setResultRowInitCtx(SResultRow* pResult, SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t* rowCellInfoOffset) {
2052 2053 2054 2055 2056 2057 2058
  for (int32_t i = 0; i < numOfOutput; ++i) {
    pCtx[i].resultInfo = getResultCell(pResult, i, rowCellInfoOffset);

    struct SResultRowEntryInfo* pResInfo = pCtx[i].resultInfo;
    if (isRowEntryCompleted(pResInfo) && isRowEntryInitialized(pResInfo)) {
      continue;
    }
2059 2060 2061 2062 2063

    if (fmIsWindowPseudoColumnFunc(pCtx[i].functionId)) {
      continue;
    }

2064 2065 2066 2067 2068 2069
    if (!pResInfo->initialized) {
      if (pCtx[i].functionId != -1) {
        pCtx[i].fpSet.init(&pCtx[i], pResInfo);
      } else {
        pResInfo->initialized = true;
      }
2070 2071 2072 2073
    }
  }
}

2074 2075 2076 2077 2078 2079
void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock) {
  if (pFilterNode == NULL) {
    return;
  }

  SFilterInfo* filter = NULL;
H
Haojun Liao 已提交
2080

H
Haojun Liao 已提交
2081
  // todo move to the initialization function
H
Haojun Liao 已提交
2082
  int32_t code = filterInitFromNode((SNode*)pFilterNode, &filter, 0);
2083 2084 2085 2086 2087 2088

  SFilterColumnParam param1 = {.numOfCols = pBlock->info.numOfCols, .pDataBlock = pBlock->pDataBlock};
  code = filterSetDataFromSlotId(filter, &param1);

  int8_t* rowRes = NULL;
  bool    keep = filterExecute(filter, pBlock, &rowRes, NULL, param1.numOfCols);
D
dapan1121 已提交
2089
  filterFreeInfo(filter);
2090

2091
  SSDataBlock* px = createOneDataBlock(pBlock, false);
2092 2093
  blockDataEnsureCapacity(px, pBlock->info.rows);

H
Haojun Liao 已提交
2094
  // todo extract method
2095 2096 2097 2098
  int32_t numOfRow = 0;
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
    SColumnInfoData* pDst = taosArrayGet(px->pDataBlock, i);
    SColumnInfoData* pSrc = taosArrayGet(pBlock->pDataBlock, i);
D
dapan1121 已提交
2099 2100 2101 2102 2103 2104 2105 2106 2107
    if (keep) {
      colDataAssign(pDst, pSrc, pBlock->info.rows);
      numOfRow = pBlock->info.rows;
    } else if (NULL != rowRes) {
      numOfRow = 0;
      for (int32_t j = 0; j < pBlock->info.rows; ++j) {
        if (rowRes[j] == 0) {
          continue;
        }
2108

D
dapan1121 已提交
2109 2110 2111 2112 2113 2114
        if (colDataIsNull_s(pSrc, j)) {
          colDataAppendNULL(pDst, numOfRow);
        } else {
          colDataAppend(pDst, numOfRow, colDataGetData(pSrc, j), false);
        }
        numOfRow += 1;
H
Haojun Liao 已提交
2115
      }
D
dapan1121 已提交
2116 2117
    } else {
      numOfRow = 0;
2118 2119 2120 2121 2122 2123
    }

    *pSrc = *pDst;
  }

  pBlock->info.rows = numOfRow;
2124
  blockDataUpdateTsWindow(pBlock);
2125 2126
}

dengyihao's avatar
dengyihao 已提交
2127 2128
void doSetTableGroupOutputBuf(SAggOperatorInfo* pAggInfo, int32_t numOfOutput, uint64_t groupId,
                              SExecTaskInfo* pTaskInfo) {
2129 2130 2131
  // for simple group by query without interval, all the tables belong to one group result.
  int64_t uid = 0;

2132 2133
  SResultRowInfo* pResultRowInfo = &pAggInfo->binfo.resultRowInfo;
  SqlFunctionCtx* pCtx = pAggInfo->binfo.pCtx;
L
Liu Jicong 已提交
2134
  int32_t*        rowCellInfoOffset = pAggInfo->binfo.rowCellInfoOffset;
2135

2136
  SResultRow* pResultRow = doSetResultOutBufByKey(pAggInfo->aggSup.pResultBuf, pResultRowInfo, (char*)&groupId,
L
Liu Jicong 已提交
2137
                                                  sizeof(groupId), true, groupId, pTaskInfo, false, &pAggInfo->aggSup);
L
Liu Jicong 已提交
2138
  assert(pResultRow != NULL);
2139 2140 2141 2142 2143 2144

  /*
   * not assign result buffer yet, add new result buffer
   * all group belong to one result set, and each group result has different group id so set the id to be one
   */
  if (pResultRow->pageId == -1) {
dengyihao's avatar
dengyihao 已提交
2145 2146
    int32_t ret =
        addNewWindowResultBuf(pResultRow, pAggInfo->aggSup.pResultBuf, groupId, pAggInfo->binfo.pRes->info.rowSize);
2147 2148 2149 2150 2151
    if (ret != TSDB_CODE_SUCCESS) {
      return;
    }
  }

2152
  setResultRowInitCtx(pResultRow, pCtx, numOfOutput, rowCellInfoOffset);
2153 2154
}

H
Haojun Liao 已提交
2155 2156
void setExecutionContext(int32_t numOfOutput, uint64_t groupId, SExecTaskInfo* pTaskInfo, SAggOperatorInfo* pAggInfo) {
  if (pAggInfo->groupId != INT32_MIN && pAggInfo->groupId == groupId) {
2157 2158 2159
    return;
  }

H
Haojun Liao 已提交
2160
  doSetTableGroupOutputBuf(pAggInfo, numOfOutput, groupId, pTaskInfo);
2161 2162

  // record the current active group id
H
Haojun Liao 已提交
2163
  pAggInfo->groupId = groupId;
2164 2165 2166 2167 2168 2169 2170 2171 2172 2173
}

/**
 * For interval query of both super table and table, copy the data in ascending order, since the output results are
 * ordered in SWindowResutl already. While handling the group by query for both table and super table,
 * all group result are completed already.
 *
 * @param pQInfo
 * @param result
 */
2174
int32_t doCopyToSDataBlock(SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf, SGroupResInfo* pGroupResInfo,
2175
                           int32_t orderType, int32_t* rowCellOffset, SqlFunctionCtx* pCtx) {
2176
  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
L
Liu Jicong 已提交
2177
  int32_t numOfResult = pBlock->info.rows;  // there are already exists result rows
2178 2179 2180 2181

  int32_t start = 0;
  int32_t step = -1;

L
Liu Jicong 已提交
2182
  // qDebug("QInfo:0x%"PRIx64" start to copy data from windowResInfo to output buf", GET_TASKID(pRuntimeEnv));
2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193
  assert(orderType == TSDB_ORDER_ASC || orderType == TSDB_ORDER_DESC);

  if (orderType == TSDB_ORDER_ASC) {
    start = pGroupResInfo->index;
    step = 1;
  } else {  // desc order copy all data
    start = numOfRows - pGroupResInfo->index - 1;
    step = -1;
  }

  for (int32_t i = start; (i < numOfRows) && (i >= 0); i += step) {
L
Liu Jicong 已提交
2194 2195
    SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
    SFilePage*  page = getBufPage(pBuf, pPos->pos.pageId);
2196

2197
    SResultRow* pRow = (SResultRow*)((char*)page + pPos->pos.offset);
2198 2199 2200 2201 2202
    if (pRow->numOfRows == 0) {
      pGroupResInfo->index += 1;
      continue;
    }

2203
    // TODO copy multiple rows?
2204
    int32_t numOfRowsToCopy = pRow->numOfRows;
2205
    if (numOfResult + numOfRowsToCopy >= pBlock->info.capacity) {
2206 2207 2208 2209 2210 2211
      break;
    }

    pGroupResInfo->index += 1;

    for (int32_t j = 0; j < pBlock->info.numOfCols; ++j) {
2212 2213
      int32_t slotId = pExprInfo[j].base.resSchema.slotId;

2214 2215
      pCtx[j].resultInfo = getResultCell(pRow, j, rowCellOffset);
      if (pCtx[j].fpSet.process) {
2216
        pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
2217 2218
      } else {
        SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, slotId);
2219

2220 2221 2222
        char* in = GET_ROWCELL_INTERBUF(pCtx[j].resultInfo);
        colDataAppend(pColInfoData, pBlock->info.rows, in, pCtx[j].resultInfo->isNullRes);
      }
2223 2224
    }

2225 2226
    releaseBufPage(pBuf, page);

2227 2228
    pBlock->info.rows += pRow->numOfRows;
    if (pBlock->info.rows >= pBlock->info.capacity) {  // output buffer is full
2229 2230 2231 2232
      break;
    }
  }

L
Liu Jicong 已提交
2233
  // qDebug("QInfo:0x%"PRIx64" copy data to query buf completed", GET_TASKID(pRuntimeEnv));
2234
  blockDataUpdateTsWindow(pBlock);
2235 2236 2237
  return 0;
}

2238
void doBuildResultDatablock(SOptrBasicInfo *pbInfo, SGroupResInfo* pGroupResInfo, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf) {
2239 2240
  assert(pGroupResInfo->currentGroup <= pGroupResInfo->totalGroup);

2241 2242 2243 2244
  int32_t* rowCellOffset = pbInfo->rowCellInfoOffset;
  SSDataBlock* pBlock    = pbInfo->pRes;
  SqlFunctionCtx* pCtx   = pbInfo->pCtx;

2245
  blockDataCleanup(pBlock);
2246 2247 2248 2249
  if (!hasRemainDataInCurrentGroup(pGroupResInfo)) {
    return;
  }

2250
  int32_t orderType = TSDB_ORDER_ASC;
2251
  doCopyToSDataBlock(pBlock, pExprInfo, pBuf, pGroupResInfo, orderType, rowCellOffset, pCtx);
2252

H
Haojun Liao 已提交
2253 2254
  // add condition (pBlock->info.rows >= 1) just to runtime happy
  blockDataUpdateTsWindow(pBlock);
2255 2256
}

L
Liu Jicong 已提交
2257 2258
static void updateNumOfRowsInResultRows(SqlFunctionCtx* pCtx, int32_t numOfOutput, SResultRowInfo* pResultRowInfo,
                                        int32_t* rowCellInfoOffset) {
2259
  // update the number of result for each, only update the number of rows for the corresponding window result.
L
Liu Jicong 已提交
2260 2261 2262
  //  if (QUERY_IS_INTERVAL_QUERY(pQueryAttr)) {
  //    return;
  //  }
H
Haojun Liao 已提交
2263
#if 0
2264
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
L
Liu Jicong 已提交
2265
    SResultRow* pResult = pResultRowInfo->pResult[i];
2266 2267 2268 2269 2270 2271 2272

    for (int32_t j = 0; j < numOfOutput; ++j) {
      int32_t functionId = pCtx[j].functionId;
      if (functionId == FUNCTION_TS || functionId == FUNCTION_TAG || functionId == FUNCTION_TAGPRJ) {
        continue;
      }

2273 2274
      SResultRowEntryInfo* pCell = getResultCell(pResult, j, rowCellInfoOffset);
      pResult->numOfRows = (uint16_t)(TMAX(pResult->numOfRows, pCell->numOfRes));
2275 2276
    }
  }
H
Haojun Liao 已提交
2277
#endif
2278 2279
}

L
Liu Jicong 已提交
2280
static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
2281 2282
  int32_t colSize = pColRes->info.bytes * numOfRows;
  return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
L
Liu Jicong 已提交
2283
                                                      colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
2284 2285
}

2286
int32_t doFillTimeIntervalGapsInResults(struct SFillInfo* pFillInfo, SSDataBlock* pBlock, int32_t capacity) {
L
Liu Jicong 已提交
2287 2288 2289 2290
  //  for(int32_t i = 0; i < pFillInfo->numOfCols; ++i) {
  //    SColumnInfoData* pColInfoData = taosArrayGet(pOutput->pDataBlock, i);
  //    p[i] = pColInfoData->pData + (pColInfoData->info.bytes * pOutput->info.rows);
  //  }
2291

2292 2293
  int32_t numOfRows = (int32_t)taosFillResultDataBlock(pFillInfo, pBlock, capacity - pBlock->info.rows);
  pBlock->info.rows += numOfRows;
2294

2295
  return pBlock->info.rows;
2296 2297
}

2298
void publishOperatorProfEvent(SOperatorInfo* pOperator, EQueryProfEventType eventType) {
2299 2300
  SQueryProfEvent event = {0};

L
Liu Jicong 已提交
2301 2302
  event.eventType = eventType;
  event.eventTime = taosGetTimestampUs();
2303
  event.operatorType = pOperator->operatorType;
dengyihao's avatar
dengyihao 已提交
2304 2305 2306
  //    if (pQInfo->summary.queryProfEvents) {
  //      taosArrayPush(pQInfo->summary.queryProfEvents, &event);
  //    }
2307 2308
}

L
Liu Jicong 已提交
2309
void publishQueryAbortEvent(SExecTaskInfo* pTaskInfo, int32_t code) {
2310 2311 2312 2313 2314
  SQueryProfEvent event;
  event.eventType = QUERY_PROF_QUERY_ABORT;
  event.eventTime = taosGetTimestampUs();
  event.abortCode = code;

2315 2316
  if (pTaskInfo->cost.queryProfEvents) {
    taosArrayPush(pTaskInfo->cost.queryProfEvents, &event);
2317 2318 2319
  }
}

L
Liu Jicong 已提交
2320
typedef struct {
2321 2322 2323 2324 2325 2326 2327
  uint8_t operatorType;
  int64_t beginTime;
  int64_t endTime;
  int64_t selfTime;
  int64_t descendantsTime;
} SOperatorStackItem;

L
Liu Jicong 已提交
2328 2329
static void doOperatorExecProfOnce(SOperatorStackItem* item, SQueryProfEvent* event, SArray* opStack,
                                   SHashObj* profResults) {
2330 2331 2332 2333 2334 2335 2336 2337
  item->endTime = event->eventTime;
  item->selfTime = (item->endTime - item->beginTime) - (item->descendantsTime);

  for (int32_t j = 0; j < taosArrayGetSize(opStack); ++j) {
    SOperatorStackItem* ancestor = taosArrayGet(opStack, j);
    ancestor->descendantsTime += item->selfTime;
  }

L
Liu Jicong 已提交
2338
  uint8_t              operatorType = item->operatorType;
2339 2340 2341 2342 2343 2344 2345 2346 2347
  SOperatorProfResult* result = taosHashGet(profResults, &operatorType, sizeof(operatorType));
  if (result != NULL) {
    result->sumRunTimes++;
    result->sumSelfTime += item->selfTime;
  } else {
    SOperatorProfResult opResult;
    opResult.operatorType = operatorType;
    opResult.sumSelfTime = item->selfTime;
    opResult.sumRunTimes = 1;
L
Liu Jicong 已提交
2348
    taosHashPut(profResults, &(operatorType), sizeof(operatorType), &opResult, sizeof(opResult));
2349 2350 2351
  }
}

2352
void calculateOperatorProfResults(void) {
dengyihao's avatar
dengyihao 已提交
2353 2354 2355 2356 2357 2358 2359 2360 2361
  //  if (pQInfo->summary.queryProfEvents == NULL) {
  //    // qDebug("QInfo:0x%"PRIx64" query prof events array is null", pQInfo->qId);
  //    return;
  //  }
  //
  //  if (pQInfo->summary.operatorProfResults == NULL) {
  //    // qDebug("QInfo:0x%"PRIx64" operator prof results hash is null", pQInfo->qId);
  //    return;
  //  }
2362 2363 2364 2365 2366

  SArray* opStack = taosArrayInit(32, sizeof(SOperatorStackItem));
  if (opStack == NULL) {
    return;
  }
2367
#if 0
L
Liu Jicong 已提交
2368
  size_t    size = taosArrayGetSize(pQInfo->summary.queryProfEvents);
2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389
  SHashObj* profResults = pQInfo->summary.operatorProfResults;

  for (int i = 0; i < size; ++i) {
    SQueryProfEvent* event = taosArrayGet(pQInfo->summary.queryProfEvents, i);
    if (event->eventType == QUERY_PROF_BEFORE_OPERATOR_EXEC) {
      SOperatorStackItem opItem;
      opItem.operatorType = event->operatorType;
      opItem.beginTime = event->eventTime;
      opItem.descendantsTime = 0;
      taosArrayPush(opStack, &opItem);
    } else if (event->eventType == QUERY_PROF_AFTER_OPERATOR_EXEC) {
      SOperatorStackItem* item = taosArrayPop(opStack);
      assert(item->operatorType == event->operatorType);
      doOperatorExecProfOnce(item, event, opStack, profResults);
    } else if (event->eventType == QUERY_PROF_QUERY_ABORT) {
      SOperatorStackItem* item;
      while ((item = taosArrayPop(opStack)) != NULL) {
        doOperatorExecProfOnce(item, event, opStack, profResults);
      }
    }
  }
2390
#endif
2391 2392 2393
  taosArrayDestroy(opStack);
}

L
Liu Jicong 已提交
2394 2395
void queryCostStatis(SExecTaskInfo* pTaskInfo) {
  STaskCostInfo* pSummary = &pTaskInfo->cost;
2396

L
Liu Jicong 已提交
2397 2398 2399
  //  uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
  //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
  //  pSummary->hashSize = hashSize;
2400 2401 2402 2403

  // add the merge time
  pSummary->elapsedTime += pSummary->firstStageMergeTime;

L
Liu Jicong 已提交
2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417
  //  SResultRowPool* p = pTaskInfo->pool;
  //  if (p != NULL) {
  //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
  //    pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
  //  } else {
  //    pSummary->winInfoSize = 0;
  //    pSummary->numOfTimeWindows = 0;
  //  }
  //
  //  calculateOperatorProfResults(pQInfo);

  qDebug("%s :cost summary: elapsed time:%" PRId64 " us, first merge:%" PRId64
         " us, total blocks:%d, "
         "load block statis:%d, load data block:%d, total rows:%" PRId64 ", check rows:%" PRId64,
2418 2419
         GET_TASKID(pTaskInfo), pSummary->elapsedTime, pSummary->firstStageMergeTime, pSummary->pRecoder->totalBlocks,
         pSummary->pRecoder->loadBlockStatis, pSummary->pRecoder->loadBlocks, pSummary->pRecoder->totalRows, pSummary->pRecoder->totalCheckedRows);
L
Liu Jicong 已提交
2420 2421 2422 2423
  //
  // qDebug("QInfo:0x%"PRIx64" :cost summary: winResPool size:%.2f Kb, numOfWin:%"PRId64", tableInfoSize:%.2f Kb,
  // hashTable:%.2f Kb", pQInfo->qId, pSummary->winInfoSize/1024.0,
  //      pSummary->numOfTimeWindows, pSummary->tableInfoSize/1024.0, pSummary->hashSize/1024.0);
2424 2425 2426 2427

  if (pSummary->operatorProfResults) {
    SOperatorProfResult* opRes = taosHashIterate(pSummary->operatorProfResults, NULL);
    while (opRes != NULL) {
L
Liu Jicong 已提交
2428 2429
      // qDebug("QInfo:0x%" PRIx64 " :cost summary: operator : %d, exec times: %" PRId64 ", self time: %" PRId64,
      //             pQInfo->qId, opRes->operatorType, opRes->sumRunTimes, opRes->sumSelfTime);
2430 2431 2432 2433 2434
      opRes = taosHashIterate(pSummary->operatorProfResults, opRes);
    }
  }
}

L
Liu Jicong 已提交
2435 2436 2437
// static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
2438
//
L
Liu Jicong 已提交
2439
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2440
//
L
Liu Jicong 已提交
2441 2442 2443 2444
//   if (pQueryAttr->limit.offset == pBlockInfo->rows) {  // current block will ignore completed
//     pTableQueryInfo->lastKey = QUERY_IS_ASC_QUERY(pQueryAttr) ? pBlockInfo->window.ekey + step :
//     pBlockInfo->window.skey + step; pQueryAttr->limit.offset = 0; return;
//   }
2445
//
L
Liu Jicong 已提交
2446 2447 2448 2449 2450
//   if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//     pQueryAttr->pos = (int32_t)pQueryAttr->limit.offset;
//   } else {
//     pQueryAttr->pos = pBlockInfo->rows - (int32_t)pQueryAttr->limit.offset - 1;
//   }
2451
//
L
Liu Jicong 已提交
2452
//   assert(pQueryAttr->pos >= 0 && pQueryAttr->pos <= pBlockInfo->rows - 1);
2453
//
L
Liu Jicong 已提交
2454 2455
//   SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//   SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2456
//
L
Liu Jicong 已提交
2457 2458
//   // update the pQueryAttr->limit.offset value, and pQueryAttr->pos value
//   TSKEY *keys = (TSKEY *) pColInfoData->pData;
2459
//
L
Liu Jicong 已提交
2460 2461 2462
//   // update the offset value
//   pTableQueryInfo->lastKey = keys[pQueryAttr->pos];
//   pQueryAttr->limit.offset = 0;
2463
//
L
Liu Jicong 已提交
2464
//   int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
2465
//
L
Liu Jicong 已提交
2466 2467 2468 2469
//   //qDebug("QInfo:0x%"PRIx64" check data block, brange:%" PRId64 "-%" PRId64 ", numBlocksOfStep:%d, numOfRes:%d,
//   lastKey:%"PRId64, GET_TASKID(pRuntimeEnv),
//          pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows, numOfRes, pQuery->current->lastKey);
// }
2470

L
Liu Jicong 已提交
2471 2472
// void skipBlocks(STaskRuntimeEnv *pRuntimeEnv) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
2473
//
L
Liu Jicong 已提交
2474 2475 2476
//   if (pQueryAttr->limit.offset <= 0 || pQueryAttr->numOfFilterCols > 0) {
//     return;
//   }
2477
//
L
Liu Jicong 已提交
2478 2479
//   pQueryAttr->pos = 0;
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2480
//
L
Liu Jicong 已提交
2481 2482
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
//   TsdbQueryHandleT pTsdbReadHandle = pRuntimeEnv->pTsdbReadHandle;
2483
//
L
Liu Jicong 已提交
2484 2485 2486 2487 2488
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pTsdbReadHandle)) {
//     if (isTaskKilled(pRuntimeEnv->qinfo)) {
//       longjmp(pRuntimeEnv->env, TSDB_CODE_TSC_QUERY_CANCELLED);
//     }
2489
//
L
Liu Jicong 已提交
2490
//     tsdbRetrieveDataBlockInfo(pTsdbReadHandle, &blockInfo);
2491
//
L
Liu Jicong 已提交
2492 2493 2494 2495
//     if (pQueryAttr->limit.offset > blockInfo.rows) {
//       pQueryAttr->limit.offset -= blockInfo.rows;
//       pTableQueryInfo->lastKey = (QUERY_IS_ASC_QUERY(pQueryAttr)) ? blockInfo.window.ekey : blockInfo.window.skey;
//       pTableQueryInfo->lastKey += step;
2496
//
L
Liu Jicong 已提交
2497 2498 2499 2500 2501 2502 2503
//       //qDebug("QInfo:0x%"PRIx64" skip rows:%d, offset:%" PRId64, GET_TASKID(pRuntimeEnv), blockInfo.rows,
//              pQuery->limit.offset);
//     } else {  // find the appropriated start position in current block
//       updateOffsetVal(pRuntimeEnv, &blockInfo);
//       break;
//     }
//   }
2504
//
L
Liu Jicong 已提交
2505 2506 2507 2508 2509 2510 2511 2512 2513
//   if (terrno != TSDB_CODE_SUCCESS) {
//     longjmp(pRuntimeEnv->env, terrno);
//   }
// }

// static TSKEY doSkipIntervalProcess(STaskRuntimeEnv* pRuntimeEnv, STimeWindow* win, SDataBlockInfo* pBlockInfo,
// STableQueryInfo* pTableQueryInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
2514
//
L
Liu Jicong 已提交
2515 2516 2517
//   assert(pQueryAttr->limit.offset == 0);
//   STimeWindow tw = *win;
//   getNextTimeWindow(pQueryAttr, &tw);
2518
//
L
Liu Jicong 已提交
2519 2520
//   if ((tw.skey <= pBlockInfo->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) ||
//       (tw.ekey >= pBlockInfo->window.skey && !QUERY_IS_ASC_QUERY(pQueryAttr))) {
2521
//
L
Liu Jicong 已提交
2522 2523 2524 2525
//     // load the data block and check data remaining in current data block
//     // TODO optimize performance
//     SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//     SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2526
//
L
Liu Jicong 已提交
2527 2528 2529 2530
//     tw = *win;
//     int32_t startPos =
//         getNextQualifiedWindow(pQueryAttr, &tw, pBlockInfo, pColInfoData->pData, binarySearchForKey, -1);
//     assert(startPos >= 0);
2531
//
L
Liu Jicong 已提交
2532 2533
//     // set the abort info
//     pQueryAttr->pos = startPos;
2534
//
L
Liu Jicong 已提交
2535 2536 2537 2538
//     // reset the query start timestamp
//     pTableQueryInfo->win.skey = ((TSKEY *)pColInfoData->pData)[startPos];
//     pQueryAttr->window.skey = pTableQueryInfo->win.skey;
//     TSKEY key = pTableQueryInfo->win.skey;
2539
//
L
Liu Jicong 已提交
2540 2541
//     pWindowResInfo->prevSKey = tw.skey;
//     int32_t index = pRuntimeEnv->resultRowInfo.curIndex;
2542
//
L
Liu Jicong 已提交
2543 2544
//     int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
//     pRuntimeEnv->resultRowInfo.curIndex = index;  // restore the window index
2545
//
L
Liu Jicong 已提交
2546 2547 2548 2549
//     //qDebug("QInfo:0x%"PRIx64" check data block, brange:%" PRId64 "-%" PRId64 ", numOfRows:%d, numOfRes:%d,
//     lastKey:%" PRId64,
//            GET_TASKID(pRuntimeEnv), pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows, numOfRes,
//            pQueryAttr->current->lastKey);
2550
//
L
Liu Jicong 已提交
2551 2552 2553 2554 2555
//     return key;
//   } else {  // do nothing
//     pQueryAttr->window.skey      = tw.skey;
//     pWindowResInfo->prevSKey = tw.skey;
//     pTableQueryInfo->lastKey = tw.skey;
2556
//
L
Liu Jicong 已提交
2557 2558
//     return tw.skey;
//   }
2559
//
L
Liu Jicong 已提交
2560 2561 2562 2563 2564 2565 2566 2567 2568 2569
//   return true;
// }

// static bool skipTimeInterval(STaskRuntimeEnv *pRuntimeEnv, TSKEY* start) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//     assert(*start <= pRuntimeEnv->current->lastKey);
//   } else {
//     assert(*start >= pRuntimeEnv->current->lastKey);
//   }
2570
//
L
Liu Jicong 已提交
2571 2572 2573 2574 2575
//   // if queried with value filter, do NOT forward query start position
//   if (pQueryAttr->limit.offset <= 0 || pQueryAttr->numOfFilterCols > 0 || pRuntimeEnv->pTsBuf != NULL ||
//   pRuntimeEnv->pFillInfo != NULL) {
//     return true;
//   }
2576
//
L
Liu Jicong 已提交
2577 2578 2579 2580 2581 2582 2583
//   /*
//    * 1. for interval without interpolation query we forward pQueryAttr->interval.interval at a time for
//    *    pQueryAttr->limit.offset times. Since hole exists, pQueryAttr->interval.interval*pQueryAttr->limit.offset
//    value is
//    *    not valid. otherwise, we only forward pQueryAttr->limit.offset number of points
//    */
//   assert(pRuntimeEnv->resultRowInfo.prevSKey == TSKEY_INITIAL_VAL);
2584
//
L
Liu Jicong 已提交
2585 2586
//   STimeWindow w = TSWINDOW_INITIALIZER;
//   bool ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
2587
//
L
Liu Jicong 已提交
2588 2589
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
//   STableQueryInfo *pTableQueryInfo = pRuntimeEnv->current;
2590
//
L
Liu Jicong 已提交
2591 2592 2593
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pRuntimeEnv->pTsdbReadHandle)) {
//     tsdbRetrieveDataBlockInfo(pRuntimeEnv->pTsdbReadHandle, &blockInfo);
2594
//
L
Liu Jicong 已提交
2595 2596 2597 2598 2599 2600 2601 2602 2603
//     if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//       if (pWindowResInfo->prevSKey == TSKEY_INITIAL_VAL) {
//         getAlignQueryTimeWindow(pQueryAttr, blockInfo.window.skey, blockInfo.window.skey, pQueryAttr->window.ekey,
//         &w); pWindowResInfo->prevSKey = w.skey;
//       }
//     } else {
//       getAlignQueryTimeWindow(pQueryAttr, blockInfo.window.ekey, pQueryAttr->window.ekey, blockInfo.window.ekey, &w);
//       pWindowResInfo->prevSKey = w.skey;
//     }
2604
//
L
Liu Jicong 已提交
2605 2606
//     // the first time window
//     STimeWindow win = getActiveTimeWindow(pWindowResInfo, pWindowResInfo->prevSKey, pQueryAttr);
2607
//
L
Liu Jicong 已提交
2608 2609
//     while (pQueryAttr->limit.offset > 0) {
//       STimeWindow tw = win;
2610
//
L
Liu Jicong 已提交
2611 2612 2613
//       if ((win.ekey <= blockInfo.window.ekey && ascQuery) || (win.ekey >= blockInfo.window.skey && !ascQuery)) {
//         pQueryAttr->limit.offset -= 1;
//         pWindowResInfo->prevSKey = win.skey;
2614
//
L
Liu Jicong 已提交
2615 2616 2617 2618 2619 2620
//         // current time window is aligned with blockInfo.window.ekey
//         // restart it from next data block by set prevSKey to be TSKEY_INITIAL_VAL;
//         if ((win.ekey == blockInfo.window.ekey && ascQuery) || (win.ekey == blockInfo.window.skey && !ascQuery)) {
//           pWindowResInfo->prevSKey = TSKEY_INITIAL_VAL;
//         }
//       }
2621
//
L
Liu Jicong 已提交
2622 2623 2624 2625
//       if (pQueryAttr->limit.offset == 0) {
//         *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
//         return true;
//       }
2626
//
L
Liu Jicong 已提交
2627 2628
//       // current window does not ended in current data block, try next data block
//       getNextTimeWindow(pQueryAttr, &tw);
2629
//
L
Liu Jicong 已提交
2630 2631 2632 2633 2634 2635 2636 2637 2638
//       /*
//        * If the next time window still starts from current data block,
//        * load the primary timestamp column first, and then find the start position for the next queried time window.
//        * Note that only the primary timestamp column is required.
//        * TODO: Optimize for this cases. All data blocks are not needed to be loaded, only if the first actually
//        required
//        * time window resides in current data block.
//        */
//       if ((tw.skey <= blockInfo.window.ekey && ascQuery) || (tw.ekey >= blockInfo.window.skey && !ascQuery)) {
2639
//
L
Liu Jicong 已提交
2640 2641
//         SArray *pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//         SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2642
//
L
Liu Jicong 已提交
2643 2644 2645
//         if ((win.ekey > blockInfo.window.ekey && ascQuery) || (win.ekey < blockInfo.window.skey && !ascQuery)) {
//           pQueryAttr->limit.offset -= 1;
//         }
2646
//
L
Liu Jicong 已提交
2647 2648 2649 2650 2651 2652 2653 2654
//         if (pQueryAttr->limit.offset == 0) {
//           *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
//           return true;
//         } else {
//           tw = win;
//           int32_t startPos =
//               getNextQualifiedWindow(pQueryAttr, &tw, &blockInfo, pColInfoData->pData, binarySearchForKey, -1);
//           assert(startPos >= 0);
2655
//
L
Liu Jicong 已提交
2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666
//           // set the abort info
//           pQueryAttr->pos = startPos;
//           pTableQueryInfo->lastKey = ((TSKEY *)pColInfoData->pData)[startPos];
//           pWindowResInfo->prevSKey = tw.skey;
//           win = tw;
//         }
//       } else {
//         break;  // offset is not 0, and next time window begins or ends in the next block.
//       }
//     }
//   }
2667
//
L
Liu Jicong 已提交
2668 2669 2670 2671
//   // check for error
//   if (terrno != TSDB_CODE_SUCCESS) {
//     longjmp(pRuntimeEnv->env, terrno);
//   }
2672
//
L
Liu Jicong 已提交
2673 2674
//   return true;
// }
2675

2676
int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t num) {
H
Haojun Liao 已提交
2677
  if (p->pDownstream == NULL) {
H
Haojun Liao 已提交
2678
    assert(p->numOfDownstream == 0);
2679 2680
  }

wafwerar's avatar
wafwerar 已提交
2681
  p->pDownstream = taosMemoryCalloc(1, num * POINTER_BYTES);
2682 2683 2684 2685 2686 2687 2688
  if (p->pDownstream == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

  memcpy(p->pDownstream, pDownstream, num * POINTER_BYTES);
  p->numOfDownstream = num;
  return TSDB_CODE_SUCCESS;
2689 2690 2691 2692
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo);

2693
static void doTableQueryInfoTimeWindowCheck(SExecTaskInfo* pTaskInfo, STableQueryInfo* pTableQueryInfo, int32_t order) {
H
Haojun Liao 已提交
2694 2695
#if 0
    if (order == TSDB_ORDER_ASC) {
2696 2697
    assert(
        (pTableQueryInfo->win.skey <= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2698 2699
        (pTableQueryInfo->lastKey >= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey >= pTaskInfo->window.skey && pTableQueryInfo->win.ekey <= pTaskInfo->window.ekey));
2700 2701 2702
  } else {
    assert(
        (pTableQueryInfo->win.skey >= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2703 2704
        (pTableQueryInfo->lastKey <= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey <= pTaskInfo->window.skey && pTableQueryInfo->win.ekey >= pTaskInfo->window.ekey));
2705
  }
H
Haojun Liao 已提交
2706
#endif
2707 2708
}

L
Liu Jicong 已提交
2709 2710 2711 2712
// static void updateTableIdInfo(STableQueryInfo* pTableQueryInfo, SSDataBlock* pBlock, SHashObj* pTableIdInfo, int32_t
// order) {
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(order);
//   pTableQueryInfo->lastKey = ((order == TSDB_ORDER_ASC)? pBlock->info.window.ekey:pBlock->info.window.skey) + step;
2713
//
L
Liu Jicong 已提交
2714 2715 2716
//   if (pTableQueryInfo->pTable == NULL) {
//     return;
//   }
2717
//
L
Liu Jicong 已提交
2718 2719 2720 2721 2722 2723 2724 2725 2726
//   STableIdInfo tidInfo = createTableIdInfo(pTableQueryInfo);
//   STableIdInfo *idinfo = taosHashGet(pTableIdInfo, &tidInfo.tid, sizeof(tidInfo.tid));
//   if (idinfo != NULL) {
//     assert(idinfo->tid == tidInfo.tid && idinfo->uid == tidInfo.uid);
//     idinfo->key = tidInfo.key;
//   } else {
//     taosHashPut(pTableIdInfo, &tidInfo.tid, sizeof(tidInfo.tid), &tidInfo, sizeof(STableIdInfo));
//   }
// }
2727

2728
int32_t loadRemoteDataCallback(void* param, const SDataBuf* pMsg, int32_t code) {
L
Liu Jicong 已提交
2729
  SSourceDataInfo* pSourceDataInfo = (SSourceDataInfo*)param;
H
Haojun Liao 已提交
2730 2731
  if (code == TSDB_CODE_SUCCESS) {
    pSourceDataInfo->pRsp = pMsg->pData;
2732

H
Haojun Liao 已提交
2733 2734
    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
    pRsp->numOfRows = htonl(pRsp->numOfRows);
dengyihao's avatar
dengyihao 已提交
2735
    pRsp->compLen = htonl(pRsp->compLen);
2736
    pRsp->numOfCols = htonl(pRsp->numOfCols);
dengyihao's avatar
dengyihao 已提交
2737
    pRsp->useconds = htobe64(pRsp->useconds);
H
Haojun Liao 已提交
2738 2739 2740
  } else {
    pSourceDataInfo->code = code;
  }
H
Haojun Liao 已提交
2741

H
Haojun Liao 已提交
2742
  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
2743
  tsem_post(&pSourceDataInfo->pEx->ready);
wmmhello's avatar
wmmhello 已提交
2744
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
2745 2746 2747 2748
}

static void destroySendMsgInfo(SMsgSendInfo* pMsgBody) {
  assert(pMsgBody != NULL);
wafwerar's avatar
wafwerar 已提交
2749 2750
  taosMemoryFreeClear(pMsgBody->msgInfo.pData);
  taosMemoryFreeClear(pMsgBody);
H
Haojun Liao 已提交
2751 2752
}

S
Shengliang Guan 已提交
2753
void qProcessFetchRsp(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
L
Liu Jicong 已提交
2754
  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->ahandle;
H
Haojun Liao 已提交
2755 2756 2757 2758 2759
  assert(pMsg->ahandle != NULL);

  SDataBuf buf = {.len = pMsg->contLen, .pData = NULL};

  if (pMsg->contLen > 0) {
wafwerar's avatar
wafwerar 已提交
2760
    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
H
Haojun Liao 已提交
2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771
    if (buf.pData == NULL) {
      terrno = TSDB_CODE_OUT_OF_MEMORY;
      pMsg->code = TSDB_CODE_OUT_OF_MEMORY;
    } else {
      memcpy(buf.pData, pMsg->pCont, pMsg->contLen);
    }
  }

  pSendInfo->fp(pSendInfo->param, &buf, pMsg->code);
  rpcFreeCont(pMsg->pCont);
  destroySendMsgInfo(pSendInfo);
2772 2773
}

L
Liu Jicong 已提交
2774
static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
2775
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
2776

wafwerar's avatar
wafwerar 已提交
2777
  SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
2778 2779 2780 2781
  if (NULL == pMsg) {
    pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return pTaskInfo->code;
  }
2782

L
Liu Jicong 已提交
2783 2784
  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
2785

L
Liu Jicong 已提交
2786 2787
  qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", %d/%" PRIzu, GET_TASKID(pTaskInfo),
         pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId, sourceIndex, totalSources);
2788 2789 2790 2791 2792 2793 2794

  pMsg->header.vgId = htonl(pSource->addr.nodeId);
  pMsg->sId = htobe64(pSource->schedId);
  pMsg->taskId = htobe64(pSource->taskId);
  pMsg->queryId = htobe64(pTaskInfo->id.queryId);

  // send the fetch remote task result reques
wafwerar's avatar
wafwerar 已提交
2795
  SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
2796
  if (NULL == pMsgSendInfo) {
wafwerar's avatar
wafwerar 已提交
2797
    taosMemoryFreeClear(pMsg);
2798 2799 2800
    qError("%s prepare message %d failed", GET_TASKID(pTaskInfo), (int32_t)sizeof(SMsgSendInfo));
    pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return pTaskInfo->code;
H
Haojun Liao 已提交
2801 2802
  }

2803 2804 2805 2806 2807
  pMsgSendInfo->param = pDataInfo;
  pMsgSendInfo->msgInfo.pData = pMsg;
  pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
  pMsgSendInfo->msgType = TDMT_VND_FETCH;
  pMsgSendInfo->fp = loadRemoteDataCallback;
2808

2809
  int64_t transporterId = 0;
L
Liu Jicong 已提交
2810
  int32_t code = asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
2811 2812 2813
  return TSDB_CODE_SUCCESS;
}

2814
// NOTE: sources columns are more than the destination SSDatablock columns.
2815
void relocateColumnData(SSDataBlock* pBlock, const SArray* pColMatchInfo, SArray* pCols) {
2816 2817 2818 2819
  size_t numOfSrcCols = taosArrayGetSize(pCols);
  ASSERT(numOfSrcCols >= pBlock->info.numOfCols);

  int32_t i = 0, j = 0;
L
Liu Jicong 已提交
2820
  while (i < numOfSrcCols && j < taosArrayGetSize(pColMatchInfo)) {
2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839
    SColumnInfoData* p = taosArrayGet(pCols, i);
    SColMatchInfo*   pmInfo = taosArrayGet(pColMatchInfo, j);
    if (!pmInfo->output) {
      j++;
      continue;
    }

    if (p->info.colId == pmInfo->colId) {
      taosArraySet(pBlock->pDataBlock, pmInfo->targetSlotId, p);
      i++;
      j++;
    } else if (p->info.colId < pmInfo->colId) {
      i++;
    } else {
      ASSERT(0);
    }
  }
}

2840
int32_t setSDataBlockFromFetchRsp(SSDataBlock* pRes, SLoadRemoteDataInfo* pLoadInfo, int32_t numOfRows, char* pData,
dengyihao's avatar
dengyihao 已提交
2841 2842
                                  int32_t compLen, int32_t numOfOutput, int64_t startTs, uint64_t* total,
                                  SArray* pColList) {
H
Haojun Liao 已提交
2843
  blockDataEnsureCapacity(pRes, numOfRows);
2844

H
Haojun Liao 已提交
2845
  if (pColList == NULL) {  // data from other sources
L
Liu Jicong 已提交
2846
    int32_t dataLen = *(int32_t*)pData;
2847 2848
    pData += sizeof(int32_t);

L
Liu Jicong 已提交
2849
    pRes->info.groupId = *(uint64_t*)pData;
2850 2851
    pData += sizeof(uint64_t);

H
Haojun Liao 已提交
2852 2853
    int32_t* colLen = (int32_t*)pData;

2854
    char* pStart = pData + sizeof(int32_t) * numOfOutput;
2855
    for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
2856
      colLen[i] = htonl(colLen[i]);
2857
      ASSERT(colLen[i] >= 0);
H
Haojun Liao 已提交
2858

2859
      SColumnInfoData* pColInfoData = taosArrayGet(pRes->pDataBlock, i);
H
Haojun Liao 已提交
2860 2861 2862
      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
        pColInfoData->varmeta.length = colLen[i];
        pColInfoData->varmeta.allocLen = colLen[i];
2863

L
Liu Jicong 已提交
2864 2865
        memcpy(pColInfoData->varmeta.offset, pStart, sizeof(int32_t) * numOfRows);
        pStart += sizeof(int32_t) * numOfRows;
H
Haojun Liao 已提交
2866

2867 2868 2869
        if (colLen[i] > 0) {
          pColInfoData->pData = taosMemoryMalloc(colLen[i]);
        }
H
Haojun Liao 已提交
2870 2871 2872
      } else {
        memcpy(pColInfoData->nullbitmap, pStart, BitmapLen(numOfRows));
        pStart += BitmapLen(numOfRows);
2873
      }
H
Haojun Liao 已提交
2874

2875 2876 2877 2878
      if (colLen[i] > 0) {
        memcpy(pColInfoData->pData, pStart, colLen[i]);
      }

L
Liu Jicong 已提交
2879 2880
      // TODO setting this flag to true temporarily so aggregate function on stable will
      // examine NULL value for non-primary key column
2881
      pColInfoData->hasNull = true;
H
Haojun Liao 已提交
2882
      pStart += colLen[i];
2883
    }
H
Haojun Liao 已提交
2884
  } else {  // extract data according to pColList
2885
    ASSERT(numOfOutput == taosArrayGetSize(pColList));
2886 2887 2888 2889 2890
    char* pStart = pData;

    int32_t numOfCols = htonl(*(int32_t*)pStart);
    pStart += sizeof(int32_t);

2891
    // todo refactor:extract method
2892
    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
dengyihao's avatar
dengyihao 已提交
2893
    for (int32_t i = 0; i < numOfCols; ++i) {
2894 2895 2896 2897 2898 2899 2900 2901
      SSysTableSchema* p = (SSysTableSchema*)pStart;

      p->colId = htons(p->colId);
      p->bytes = htonl(p->bytes);
      pStart += sizeof(SSysTableSchema);
    }

    SSDataBlock block = {.pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData)), .info.numOfCols = numOfCols};
dengyihao's avatar
dengyihao 已提交
2902
    for (int32_t i = 0; i < numOfCols; ++i) {
2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915
      SColumnInfoData idata = {0};
      idata.info.type = pSchema[i].type;
      idata.info.bytes = pSchema[i].bytes;
      idata.info.colId = pSchema[i].colId;

      taosArrayPush(block.pDataBlock, &idata);
      if (IS_VAR_DATA_TYPE(idata.info.type)) {
        block.info.hasVarCol = true;
      }
    }

    blockDataEnsureCapacity(&block, numOfRows);

L
Liu Jicong 已提交
2916 2917
    int32_t  dataLen = *(int32_t*)pStart;
    uint64_t groupId = *(uint64_t*)(pStart + sizeof(int32_t));
2918 2919
    pStart += sizeof(int32_t) + sizeof(uint64_t);

L
Liu Jicong 已提交
2920
    int32_t* colLen = (int32_t*)(pStart);
2921 2922 2923 2924
    pStart += sizeof(int32_t) * numOfCols;

    for (int32_t i = 0; i < numOfCols; ++i) {
      colLen[i] = htonl(colLen[i]);
2925
      ASSERT(colLen[i] >= 0);
2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943

      SColumnInfoData* pColInfoData = taosArrayGet(block.pDataBlock, i);
      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
        pColInfoData->varmeta.length = colLen[i];
        pColInfoData->varmeta.allocLen = colLen[i];

        memcpy(pColInfoData->varmeta.offset, pStart, sizeof(int32_t) * numOfRows);
        pStart += sizeof(int32_t) * numOfRows;

        pColInfoData->pData = taosMemoryMalloc(colLen[i]);
      } else {
        memcpy(pColInfoData->nullbitmap, pStart, BitmapLen(numOfRows));
        pStart += BitmapLen(numOfRows);
      }

      memcpy(pColInfoData->pData, pStart, colLen[i]);
      pStart += colLen[i];
    }
H
Haojun Liao 已提交
2944 2945

    // data from mnode
L
Liu Jicong 已提交
2946
    relocateColumnData(pRes, pColList, block.pDataBlock);
2947
  }
2948

H
Haojun Liao 已提交
2949
  pRes->info.rows = numOfRows;
2950
  blockDataUpdateTsWindow(pRes);
2951

2952
  int64_t el = taosGetTimestampUs() - startTs;
2953

H
Haojun Liao 已提交
2954 2955
  pLoadInfo->totalRows += numOfRows;
  pLoadInfo->totalSize += compLen;
2956

H
Haojun Liao 已提交
2957 2958 2959
  if (total != NULL) {
    *total += numOfRows;
  }
2960

H
Haojun Liao 已提交
2961
  pLoadInfo->totalElapsed += el;
2962 2963
  return TSDB_CODE_SUCCESS;
}
2964

L
Liu Jicong 已提交
2965 2966
static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
2967
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
2968

L
Liu Jicong 已提交
2969
  int64_t              el = taosGetTimestampUs() - startTs;
H
Haojun Liao 已提交
2970 2971
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
  pLoadInfo->totalElapsed += el;
H
Haojun Liao 已提交
2972

2973
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
L
Liu Jicong 已提交
2974 2975 2976
  qDebug("%s all %" PRIzu " sources are exhausted, total rows: %" PRIu64 " bytes:%" PRIu64 ", elapsed:%.2f ms",
         GET_TASKID(pTaskInfo), totalSources, pLoadInfo->totalRows, pLoadInfo->totalSize,
         pLoadInfo->totalElapsed / 1000.0);
2977 2978 2979 2980 2981

  doSetOperatorCompleted(pOperator);
  return NULL;
}

L
Liu Jicong 已提交
2982 2983
static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
                                                   SExecTaskInfo* pTaskInfo) {
2984 2985 2986 2987 2988 2989 2990 2991 2992
  int32_t code = 0;
  int64_t startTs = taosGetTimestampUs();
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);

  while (1) {
    int32_t completed = 0;
    for (int32_t i = 0; i < totalSources; ++i) {
      SSourceDataInfo* pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, i);

2993
      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
2994
        completed += 1;
H
Haojun Liao 已提交
2995 2996
        continue;
      }
2997

2998
      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
2999 3000 3001
        continue;
      }

L
Liu Jicong 已提交
3002
      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3003
      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
3004

L
Liu Jicong 已提交
3005
      SSDataBlock*         pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3006
      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3007
      if (pRsp->numOfRows == 0) {
L
Liu Jicong 已提交
3008 3009
        qDebug("%s vgId:%d, taskID:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
               " try next",
3010
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i + 1, pDataInfo->totalRows,
H
Haojun Liao 已提交
3011
               pExchangeInfo->loadInfo.totalRows);
3012
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3013 3014 3015
        completed += 1;
        continue;
      }
H
Haojun Liao 已提交
3016

H
Haojun Liao 已提交
3017
      SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3018 3019
      code = setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
                                       pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3020
      if (code != 0) {
3021 3022 3023
        goto _error;
      }

3024 3025 3026
      if (pRsp->completed == 1) {
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
               ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
L
Liu Jicong 已提交
3027 3028
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
               pLoadInfo->totalRows, pLoadInfo->totalSize, i + 1, totalSources);
3029
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3030
      } else {
dengyihao's avatar
dengyihao 已提交
3031 3032
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
               ", totalBytes:%" PRIu64,
H
Haojun Liao 已提交
3033 3034
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
               pLoadInfo->totalSize);
3035 3036
      }

3037 3038
      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057
        code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
        if (code != TSDB_CODE_SUCCESS) {
          goto _error;
        }
      }

      return pExchangeInfo->pResult;
    }

    if (completed == totalSources) {
      return setAllSourcesCompleted(pOperator, startTs);
    }
  }

_error:
  pTaskInfo->code = code;
  return NULL;
}

L
Liu Jicong 已提交
3058 3059 3060
static SSDataBlock* concurrentlyLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3061

H
Haojun Liao 已提交
3062 3063 3064
  if (pOperator->status == OP_RES_TO_RETURN) {
    return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
  }
3065

L
Liu Jicong 已提交
3066
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3067 3068 3069
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3070
  for (int32_t i = 0; i < totalSources; ++i) {
3071 3072
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3073
      return NULL;
3074 3075 3076 3077
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3078 3079
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3080 3081

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3082 3083
  pOperator->status = OP_RES_TO_RETURN;
  return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
3084 3085
}

L
Liu Jicong 已提交
3086 3087 3088
static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3089

L
Liu Jicong 已提交
3090
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3091 3092 3093
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3094
  for (int32_t i = 0; i < totalSources; ++i) {
3095 3096
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3097 3098
      pTaskInfo->code = code;
      return code;
3099 3100 3101 3102
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3103 3104
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3105 3106

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3107
  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
3108

H
Haojun Liao 已提交
3109
  return TSDB_CODE_SUCCESS;
3110 3111
}

L
Liu Jicong 已提交
3112 3113 3114
static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3115

L
Liu Jicong 已提交
3116
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3117
  int64_t startTs = taosGetTimestampUs();
3118

L
Liu Jicong 已提交
3119
  while (1) {
3120 3121
    if (pExchangeInfo->current >= totalSources) {
      return setAllSourcesCompleted(pOperator, startTs);
3122
    }
3123

3124 3125 3126
    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
    tsem_wait(&pExchangeInfo->ready);

dengyihao's avatar
dengyihao 已提交
3127
    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
X
Xiaoyu Wang 已提交
3128
    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
3129

H
Haojun Liao 已提交
3130
    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
dengyihao's avatar
dengyihao 已提交
3131 3132
      qError("%s vgId:%d, taskID:0x%" PRIx64 " error happens, code:%s", GET_TASKID(pTaskInfo), pSource->addr.nodeId,
             pSource->taskId, tstrerror(pDataInfo->code));
H
Haojun Liao 已提交
3133 3134 3135 3136
      pOperator->pTaskInfo->code = pDataInfo->code;
      return NULL;
    }

L
Liu Jicong 已提交
3137
    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
H
Haojun Liao 已提交
3138
    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3139
    if (pRsp->numOfRows == 0) {
dengyihao's avatar
dengyihao 已提交
3140 3141
      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " %d of total completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
             " try next",
3142
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pExchangeInfo->current + 1,
H
Haojun Liao 已提交
3143
             pDataInfo->totalRows, pLoadInfo->totalRows);
H
Haojun Liao 已提交
3144

3145
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3146 3147 3148
      pExchangeInfo->current += 1;
      continue;
    }
H
Haojun Liao 已提交
3149

L
Liu Jicong 已提交
3150
    SSDataBlock*       pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3151
    SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
L
Liu Jicong 已提交
3152 3153
    int32_t            code =
        setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
3154
                                  pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3155 3156

    if (pRsp->completed == 1) {
H
Haojun Liao 已提交
3157
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
L
Liu Jicong 已提交
3158 3159 3160
             ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
             pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1, totalSources);
3161

3162
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3163 3164
      pExchangeInfo->current += 1;
    } else {
L
Liu Jicong 已提交
3165 3166 3167 3168
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
             ", totalBytes:%" PRIu64,
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
             pLoadInfo->totalSize);
3169 3170 3171 3172
    }

    return pExchangeInfo->pResult;
  }
3173 3174
}

L
Liu Jicong 已提交
3175
static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
3176
  if (OPTR_IS_OPENED(pOperator)) {
H
Haojun Liao 已提交
3177 3178 3179
    return TSDB_CODE_SUCCESS;
  }

L
Liu Jicong 已提交
3180
  SExchangeInfo* pExchangeInfo = pOperator->info;
H
Haojun Liao 已提交
3181 3182 3183 3184 3185 3186 3187 3188 3189
  if (pExchangeInfo->seqLoadData) {
    // do nothing for sequentially load data
  } else {
    int32_t code = prepareConcurrentlyLoad(pOperator);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
    }
  }

3190
  OPTR_SET_OPENED(pOperator);
H
Haojun Liao 已提交
3191 3192 3193
  return TSDB_CODE_SUCCESS;
}

3194
static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3195 3196
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3197

3198
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
3199
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3200 3201
    return NULL;
  }
3202

L
Liu Jicong 已提交
3203
  size_t               totalSources = taosArrayGetSize(pExchangeInfo->pSources);
H
Haojun Liao 已提交
3204
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
H
Haojun Liao 已提交
3205

3206
  if (pOperator->status == OP_EXEC_DONE) {
L
Liu Jicong 已提交
3207 3208 3209
    qDebug("%s all %" PRIzu " source(s) are exhausted, total rows:%" PRIu64 " bytes:%" PRIu64 ", elapsed:%.2f ms",
           GET_TASKID(pTaskInfo), totalSources, pLoadInfo->totalRows, pLoadInfo->totalSize,
           pLoadInfo->totalElapsed / 1000.0);
3210 3211 3212 3213 3214 3215
    return NULL;
  }

  if (pExchangeInfo->seqLoadData) {
    return seqLoadRemoteData(pOperator);
  } else {
H
Haojun Liao 已提交
3216
    return concurrentlyLoadRemoteData(pOperator);
3217
  }
H
Haojun Liao 已提交
3218

3219
#if 0
H
Haojun Liao 已提交
3220
  _error:
wafwerar's avatar
wafwerar 已提交
3221 3222
  taosMemoryFreeClear(pMsg);
  taosMemoryFreeClear(pMsgSendInfo);
H
Haojun Liao 已提交
3223 3224 3225

  terrno = pTaskInfo->code;
  return NULL;
3226
#endif
H
Haojun Liao 已提交
3227
}
3228

H
Haojun Liao 已提交
3229
static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo) {
3230
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
H
Haojun Liao 已提交
3231 3232
  if (pInfo->pSourceDataInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
3233 3234
  }

L
Liu Jicong 已提交
3235
  for (int32_t i = 0; i < numOfSources; ++i) {
3236
    SSourceDataInfo dataInfo = {0};
H
Haojun Liao 已提交
3237
    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
L
Liu Jicong 已提交
3238 3239
    dataInfo.pEx = pInfo;
    dataInfo.index = i;
3240

H
Haojun Liao 已提交
3241 3242 3243 3244 3245 3246 3247 3248 3249 3250
    void* ret = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
    if (ret == NULL) {
      taosArrayDestroy(pInfo->pSourceDataInfo);
      return TSDB_CODE_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
}

H
Haojun Liao 已提交
3251
SOperatorInfo* createExchangeOperatorInfo(const SNodeList* pSources, SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
3252
  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
wafwerar's avatar
wafwerar 已提交
3253
  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
3254 3255

  if (pInfo == NULL || pOperator == NULL) {
wafwerar's avatar
wafwerar 已提交
3256 3257
    taosMemoryFreeClear(pInfo);
    taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3258 3259
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return NULL;
H
Haojun Liao 已提交
3260 3261
  }

H
Haojun Liao 已提交
3262
  size_t numOfSources = LIST_LENGTH(pSources);
H
Haojun Liao 已提交
3263
  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
H
Haojun Liao 已提交
3264 3265 3266
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
  if (pInfo->pSourceDataInfo == NULL || pInfo->pSources == NULL) {
    goto _error;
H
Haojun Liao 已提交
3267 3268
  }

L
Liu Jicong 已提交
3269 3270
  for (int32_t i = 0; i < numOfSources; ++i) {
    SNodeListNode* pNode = nodesListGetNode((SNodeList*)pSources, i);
H
Haojun Liao 已提交
3271 3272
    taosArrayPush(pInfo->pSources, pNode);
  }
3273

H
Haojun Liao 已提交
3274 3275 3276
  int32_t code = initDataSource(numOfSources, pInfo);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3277
  }
H
Haojun Liao 已提交
3278

dengyihao's avatar
dengyihao 已提交
3279
  pInfo->pResult = pBlock;
3280 3281 3282
  pInfo->seqLoadData = true;

  tsem_init(&pInfo->ready, 0, 0);
3283

dengyihao's avatar
dengyihao 已提交
3284
  pOperator->name = "ExchangeOperator";
X
Xiaoyu Wang 已提交
3285
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
3286
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
3287 3288 3289 3290
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->numOfOutput = pBlock->info.numOfCols;
  pOperator->pTaskInfo = pTaskInfo;
3291

L
Liu Jicong 已提交
3292 3293
  pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                         destroyExchangeOperatorInfo, NULL, NULL, NULL);
3294

S
Shengliang Guan 已提交
3295
#if 1
L
Liu Jicong 已提交
3296
  {  // todo refactor
H
Haojun Liao 已提交
3297 3298 3299
    SRpcInit rpcInit;
    memset(&rpcInit, 0, sizeof(rpcInit));
    rpcInit.localPort = 0;
H
Haojun Liao 已提交
3300
    rpcInit.label = "EX";
H
Haojun Liao 已提交
3301
    rpcInit.numOfThreads = 1;
S
Shengliang Guan 已提交
3302
    rpcInit.cfp = qProcessFetchRsp;
S
Shengliang Guan 已提交
3303
    rpcInit.sessions = tsMaxConnections;
H
Haojun Liao 已提交
3304
    rpcInit.connType = TAOS_CONN_CLIENT;
L
Liu Jicong 已提交
3305
    rpcInit.user = (char*)"root";
S
Shengliang Guan 已提交
3306
    rpcInit.idleTime = tsShellActivityTimer * 1000;
H
Haojun Liao 已提交
3307
    rpcInit.ckey = "key";
S
Shengliang Guan 已提交
3308
    rpcInit.spi = 1;
L
Liu Jicong 已提交
3309
    rpcInit.secret = (char*)"dcc5bed04851fec854c035b2e40263b6";
H
Haojun Liao 已提交
3310 3311 3312

    pInfo->pTransporter = rpcOpen(&rpcInit);
    if (pInfo->pTransporter == NULL) {
L
Liu Jicong 已提交
3313
      return NULL;  // todo
H
Haojun Liao 已提交
3314 3315
    }
  }
S
Shengliang 已提交
3316
#endif
3317

3318
  return pOperator;
H
Haojun Liao 已提交
3319

L
Liu Jicong 已提交
3320
_error:
H
Haojun Liao 已提交
3321
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3322
    destroyExchangeOperatorInfo(pInfo, numOfSources);
H
Haojun Liao 已提交
3323 3324
  }

wafwerar's avatar
wafwerar 已提交
3325 3326
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3327
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
H
Haojun Liao 已提交
3328
  return NULL;
3329 3330
}

dengyihao's avatar
dengyihao 已提交
3331 3332
static int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                                const char* pKey);
3333

3334
static void destroySortedMergeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
3335
  SSortedMergeOperatorInfo* pInfo = (SSortedMergeOperatorInfo*)param;
H
Haojun Liao 已提交
3336
  taosArrayDestroy(pInfo->pSortInfo);
3337 3338 3339
  taosArrayDestroy(pInfo->groupInfo);

  if (pInfo->pSortHandle != NULL) {
H
Haojun Liao 已提交
3340
    tsortDestroySortHandle(pInfo->pSortHandle);
3341 3342
  }

H
Haojun Liao 已提交
3343
  blockDataDestroy(pInfo->binfo.pRes);
H
Haojun Liao 已提交
3344
  cleanupAggSup(&pInfo->aggSup);
3345
}
H
Haojun Liao 已提交
3346

H
Haojun Liao 已提交
3347
// TODO merge aggregate super table
L
Liu Jicong 已提交
3348
static void appendOneRowToDataBlock(SSDataBlock* pBlock, STupleHandle* pTupleHandle) {
3349 3350
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, i);
3351

H
Haojun Liao 已提交
3352
    bool isNull = tsortIsNullVal(pTupleHandle, i);
H
Haojun Liao 已提交
3353 3354 3355
    if (isNull) {
      colDataAppend(pColInfo, pBlock->info.rows, NULL, true);
    } else {
H
Haojun Liao 已提交
3356
      char* pData = tsortGetValue(pTupleHandle, i);
H
Haojun Liao 已提交
3357 3358
      colDataAppend(pColInfo, pBlock->info.rows, pData, false);
    }
3359 3360
  }

3361 3362
  pBlock->info.rows += 1;
}
3363

H
Haojun Liao 已提交
3364
SSDataBlock* getSortedBlockData(SSortHandle* pHandle, SSDataBlock* pDataBlock, int32_t capacity) {
3365
  blockDataCleanup(pDataBlock);
D
dapan1121 已提交
3366
  blockDataEnsureCapacity(pDataBlock, capacity);
3367

wmmhello's avatar
wmmhello 已提交
3368 3369
  blockDataEnsureCapacity(pDataBlock, capacity);

L
Liu Jicong 已提交
3370
  while (1) {
H
Haojun Liao 已提交
3371
    STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
3372 3373 3374
    if (pTupleHandle == NULL) {
      break;
    }
3375

3376 3377 3378
    appendOneRowToDataBlock(pDataBlock, pTupleHandle);
    if (pDataBlock->info.rows >= capacity) {
      return pDataBlock;
H
Haojun Liao 已提交
3379 3380
    }
  }
3381

L
Liu Jicong 已提交
3382
  return (pDataBlock->info.rows > 0) ? pDataBlock : NULL;
3383
}
3384

3385
SSDataBlock* loadNextDataBlock(void* param) {
L
Liu Jicong 已提交
3386
  SOperatorInfo* pOperator = (SOperatorInfo*)param;
3387
  return pOperator->fpSet.getNextFn(pOperator);
3388 3389
}

L
Liu Jicong 已提交
3390
static bool needToMerge(SSDataBlock* pBlock, SArray* groupInfo, char** buf, int32_t rowIndex) {
3391 3392 3393 3394
  size_t size = taosArrayGetSize(groupInfo);
  if (size == 0) {
    return true;
  }
3395

3396 3397
  for (int32_t i = 0; i < size; ++i) {
    int32_t* index = taosArrayGet(groupInfo, i);
3398

3399
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
L
Liu Jicong 已提交
3400
    bool             isNull = colDataIsNull(pColInfo, rowIndex, pBlock->info.rows, NULL);
3401

3402 3403 3404
    if ((isNull && buf[i] != NULL) || (!isNull && buf[i] == NULL)) {
      return false;
    }
3405

3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418
    char* pCell = colDataGetData(pColInfo, rowIndex);
    if (IS_VAR_DATA_TYPE(pColInfo->info.type)) {
      if (varDataLen(pCell) != varDataLen(buf[i])) {
        return false;
      } else {
        if (memcmp(varDataVal(pCell), varDataVal(buf[i]), varDataLen(pCell)) != 0) {
          return false;
        }
      }
    } else {
      if (memcmp(pCell, buf[i], pColInfo->info.bytes) != 0) {
        return false;
      }
3419 3420 3421
    }
  }

3422
  return 0;
3423 3424
}

L
Liu Jicong 已提交
3425 3426 3427
static void doMergeResultImpl(SSortedMergeOperatorInfo* pInfo, SqlFunctionCtx* pCtx, int32_t numOfExpr,
                              int32_t rowIndex) {
  for (int32_t j = 0; j < numOfExpr; ++j) {  // TODO set row index
3428
    pCtx[j].startRow = rowIndex;
3429 3430
  }

3431 3432
  for (int32_t j = 0; j < numOfExpr; ++j) {
    int32_t functionId = pCtx[j].functionId;
L
Liu Jicong 已提交
3433 3434 3435 3436 3437 3438 3439 3440 3441
    //    pCtx[j].fpSet->addInput(&pCtx[j]);

    //    if (functionId < 0) {
    //      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
    //      doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
    //    } else {
    //      assert(!TSDB_FUNC_IS_SCALAR(functionId));
    //      aAggs[functionId].mergeFunc(&pCtx[j]);
    //    }
3442
  }
3443
}
3444

L
Liu Jicong 已提交
3445 3446
static void doFinalizeResultImpl(SqlFunctionCtx* pCtx, int32_t numOfExpr) {
  for (int32_t j = 0; j < numOfExpr; ++j) {
3447 3448 3449 3450
    int32_t functionId = pCtx[j].functionId;
    //    if (functionId == FUNC_TAG_DUMMY || functionId == FUNC_TS_DUMMY) {
    //      continue;
    //    }
3451

3452 3453 3454 3455
    //    if (functionId < 0) {
    //      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
    //      doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_FINALIZE);
    //    } else {
dengyihao's avatar
dengyihao 已提交
3456
    //    pCtx[j].fpSet.finalize(&pCtx[j]);
3457 3458
  }
}
3459

3460
static bool saveCurrentTuple(char** rowColData, SArray* pColumnList, SSDataBlock* pBlock, int32_t rowIndex) {
L
Liu Jicong 已提交
3461
  int32_t size = (int32_t)taosArrayGetSize(pColumnList);
3462

L
Liu Jicong 已提交
3463 3464
  for (int32_t i = 0; i < size; ++i) {
    int32_t*         index = taosArrayGet(pColumnList, i);
3465
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
H
Haojun Liao 已提交
3466

3467 3468 3469
    char* data = colDataGetData(pColInfo, rowIndex);
    memcpy(rowColData[i], data, colDataGetLength(pColInfo, rowIndex));
  }
3470

3471 3472
  return true;
}
3473

3474 3475
static void doMergeImpl(SOperatorInfo* pOperator, int32_t numOfExpr, SSDataBlock* pBlock) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
3476

3477
  SqlFunctionCtx* pCtx = pInfo->binfo.pCtx;
L
Liu Jicong 已提交
3478
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
3479 3480
    pCtx[i].size = 1;
  }
3481

L
Liu Jicong 已提交
3482
  for (int32_t i = 0; i < pBlock->info.rows; ++i) {
3483 3484 3485 3486 3487 3488 3489 3490 3491
    if (!pInfo->hasGroupVal) {
      ASSERT(i == 0);
      doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
      pInfo->hasGroupVal = saveCurrentTuple(pInfo->groupVal, pInfo->groupInfo, pBlock, i);
    } else {
      if (needToMerge(pBlock, pInfo->groupInfo, pInfo->groupVal, i)) {
        doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
      } else {
        doFinalizeResultImpl(pCtx, numOfExpr);
H
Haojun Liao 已提交
3492
        int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfOutput, NULL);
3493
        //        setTagValueForMultipleRows(pCtx, pOperator->numOfOutput, numOfRows);
3494

3495
        // TODO check for available buffer;
H
Haojun Liao 已提交
3496

3497 3498 3499 3500 3501
        // next group info data
        pInfo->binfo.pRes->info.rows += numOfRows;
        for (int32_t j = 0; j < numOfExpr; ++j) {
          if (pCtx[j].functionId < 0) {
            continue;
3502
          }
3503

H
Haojun Liao 已提交
3504
          pCtx[j].fpSet.process(&pCtx[j]);
3505
        }
3506 3507 3508

        doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
        pInfo->hasGroupVal = saveCurrentTuple(pInfo->groupVal, pInfo->groupInfo, pBlock, i);
H
Haojun Liao 已提交
3509
      }
3510 3511 3512 3513
    }
  }
}

3514 3515
static SSDataBlock* doMerge(SOperatorInfo* pOperator) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
L
Liu Jicong 已提交
3516
  SSortHandle*              pHandle = pInfo->pSortHandle;
3517

3518
  SSDataBlock* pDataBlock = createOneDataBlock(pInfo->binfo.pRes, false);
3519
  blockDataEnsureCapacity(pDataBlock, pOperator->resultInfo.capacity);
3520

L
Liu Jicong 已提交
3521
  while (1) {
3522
    blockDataCleanup(pDataBlock);
3523
    while (1) {
H
Haojun Liao 已提交
3524
      STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
3525 3526
      if (pTupleHandle == NULL) {
        break;
3527
      }
3528

3529 3530
      // build datablock for merge for one group
      appendOneRowToDataBlock(pDataBlock, pTupleHandle);
3531
      if (pDataBlock->info.rows >= pOperator->resultInfo.capacity) {
3532 3533
        break;
      }
3534
    }
3535

3536 3537 3538
    if (pDataBlock->info.rows == 0) {
      break;
    }
3539

3540
    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pDataBlock, TSDB_ORDER_ASC, true);
L
Liu Jicong 已提交
3541 3542
    //  updateOutputBuf(&pInfo->binfo, &pAggInfo->bufCapacity, pBlock->info.rows * pAggInfo->resultRowFactor,
    //  pOperator->pRuntimeEnv, true);
3543 3544 3545
    doMergeImpl(pOperator, pOperator->numOfOutput, pDataBlock);
    // flush to tuple store, and after all data have been handled, return to upstream node or sink node
  }
3546

3547
  doFinalizeResultImpl(pInfo->binfo.pCtx, pOperator->numOfOutput);
H
Haojun Liao 已提交
3548
  int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfOutput, NULL);
3549
  //        setTagValueForMultipleRows(pCtx, pOperator->numOfOutput, numOfRows);
3550

3551
  // TODO check for available buffer;
3552

3553 3554
  // next group info data
  pInfo->binfo.pRes->info.rows += numOfRows;
L
Liu Jicong 已提交
3555
  return (pInfo->binfo.pRes->info.rows > 0) ? pInfo->binfo.pRes : NULL;
3556
}
3557

3558
static SSDataBlock* doSortedMerge(SOperatorInfo* pOperator) {
3559 3560
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
3561 3562
  }

L
Liu Jicong 已提交
3563
  SExecTaskInfo*            pTaskInfo = pOperator->pTaskInfo;
3564
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
H
Haojun Liao 已提交
3565
  if (pOperator->status == OP_RES_TO_RETURN) {
3566
    return getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity);
3567 3568
  }

3569
  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
dengyihao's avatar
dengyihao 已提交
3570 3571
  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, NULL, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize,
                                             numOfBufPage, pInfo->binfo.pRes, "GET_TASKID(pTaskInfo)");
H
Haojun Liao 已提交
3572

H
Haojun Liao 已提交
3573
  tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock);
3574

L
Liu Jicong 已提交
3575
  for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
wmmhello's avatar
wmmhello 已提交
3576
    SSortSource* ps = taosMemoryCalloc(1, sizeof(SSortSource));
H
Haojun Liao 已提交
3577
    ps->param = pOperator->pDownstream[i];
H
Haojun Liao 已提交
3578
    tsortAddSource(pInfo->pSortHandle, ps);
3579 3580
  }

H
Haojun Liao 已提交
3581
  int32_t code = tsortOpen(pInfo->pSortHandle);
3582
  if (code != TSDB_CODE_SUCCESS) {
3583
    longjmp(pTaskInfo->env, terrno);
3584 3585
  }

H
Haojun Liao 已提交
3586
  pOperator->status = OP_RES_TO_RETURN;
3587
  return doMerge(pOperator);
3588
}
3589

L
Liu Jicong 已提交
3590 3591
static int32_t initGroupCol(SExprInfo* pExprInfo, int32_t numOfCols, SArray* pGroupInfo,
                            SSortedMergeOperatorInfo* pInfo) {
3592 3593
  if (pGroupInfo == NULL || taosArrayGetSize(pGroupInfo) == 0) {
    return 0;
H
Haojun Liao 已提交
3594 3595
  }

3596 3597 3598 3599 3600 3601 3602 3603
  int32_t len = 0;
  SArray* plist = taosArrayInit(3, sizeof(SColumn));
  pInfo->groupInfo = taosArrayInit(3, sizeof(int32_t));

  if (plist == NULL || pInfo->groupInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

L
Liu Jicong 已提交
3604 3605
  size_t numOfGroupCol = taosArrayGetSize(pInfo->groupInfo);
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3606
    SColumn* pCol = taosArrayGet(pGroupInfo, i);
L
Liu Jicong 已提交
3607
    for (int32_t j = 0; j < numOfCols; ++j) {
H
Haojun Liao 已提交
3608
      SExprInfo* pe = &pExprInfo[j];
3609
      if (pe->base.resSchema.slotId == pCol->colId) {
3610 3611
        taosArrayPush(plist, pCol);
        taosArrayPush(pInfo->groupInfo, &j);
H
Haojun Liao 已提交
3612
        len += pCol->bytes;
3613 3614
        break;
      }
H
Haojun Liao 已提交
3615 3616 3617
    }
  }

3618
  ASSERT(taosArrayGetSize(pGroupInfo) == taosArrayGetSize(plist));
H
Haojun Liao 已提交
3619

wafwerar's avatar
wafwerar 已提交
3620
  pInfo->groupVal = taosMemoryCalloc(1, (POINTER_BYTES * numOfGroupCol + len));
3621 3622 3623 3624
  if (pInfo->groupVal == NULL) {
    taosArrayDestroy(plist);
    return TSDB_CODE_OUT_OF_MEMORY;
  }
H
Haojun Liao 已提交
3625

3626
  int32_t offset = 0;
L
Liu Jicong 已提交
3627 3628
  char*   start = (char*)(pInfo->groupVal + (POINTER_BYTES * numOfGroupCol));
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3629 3630
    pInfo->groupVal[i] = start + offset;
    SColumn* pCol = taosArrayGet(plist, i);
H
Haojun Liao 已提交
3631
    offset += pCol->bytes;
3632
  }
H
Haojun Liao 已提交
3633

3634
  taosArrayDestroy(plist);
H
Haojun Liao 已提交
3635

3636 3637
  return TSDB_CODE_SUCCESS;
}
H
Haojun Liao 已提交
3638

L
Liu Jicong 已提交
3639 3640 3641
SOperatorInfo* createSortedMergeOperatorInfo(SOperatorInfo** downstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
                                             int32_t num, SArray* pSortInfo, SArray* pGroupInfo,
                                             SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
3642
  SSortedMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortedMergeOperatorInfo));
L
Liu Jicong 已提交
3643
  SOperatorInfo*            pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
3644
  if (pInfo == NULL || pOperator == NULL) {
3645
    goto _error;
3646
  }
H
Haojun Liao 已提交
3647

3648
  pInfo->binfo.pCtx = createSqlFunctionCtx(pExprInfo, num, &pInfo->binfo.rowCellInfoOffset);
3649
  initResultRowInfo(&pInfo->binfo.resultRowInfo, (int32_t)1);
H
Haojun Liao 已提交
3650

3651 3652 3653
  if (pInfo->binfo.pCtx == NULL || pInfo->binfo.pRes == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
3654

dengyihao's avatar
dengyihao 已提交
3655
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
3656
  int32_t code = doInitAggInfoSup(&pInfo->aggSup, pInfo->binfo.pCtx, num, keyBufSize, pTaskInfo->id.str);
3657 3658 3659
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3660

H
Haojun Liao 已提交
3661
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, pTaskInfo);
H
Haojun Liao 已提交
3662
  code = initGroupCol(pExprInfo, num, pGroupInfo, pInfo);
3663 3664 3665
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3666

L
Liu Jicong 已提交
3667 3668 3669 3670 3671
  //  pInfo->resultRowFactor = (int32_t)(getRowNumForMultioutput(pRuntimeEnv->pQueryAttr,
  //      pRuntimeEnv->pQueryAttr->topBotQuery, false));
  pInfo->sortBufSize = 1024 * 16;  // 1MB
  pInfo->bufPageSize = 1024;
  pInfo->pSortInfo = pSortInfo;
H
Haojun Liao 已提交
3672

3673
  pOperator->resultInfo.capacity = blockDataGetCapacityInRow(pInfo->binfo.pRes, pInfo->bufPageSize);
H
Haojun Liao 已提交
3674

L
Liu Jicong 已提交
3675
  pOperator->name = "SortedMerge";
X
Xiaoyu Wang 已提交
3676
  // pOperator->operatorType = OP_SortedMerge;
3677
  pOperator->blocking = true;
L
Liu Jicong 已提交
3678 3679 3680 3681
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->numOfOutput = num;
  pOperator->pExpr = pExprInfo;
H
Haojun Liao 已提交
3682

L
Liu Jicong 已提交
3683
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
3684

3685 3686
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSortedMerge, NULL, NULL, destroySortedMergeOperatorInfo,
                                         NULL, NULL, NULL);
3687 3688 3689
  code = appendDownstream(pOperator, downstream, numOfDownstream);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3690
  }
H
Haojun Liao 已提交
3691

3692
  return pOperator;
H
Haojun Liao 已提交
3693

L
Liu Jicong 已提交
3694
_error:
3695
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3696
    destroySortedMergeOperatorInfo(pInfo, num);
H
Haojun Liao 已提交
3697 3698
  }

wafwerar's avatar
wafwerar 已提交
3699 3700
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
3701 3702
  terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
  return NULL;
H
Haojun Liao 已提交
3703 3704
}

3705
static SSDataBlock* doSort(SOperatorInfo* pOperator) {
3706 3707 3708 3709
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }

L
Liu Jicong 已提交
3710
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3711 3712
  SSortOperatorInfo* pInfo = pOperator->info;

H
Haojun Liao 已提交
3713
  if (pOperator->status == OP_RES_TO_RETURN) {
H
Haojun Liao 已提交
3714
    return getSortedBlockData(pInfo->pSortHandle, pInfo->pDataBlock, pInfo->numOfRowsInRes);
H
Haojun Liao 已提交
3715 3716
  }

3717
  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
dengyihao's avatar
dengyihao 已提交
3718 3719
  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, pInfo->inputSlotMap, SORT_SINGLESOURCE_SORT,
                                             pInfo->bufPageSize, numOfBufPage, pInfo->pDataBlock, pTaskInfo->id.str);
3720

H
Haojun Liao 已提交
3721
  tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock);
H
Haojun Liao 已提交
3722

wmmhello's avatar
wmmhello 已提交
3723
  SSortSource* ps = taosMemoryCalloc(1, sizeof(SSortSource));
H
Haojun Liao 已提交
3724
  ps->param = pOperator->pDownstream[0];
H
Haojun Liao 已提交
3725
  tsortAddSource(pInfo->pSortHandle, ps);
3726

H
Haojun Liao 已提交
3727
  int32_t code = tsortOpen(pInfo->pSortHandle);
3728
  taosMemoryFreeClear(ps);
3729
  if (code != TSDB_CODE_SUCCESS) {
3730
    longjmp(pTaskInfo->env, terrno);
3731
  }
3732

H
Haojun Liao 已提交
3733
  pOperator->status = OP_RES_TO_RETURN;
H
Haojun Liao 已提交
3734
  return getSortedBlockData(pInfo->pSortHandle, pInfo->pDataBlock, pInfo->numOfRowsInRes);
3735 3736
}

dengyihao's avatar
dengyihao 已提交
3737 3738
SOperatorInfo* createSortOperatorInfo(SOperatorInfo* downstream, SSDataBlock* pResBlock, SArray* pSortInfo,
                                      SArray* pIndexMap, SExecTaskInfo* pTaskInfo) {
H
Haojun Liao 已提交
3739
  SSortOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortOperatorInfo));
L
Liu Jicong 已提交
3740
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
dengyihao's avatar
dengyihao 已提交
3741
  int32_t            rowSize = pResBlock->info.rowSize;
3742 3743

  if (pInfo == NULL || pOperator == NULL || rowSize > 100 * 1024 * 1024) {
wafwerar's avatar
wafwerar 已提交
3744 3745
    taosMemoryFreeClear(pInfo);
    taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3746 3747 3748
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return NULL;
  }
3749

dengyihao's avatar
dengyihao 已提交
3750
  pInfo->bufPageSize = rowSize < 1024 ? 1024 * 2 : rowSize * 2;  // there are headers, so pageSize = rowSize + header
3751

wmmhello's avatar
wmmhello 已提交
3752
  pInfo->sortBufSize = pInfo->bufPageSize * 16;  // TODO dynamic set the available sort buffer
L
Liu Jicong 已提交
3753 3754 3755
  pInfo->numOfRowsInRes = 1024;
  pInfo->pDataBlock = pResBlock;
  pInfo->pSortInfo = pSortInfo;
3756
  pInfo->inputSlotMap = pIndexMap;
H
Haojun Liao 已提交
3757

dengyihao's avatar
dengyihao 已提交
3758
  pOperator->name = "SortOperator";
L
Liu Jicong 已提交
3759
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_SORT;
3760
  pOperator->blocking = true;
dengyihao's avatar
dengyihao 已提交
3761 3762
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3763

dengyihao's avatar
dengyihao 已提交
3764
  pOperator->pTaskInfo = pTaskInfo;
L
Liu Jicong 已提交
3765 3766
  pOperator->fpSet =
      createOperatorFpSet(operatorDummyOpenFn, doSort, NULL, NULL, destroyOrderOperatorInfo, NULL, NULL, NULL);
3767

3768
  int32_t code = appendDownstream(pOperator, &downstream, 1);
3769
  return pOperator;
H
Haojun Liao 已提交
3770

dengyihao's avatar
dengyihao 已提交
3771
_error:
H
Haojun Liao 已提交
3772 3773 3774 3775
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  taosMemoryFree(pInfo);
  taosMemoryFree(pOperator);
  return NULL;
3776 3777
}

3778 3779
int32_t getTableScanOrder(SOperatorInfo* pOperator) {
  if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
H
Haojun Liao 已提交
3780
    if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
3781
      return TSDB_ORDER_ASC;
H
Haojun Liao 已提交
3782 3783
    } else {
      return getTableScanOrder(pOperator->pDownstream[0]);
3784 3785 3786 3787 3788 3789
    }
  }

  STableScanInfo* pTableScanInfo = pOperator->info;
  return pTableScanInfo->cond.order;
}
3790 3791

// this is a blocking operator
L
Liu Jicong 已提交
3792
static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
H
Haojun Liao 已提交
3793 3794
  if (OPTR_IS_OPENED(pOperator)) {
    return TSDB_CODE_SUCCESS;
3795 3796
  }

H
Haojun Liao 已提交
3797
  SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
3798
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3799

dengyihao's avatar
dengyihao 已提交
3800
  SOptrBasicInfo* pInfo = &pAggInfo->binfo;
3801

H
Haojun Liao 已提交
3802
  int32_t        order = TSDB_ORDER_ASC;
H
Haojun Liao 已提交
3803
  SOperatorInfo* downstream = pOperator->pDownstream[0];
3804

H
Haojun Liao 已提交
3805 3806
  bool newgroup = true;
  while (1) {
H
Haojun Liao 已提交
3807
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
3808
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
3809
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
3810 3811 3812 3813

    if (pBlock == NULL) {
      break;
    }
H
Haojun Liao 已提交
3814 3815 3816
    //    if (pAggInfo->current != NULL) {
    //      setTagValue(pOperator, pAggInfo->current->pTable, pInfo->pCtx, pOperator->numOfOutput);
    //    }
3817

3818 3819
    // there is an scalar expression that needs to be calculated before apply the group aggregation.
    if (pAggInfo->pScalarExprInfo != NULL) {
X
Xiaoyu Wang 已提交
3820 3821
      int32_t code = projectApplyFunctions(pAggInfo->pScalarExprInfo, pBlock, pBlock, pAggInfo->pScalarCtx,
                                           pAggInfo->numOfScalarExpr, NULL);
3822 3823 3824 3825
      if (code != TSDB_CODE_SUCCESS) {
        pTaskInfo->code = code;
        longjmp(pTaskInfo->env, pTaskInfo->code);
      }
3826 3827
    }

3828
    // the pDataBlock are always the same one, no need to call this again
H
Haojun Liao 已提交
3829
    setExecutionContext(pOperator->numOfOutput, pBlock->info.groupId, pTaskInfo, pAggInfo);
3830
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, true);
H
Haojun Liao 已提交
3831
    doAggregateImpl(pOperator, 0, pInfo->pCtx);
3832

dengyihao's avatar
dengyihao 已提交
3833
#if 0  // test for encode/decode result info
3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844
    if(pOperator->encodeResultRow){
      char *result = NULL;
      int32_t length = 0;
      SAggSupporter   *pSup = &pAggInfo->aggSup;
      pOperator->encodeResultRow(pOperator, pSup, pInfo, &result, &length);
      taosHashClear(pSup->pResultRowHashTable);
      pInfo->resultRowInfo.size = 0;
      pOperator->decodeResultRow(pOperator, pSup, pInfo, result, length);
      if(result){
        taosMemoryFree(result);
      }
3845
    }
3846
#endif
3847 3848
  }

H
Haojun Liao 已提交
3849 3850 3851
  closeAllResultRows(&pAggInfo->binfo.resultRowInfo);
  finalizeMultiTupleQueryResult(pAggInfo->binfo.pCtx, pOperator->numOfOutput, pAggInfo->aggSup.pResultBuf,
                                &pAggInfo->binfo.resultRowInfo, pAggInfo->binfo.rowCellInfoOffset);
H
Haojun Liao 已提交
3852

3853
  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, false);
H
Haojun Liao 已提交
3854 3855 3856 3857
  OPTR_SET_OPENED(pOperator);
  return TSDB_CODE_SUCCESS;
}

3858
static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3859
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3860 3861 3862 3863 3864 3865
  SOptrBasicInfo*   pInfo = &pAggInfo->binfo;

  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }

L
Liu Jicong 已提交
3866
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3867
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
H
Haojun Liao 已提交
3868 3869 3870 3871
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
    return NULL;
  }

H
Haojun Liao 已提交
3872
  blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
3873
  doBuildResultDatablock(pInfo, &pAggInfo->groupResInfo, pOperator->pExpr, pAggInfo->aggSup.pResultBuf);
H
Haojun Liao 已提交
3874 3875 3876
  if (pInfo->pRes->info.rows == 0 || !hasRemainDataInCurrentGroup(&pAggInfo->groupResInfo)) {
    doSetOperatorCompleted(pOperator);
  }
3877

H
Haojun Liao 已提交
3878
  doSetOperatorCompleted(pOperator);
L
Liu Jicong 已提交
3879
  return (blockDataGetNumOfRows(pInfo->pRes) != 0) ? pInfo->pRes : NULL;
3880 3881
}

dengyihao's avatar
dengyihao 已提交
3882 3883
void aggEncodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char** result,
                        int32_t* length) {
wmmhello's avatar
wmmhello 已提交
3884
  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
3885
  size_t  keyLen = sizeof(uint64_t) * 2;  // estimate the key length
wmmhello's avatar
wmmhello 已提交
3886
  int32_t totalSize = sizeof(int32_t) + size * (sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize);
wafwerar's avatar
wafwerar 已提交
3887
  *result = taosMemoryCalloc(1, totalSize);
L
Liu Jicong 已提交
3888
  if (*result == NULL) {
3889
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
wmmhello's avatar
wmmhello 已提交
3890 3891 3892
  }
  *(int32_t*)(*result) = size;
  int32_t offset = sizeof(int32_t);
3893 3894

  // prepare memory
3895
  SResultRowPosition* pos = &pInfo->resultRowInfo.cur;
dengyihao's avatar
dengyihao 已提交
3896 3897
  void*               pPage = getBufPage(pSup->pResultBuf, pos->pageId);
  SResultRow*         pRow = (SResultRow*)((char*)pPage + pos->offset);
3898 3899 3900
  setBufPageDirty(pPage, true);
  releaseBufPage(pSup->pResultBuf, pPage);

dengyihao's avatar
dengyihao 已提交
3901
  void* pIter = taosHashIterate(pSup->pResultRowHashTable, NULL);
wmmhello's avatar
wmmhello 已提交
3902
  while (pIter) {
dengyihao's avatar
dengyihao 已提交
3903
    void*               key = taosHashGetKey(pIter, &keyLen);
3904
    SResultRowPosition* p1 = (SResultRowPosition*)pIter;
3905

dengyihao's avatar
dengyihao 已提交
3906
    pPage = (SFilePage*)getBufPage(pSup->pResultBuf, p1->pageId);
3907
    pRow = (SResultRow*)((char*)pPage + p1->offset);
3908 3909
    setBufPageDirty(pPage, true);
    releaseBufPage(pSup->pResultBuf, pPage);
wmmhello's avatar
wmmhello 已提交
3910 3911 3912

    // recalculate the result size
    int32_t realTotalSize = offset + sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize;
L
Liu Jicong 已提交
3913 3914 3915
    if (realTotalSize > totalSize) {
      char* tmp = taosMemoryRealloc(*result, realTotalSize);
      if (tmp == NULL) {
wmmhello's avatar
wmmhello 已提交
3916
        terrno = TSDB_CODE_OUT_OF_MEMORY;
wafwerar's avatar
wafwerar 已提交
3917
        taosMemoryFree(*result);
wmmhello's avatar
wmmhello 已提交
3918
        *result = NULL;
3919
        longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
L
Liu Jicong 已提交
3920
      } else {
wmmhello's avatar
wmmhello 已提交
3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932
        *result = tmp;
      }
    }
    // save key
    *(int32_t*)(*result + offset) = keyLen;
    offset += sizeof(int32_t);
    memcpy(*result + offset, key, keyLen);
    offset += keyLen;

    // save value
    *(int32_t*)(*result + offset) = pSup->resultRowSize;
    offset += sizeof(int32_t);
3933
    memcpy(*result + offset, pRow, pSup->resultRowSize);
wmmhello's avatar
wmmhello 已提交
3934 3935 3936 3937 3938
    offset += pSup->resultRowSize;

    pIter = taosHashIterate(pSup->pResultRowHashTable, pIter);
  }

L
Liu Jicong 已提交
3939
  if (length) {
wmmhello's avatar
wmmhello 已提交
3940 3941 3942 3943 3944
    *length = offset;
  }
  return;
}

dengyihao's avatar
dengyihao 已提交
3945 3946
bool aggDecodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char* result,
                        int32_t length) {
L
Liu Jicong 已提交
3947
  if (!result || length <= 0) {
wmmhello's avatar
wmmhello 已提交
3948 3949 3950 3951 3952 3953 3954
    return false;
  }

  //  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
  int32_t count = *(int32_t*)(result);

  int32_t offset = sizeof(int32_t);
L
Liu Jicong 已提交
3955
  while (count-- > 0 && length > offset) {
wmmhello's avatar
wmmhello 已提交
3956 3957 3958
    int32_t keyLen = *(int32_t*)(result + offset);
    offset += sizeof(int32_t);

L
Liu Jicong 已提交
3959 3960 3961
    uint64_t    tableGroupId = *(uint64_t*)(result + offset);
    SResultRow* resultRow = getNewResultRow_rv(pSup->pResultBuf, tableGroupId, pSup->resultRowSize);
    if (!resultRow) {
3962
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3963
    }
3964

wmmhello's avatar
wmmhello 已提交
3965
    // add a new result set for a new group
3966 3967
    SResultRowPosition pos = {.pageId = resultRow->pageId, .offset = resultRow->offset};
    taosHashPut(pSup->pResultRowHashTable, result + offset, keyLen, &pos, sizeof(SResultRowPosition));
wmmhello's avatar
wmmhello 已提交
3968 3969 3970

    offset += keyLen;
    int32_t valueLen = *(int32_t*)(result + offset);
L
Liu Jicong 已提交
3971
    if (valueLen != pSup->resultRowSize) {
3972
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3973 3974 3975 3976 3977 3978 3979 3980 3981 3982
    }
    offset += sizeof(int32_t);
    int32_t pageId = resultRow->pageId;
    int32_t pOffset = resultRow->offset;
    memcpy(resultRow, result + offset, valueLen);
    resultRow->pageId = pageId;
    resultRow->offset = pOffset;
    offset += valueLen;

    initResultRow(resultRow);
3983
    prepareResultListBuffer(&pInfo->resultRowInfo, pOperator->pTaskInfo->env);
dengyihao's avatar
dengyihao 已提交
3984
    //    pInfo->resultRowInfo.cur = pInfo->resultRowInfo.size;
L
Liu Jicong 已提交
3985 3986
    //    pInfo->resultRowInfo.pPosition[pInfo->resultRowInfo.size++] =
    //        (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
dengyihao's avatar
dengyihao 已提交
3987
    pInfo->resultRowInfo.cur = (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
wmmhello's avatar
wmmhello 已提交
3988 3989
  }

L
Liu Jicong 已提交
3990
  if (offset != length) {
3991
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3992 3993 3994 3995
  }
  return true;
}

3996 3997
enum {
  PROJECT_RETRIEVE_CONTINUE = 0x1,
L
Liu Jicong 已提交
3998
  PROJECT_RETRIEVE_DONE = 0x2,
3999 4000 4001 4002 4003
};

static int32_t handleLimitOffset(SOperatorInfo* pOperator, SSDataBlock* pBlock) {
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
L
Liu Jicong 已提交
4004
  SSDataBlock*          pRes = pInfo->pRes;
4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052

  if (pProjectInfo->curSOffset > 0) {
    if (pProjectInfo->groupId == 0) {  // it is the first group
      pProjectInfo->groupId = pBlock->info.groupId;
      blockDataCleanup(pInfo->pRes);
      return PROJECT_RETRIEVE_CONTINUE;
    } else if (pProjectInfo->groupId != pBlock->info.groupId) {
      pProjectInfo->curSOffset -= 1;

      // ignore data block in current group
      if (pProjectInfo->curSOffset > 0) {
        blockDataCleanup(pInfo->pRes);
        return PROJECT_RETRIEVE_CONTINUE;
      }
    }

    // set current group id of the project operator
    pProjectInfo->groupId = pBlock->info.groupId;
  }

  if (pProjectInfo->groupId != 0 && pProjectInfo->groupId != pBlock->info.groupId) {
    pProjectInfo->curGroupOutput += 1;
    if ((pProjectInfo->slimit.limit > 0) && (pProjectInfo->slimit.limit <= pProjectInfo->curGroupOutput)) {
      pOperator->status = OP_EXEC_DONE;
      blockDataCleanup(pRes);

      return PROJECT_RETRIEVE_DONE;
    }

    // reset the value for a new group data
    pProjectInfo->curOffset = 0;
    pProjectInfo->curOutput = 0;
  }

  // here we reach the start position, according to the limit/offset requirements.

  // set current group id
  pProjectInfo->groupId = pBlock->info.groupId;

  if (pProjectInfo->curOffset >= pRes->info.rows) {
    pProjectInfo->curOffset -= pRes->info.rows;
    blockDataCleanup(pRes);
    return PROJECT_RETRIEVE_CONTINUE;
  } else if (pProjectInfo->curOffset < pRes->info.rows && pProjectInfo->curOffset > 0) {
    blockDataTrimFirstNRows(pRes, pProjectInfo->curOffset);
    pProjectInfo->curOffset = 0;
  }

4053 4054 4055
  // check for the limitation in each group
  if (pProjectInfo->limit.limit > 0 && pProjectInfo->curOutput + pRes->info.rows >= pProjectInfo->limit.limit) {
    pRes->info.rows = (int32_t)(pProjectInfo->limit.limit - pProjectInfo->curOutput);
4056 4057 4058 4059 4060

    if (pProjectInfo->slimit.limit == -1 || pProjectInfo->slimit.limit <= pProjectInfo->curGroupOutput) {
      pOperator->status = OP_EXEC_DONE;
    }

4061
    return PROJECT_RETRIEVE_DONE;
4062
  }
4063

4064
  // todo optimize performance
4065 4066
  // If there are slimit/soffset value exists, multi-round result can not be packed into one group, since the
  // they may not belong to the same group the limit/offset value is not valid in this case.
L
Liu Jicong 已提交
4067 4068
  if (pRes->info.rows >= pOperator->resultInfo.threshold || pProjectInfo->slimit.offset != -1 ||
      pProjectInfo->slimit.limit != -1) {
4069
    return PROJECT_RETRIEVE_DONE;
L
Liu Jicong 已提交
4070
  } else {  // not full enough, continue to accumulate the output data in the buffer.
4071 4072 4073 4074
    return PROJECT_RETRIEVE_CONTINUE;
  }
}

4075
static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
4076
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
L
Liu Jicong 已提交
4077
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
4078 4079

  SSDataBlock* pRes = pInfo->pRes;
4080
  blockDataCleanup(pRes);
4081 4082 4083 4084

  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }
dengyihao's avatar
dengyihao 已提交
4085

H
Haojun Liao 已提交
4086
#if 0
4087 4088 4089 4090 4091 4092
  if (pProjectInfo->existDataBlock) {  // TODO refactor
    SSDataBlock* pBlock = pProjectInfo->existDataBlock;
    pProjectInfo->existDataBlock = NULL;
    *newgroup = true;

    // todo dynamic set tags
L
Liu Jicong 已提交
4093 4094 4095
    //    if (pTableQueryInfo != NULL) {
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfOutput);
    //    }
4096 4097

    // the pDataBlock are always the same one, no need to call this again
H
Haojun Liao 已提交
4098
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, TSDB_ORDER_ASC);
4099

H
Haojun Liao 已提交
4100
    blockDataEnsureCapacity(pInfo->pRes, pBlock->info.rows);
4101
    projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfOutput);
L
Liu Jicong 已提交
4102
    if (pRes->info.rows >= pProjectInfo->binfo.capacity * 0.8) {
4103 4104 4105 4106 4107
      copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfOutput);
      resetResultRowEntryResult(pInfo->pCtx, pOperator->numOfOutput);
      return pRes;
    }
  }
H
Haojun Liao 已提交
4108
#endif
4109

H
Haojun Liao 已提交
4110 4111
  SOperatorInfo* downstream = pOperator->pDownstream[0];

L
Liu Jicong 已提交
4112
  while (1) {
H
Haojun Liao 已提交
4113
    // The downstream exec may change the value of the newgroup, so use a local variable instead.
H
Haojun Liao 已提交
4114
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4115
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
4116
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4117 4118

    if (pBlock == NULL) {
4119
      setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
4120 4121 4122 4123
      break;
    }

    // Return result of the previous group in the firstly.
4124
    if (false) {
4125 4126 4127
      if (pRes->info.rows > 0) {
        pProjectInfo->existDataBlock = pBlock;
        break;
L
Liu Jicong 已提交
4128
      } else {  // init output buffer for a new group data
4129 4130 4131 4132 4133
        initCtxOutputBuffer(pInfo->pCtx, pOperator->numOfOutput);
      }
    }

    // todo dynamic set tags
H
Haojun Liao 已提交
4134 4135 4136 4137
    //    STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
    //    if (pTableQueryInfo != NULL) {
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfOutput);
    //    }
4138 4139

    // the pDataBlock are always the same one, no need to call this again
4140 4141 4142
    int32_t order = getTableScanOrder(pOperator->pDownstream[0]);

    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, false);
4143 4144
    blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);

dengyihao's avatar
dengyihao 已提交
4145 4146
    projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfOutput,
                          pProjectInfo->pPseudoColInfo);
H
Haojun Liao 已提交
4147

4148 4149
    int32_t status = handleLimitOffset(pOperator, pBlock);
    if (status == PROJECT_RETRIEVE_CONTINUE) {
H
Haojun Liao 已提交
4150
      continue;
L
Liu Jicong 已提交
4151
    } else if (status == PROJECT_RETRIEVE_DONE) {
4152 4153 4154
      break;
    }
  }
dengyihao's avatar
dengyihao 已提交
4155

H
Haojun Liao 已提交
4156
  pProjectInfo->curOutput += pInfo->pRes->info.rows;
H
Haojun Liao 已提交
4157

L
Liu Jicong 已提交
4158 4159
  //  copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfOutput);
  return (pInfo->pRes->info.rows > 0) ? pInfo->pRes : NULL;
4160 4161
}

L
Liu Jicong 已提交
4162 4163
static void doHandleRemainBlockForNewGroupImpl(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                               SExecTaskInfo* pTaskInfo) {
4164
  pInfo->totalInputRows = pInfo->existNewGroupBlock->info.rows;
H
Haojun Liao 已提交
4165

L
Liu Jicong 已提交
4166 4167
  int64_t ekey = Q_STATUS_EQUAL(pTaskInfo->status, TASK_COMPLETED) ? pTaskInfo->window.ekey
                                                                   : pInfo->existNewGroupBlock->info.window.ekey;
4168 4169
  taosResetFillInfo(pInfo->pFillInfo, getFillInfoStart(pInfo->pFillInfo));

4170
  taosFillSetStartInfo(pInfo->pFillInfo, pInfo->existNewGroupBlock->info.rows, ekey);
4171 4172
  taosFillSetInputDataBlock(pInfo->pFillInfo, pInfo->existNewGroupBlock);

4173
  doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, pResultInfo->capacity);
4174 4175 4176 4177
  pInfo->existNewGroupBlock = NULL;
  *newgroup = true;
}

L
Liu Jicong 已提交
4178 4179
static void doHandleRemainBlockFromNewGroup(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                            SExecTaskInfo* pTaskInfo) {
4180 4181
  if (taosFillHasMoreResults(pInfo->pFillInfo)) {
    *newgroup = false;
4182
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, (int32_t)pResultInfo->capacity);
H
Haojun Liao 已提交
4183
    if (pInfo->pRes->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult)) {
4184 4185 4186 4187 4188 4189
      return;
    }
  }

  // handle the cached new group data block
  if (pInfo->existNewGroupBlock) {
4190
    doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4191 4192 4193
  }
}

4194
static SSDataBlock* doFill(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
4195 4196
  SFillOperatorInfo* pInfo = pOperator->info;
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
4197

H
Haojun Liao 已提交
4198
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4199 4200 4201
  SSDataBlock* pResBlock = pInfo->pRes;

  blockDataCleanup(pResBlock);
4202 4203 4204 4205
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }

4206 4207 4208
  // todo handle different group data interpolation
  bool n = false;
  bool *newgroup = &n;
4209
  doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4210 4211
  if (pResBlock->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult && pResBlock->info.rows > 0)) {
    return pResBlock;
H
Haojun Liao 已提交
4212
  }
4213

H
Haojun Liao 已提交
4214
  SOperatorInfo* pDownstream = pOperator->pDownstream[0];
L
Liu Jicong 已提交
4215
  while (1) {
H
Haojun Liao 已提交
4216
    publishOperatorProfEvent(pDownstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4217
    SSDataBlock* pBlock = pDownstream->fpSet.getNextFn(pDownstream);
H
Haojun Liao 已提交
4218
    publishOperatorProfEvent(pDownstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229

    if (*newgroup) {
      assert(pBlock != NULL);
    }

    if (*newgroup && pInfo->totalInputRows > 0) {  // there are already processed current group data block
      pInfo->existNewGroupBlock = pBlock;
      *newgroup = false;

      // Fill the previous group data block, before handle the data block of new group.
      // Close the fill operation for previous group data block
4230
      taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4231 4232 4233 4234 4235 4236 4237
    } else {
      if (pBlock == NULL) {
        if (pInfo->totalInputRows == 0) {
          pOperator->status = OP_EXEC_DONE;
          return NULL;
        }

4238
        taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4239 4240 4241 4242 4243 4244 4245
      } else {
        pInfo->totalInputRows += pBlock->info.rows;
        taosFillSetStartInfo(pInfo->pFillInfo, pBlock->info.rows, pBlock->info.window.ekey);
        taosFillSetInputDataBlock(pInfo->pFillInfo, pBlock);
      }
    }

4246 4247
    blockDataEnsureCapacity(pResBlock, pOperator->resultInfo.capacity);
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pResBlock, pOperator->resultInfo.capacity);
4248 4249

    // current group has no more result to return
4250
    if (pResBlock->info.rows > 0) {
4251 4252
      // 1. The result in current group not reach the threshold of output result, continue
      // 2. If multiple group results existing in one SSDataBlock is not allowed, return immediately
4253 4254
      if (pResBlock->info.rows > pResultInfo->threshold || pBlock == NULL || (!pInfo->multigroupResult)) {
        return pResBlock;
4255 4256
      }

4257
      doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4258 4259
      if (pResBlock->info.rows > pOperator->resultInfo.threshold || pBlock == NULL) {
        return pResBlock;
4260 4261 4262
      }
    } else if (pInfo->existNewGroupBlock) {  // try next group
      assert(pBlock != NULL);
4263
      doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4264 4265
      if (pResBlock->info.rows > pResultInfo->threshold) {
        return pResBlock;
4266 4267 4268 4269 4270 4271 4272 4273
      }
    } else {
      return NULL;
    }
  }
}

// todo set the attribute of query scan count
H
Haojun Liao 已提交
4274
static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr) {
L
Liu Jicong 已提交
4275
  for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289
    int32_t functionId = getExprFunctionId(&pQueryAttr->pExpr1[i]);
    if (functionId == FUNCTION_STDDEV || functionId == FUNCTION_PERCT) {
      return 2;
    }
  }

  return 1;
}

static void destroyOperatorInfo(SOperatorInfo* pOperator) {
  if (pOperator == NULL) {
    return;
  }

4290 4291
  if (pOperator->fpSet.closeFn != NULL) {
    pOperator->fpSet.closeFn(pOperator->info, pOperator->numOfOutput);
4292 4293
  }

H
Haojun Liao 已提交
4294
  if (pOperator->pDownstream != NULL) {
L
Liu Jicong 已提交
4295
    for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
H
Haojun Liao 已提交
4296
      destroyOperatorInfo(pOperator->pDownstream[i]);
4297 4298
    }

wafwerar's avatar
wafwerar 已提交
4299
    taosMemoryFreeClear(pOperator->pDownstream);
H
Haojun Liao 已提交
4300
    pOperator->numOfDownstream = 0;
4301 4302
  }

wafwerar's avatar
wafwerar 已提交
4303 4304
  taosMemoryFreeClear(pOperator->info);
  taosMemoryFreeClear(pOperator);
4305 4306
}

dengyihao's avatar
dengyihao 已提交
4307 4308
int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                         const char* pKey) {
4309 4310
  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);

dengyihao's avatar
dengyihao 已提交
4311 4312
  pAggSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
  pAggSup->keyBuf = taosMemoryCalloc(1, keyBufSize + POINTER_BYTES + sizeof(int64_t));
4313
  pAggSup->pResultRowHashTable = taosHashInit(10, hashFn, true, HASH_NO_LOCK);
L
Liu Jicong 已提交
4314 4315
  //  pAggSup->pResultRowListSet = taosHashInit(100, hashFn, false, HASH_NO_LOCK);
  //  pAggSup->pResultRowArrayList = taosArrayInit(10, sizeof(SResultRowCell));
4316

4317
  if (pAggSup->keyBuf == NULL /*|| pAggSup->pResultRowArrayList == NULL || pAggSup->pResultRowListSet == NULL*/ ||
4318
      pAggSup->pResultRowHashTable == NULL) {
4319 4320 4321
    return TSDB_CODE_OUT_OF_MEMORY;
  }

H
Haojun Liao 已提交
4322 4323 4324 4325 4326
  int32_t code = createDiskbasedBuf(&pAggSup->pResultBuf, 4096, 4096 * 256, pKey, "/tmp/");
  if (code != TSDB_CODE_SUCCESS) {
    return code;
  }

4327 4328 4329
  return TSDB_CODE_SUCCESS;
}

4330
void cleanupAggSup(SAggSupporter* pAggSup) {
wafwerar's avatar
wafwerar 已提交
4331
  taosMemoryFreeClear(pAggSup->keyBuf);
4332
  taosHashCleanup(pAggSup->pResultRowHashTable);
H
Haojun Liao 已提交
4333
  destroyDiskbasedBuf(pAggSup->pResultBuf);
4334 4335
}

H
Haojun Liao 已提交
4336
int32_t initAggInfo(SOptrBasicInfo* pBasicInfo, SAggSupporter* pAggSup, SExprInfo* pExprInfo, int32_t numOfCols,
4337
                    SSDataBlock* pResultBlock, size_t keyBufSize, const char* pkey) {
4338
  pBasicInfo->pCtx = createSqlFunctionCtx(pExprInfo, numOfCols, &pBasicInfo->rowCellInfoOffset);
H
Haojun Liao 已提交
4339 4340
  pBasicInfo->pRes = pResultBlock;

4341
  doInitAggInfoSup(pAggSup, pBasicInfo->pCtx, numOfCols, keyBufSize, pkey);
4342

L
Liu Jicong 已提交
4343
  for (int32_t i = 0; i < numOfCols; ++i) {
4344 4345 4346
    pBasicInfo->pCtx[i].pBuf = pAggSup->pResultBuf;
  }

4347
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
4348 4349
}

4350 4351 4352 4353 4354 4355 4356 4357 4358
void initResultSizeInfo(SOperatorInfo* pOperator, int32_t numOfRows) {
  pOperator->resultInfo.capacity = numOfRows;
  pOperator->resultInfo.threshold = numOfRows * 0.75;

  if (pOperator->resultInfo.threshold == 0) {
    pOperator->resultInfo.capacity = numOfRows;
  }
}

H
Haojun Liao 已提交
4359
static STableQueryInfo* initTableQueryInfo(const STableGroupInfo* pTableGroupInfo) {
L
Liu Jicong 已提交
4360 4361 4362 4363
  if (pTableGroupInfo->numOfTables == 0) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
4364
  STableQueryInfo* pTableQueryInfo = taosMemoryCalloc(pTableGroupInfo->numOfTables, sizeof(STableQueryInfo));
H
Haojun Liao 已提交
4365 4366
  if (pTableQueryInfo == NULL) {
    return NULL;
H
Haojun Liao 已提交
4367
  }
H
Haojun Liao 已提交
4368 4369

  int32_t index = 0;
L
Liu Jicong 已提交
4370
  for (int32_t i = 0; i < taosArrayGetSize(pTableGroupInfo->pGroupList); ++i) {
H
Haojun Liao 已提交
4371
    SArray* pa = taosArrayGetP(pTableGroupInfo->pGroupList, i);
L
Liu Jicong 已提交
4372
    for (int32_t j = 0; j < taosArrayGetSize(pa); ++j) {
H
Haojun Liao 已提交
4373 4374
      STableKeyInfo* pk = taosArrayGet(pa, j);

H
Haojun Liao 已提交
4375
      STableQueryInfo* pTQueryInfo = &pTableQueryInfo[index++];
dengyihao's avatar
dengyihao 已提交
4376
      //      pTQueryInfo->uid = pk->uid;
L
Liu Jicong 已提交
4377
      pTQueryInfo->lastKey = pk->lastKey;
dengyihao's avatar
dengyihao 已提交
4378
      //      pTQueryInfo->groupIndex = i;
H
Haojun Liao 已提交
4379 4380
    }
  }
H
Haojun Liao 已提交
4381 4382

  STimeWindow win = {0, INT64_MAX};
4383
  createTableQueryInfo(pTableQueryInfo, win);
H
Haojun Liao 已提交
4384
  return pTableQueryInfo;
H
Haojun Liao 已提交
4385 4386
}

L
Liu Jicong 已提交
4387
SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
4388
                                           SSDataBlock* pResultBlock, SExprInfo* pScalarExprInfo,
L
Liu Jicong 已提交
4389 4390
                                           int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo,
                                           const STableGroupInfo* pTableGroupInfo) {
wafwerar's avatar
wafwerar 已提交
4391
  SAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SAggOperatorInfo));
L
Liu Jicong 已提交
4392
  SOperatorInfo*    pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4393 4394 4395
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
4396

H
Haojun Liao 已提交
4397
  int32_t numOfRows = 1;
dengyihao's avatar
dengyihao 已提交
4398
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4399 4400

  initResultSizeInfo(pOperator, numOfRows);
dengyihao's avatar
dengyihao 已提交
4401 4402
  int32_t code =
      initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResultBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4403
  pInfo->pTableQueryInfo = initTableQueryInfo(pTableGroupInfo);
L
Liu Jicong 已提交
4404
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4405 4406
    goto _error;
  }
H
Haojun Liao 已提交
4407

H
Haojun Liao 已提交
4408 4409 4410 4411 4412 4413 4414
  pOperator->resultInfo.capacity = 4096;
  pOperator->resultInfo.threshold = 4096 * 0.75;

  int32_t numOfGroup = 10;  // todo replaced with true value
  pInfo->groupId = INT32_MIN;
  initResultRowInfo(&pInfo->binfo.resultRowInfo, numOfGroup);

4415 4416
  pInfo->pScalarExprInfo = pScalarExprInfo;
  pInfo->numOfScalarExpr = numOfScalarExpr;
4417 4418 4419
  if (pInfo->pScalarExprInfo != NULL) {
    pInfo->pScalarCtx = createSqlFunctionCtx(pScalarExprInfo, numOfCols, &pInfo->rowCellInfoOffset);
  }
4420

dengyihao's avatar
dengyihao 已提交
4421
  pOperator->name = "TableAggregate";
X
Xiaoyu Wang 已提交
4422
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_AGG;
4423
  pOperator->blocking = true;
dengyihao's avatar
dengyihao 已提交
4424 4425 4426 4427 4428
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
  pOperator->numOfOutput = numOfCols;
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
4429

4430 4431
  pOperator->fpSet = createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, NULL, destroyAggOperatorInfo,
                                         aggEncodeResultRow, aggDecodeResultRow, NULL);
H
Haojun Liao 已提交
4432 4433 4434 4435 4436

  code = appendDownstream(pOperator, &downstream, 1);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4437 4438

  return pOperator;
L
Liu Jicong 已提交
4439
_error:
H
Haojun Liao 已提交
4440
  destroyAggOperatorInfo(pInfo, numOfCols);
wafwerar's avatar
wafwerar 已提交
4441 4442
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
4443 4444
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4445 4446
}

H
Haojun Liao 已提交
4447
void doDestroyBasicInfo(SOptrBasicInfo* pInfo, int32_t numOfOutput) {
4448 4449
  assert(pInfo != NULL);

4450
  destroySqlFunctionCtx(pInfo->pCtx, numOfOutput);
wafwerar's avatar
wafwerar 已提交
4451
  taosMemoryFreeClear(pInfo->rowCellInfoOffset);
4452 4453

  cleanupResultRowInfo(&pInfo->resultRowInfo);
H
Haojun Liao 已提交
4454
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
4455 4456
}

H
Haojun Liao 已提交
4457
void destroyBasicOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4458
  SOptrBasicInfo* pInfo = (SOptrBasicInfo*)param;
4459 4460
  doDestroyBasicInfo(pInfo, numOfOutput);
}
H
Haojun Liao 已提交
4461 4462

void destroyAggOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4463
  SAggOperatorInfo* pInfo = (SAggOperatorInfo*)param;
4464 4465
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
}
4466

H
Haojun Liao 已提交
4467
void destroySFillOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4468
  SFillOperatorInfo* pInfo = (SFillOperatorInfo*)param;
4469
  pInfo->pFillInfo = taosDestroyFillInfo(pInfo->pFillInfo);
H
Haojun Liao 已提交
4470
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
wafwerar's avatar
wafwerar 已提交
4471
  taosMemoryFreeClear(pInfo->p);
4472 4473
}

H
Haojun Liao 已提交
4474
static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4475
  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
4476 4477 4478
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
}

H
Haojun Liao 已提交
4479
static void destroyOrderOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4480
  SSortOperatorInfo* pInfo = (SSortOperatorInfo*)param;
H
Haojun Liao 已提交
4481 4482
  pInfo->pDataBlock = blockDataDestroy(pInfo->pDataBlock);

H
Haojun Liao 已提交
4483
  taosArrayDestroy(pInfo->pSortInfo);
4484
  taosArrayDestroy(pInfo->inputSlotMap);
4485 4486
}

H
Haojun Liao 已提交
4487
void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4488
  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
H
Haojun Liao 已提交
4489 4490 4491 4492 4493 4494 4495 4496 4497
  taosArrayDestroy(pExInfo->pSources);
  taosArrayDestroy(pExInfo->pSourceDataInfo);
  if (pExInfo->pResult != NULL) {
    blockDataDestroy(pExInfo->pResult);
  }

  tsem_destroy(&pExInfo->ready);
}

H
Haojun Liao 已提交
4498 4499
static SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
  SArray* pList = taosArrayInit(4, sizeof(int32_t));
dengyihao's avatar
dengyihao 已提交
4500
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
4501 4502 4503 4504 4505 4506 4507 4508
    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
      taosArrayPush(pList, &i);
    }
  }

  return pList;
}

L
Liu Jicong 已提交
4509
SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t num,
dengyihao's avatar
dengyihao 已提交
4510 4511
                                         SSDataBlock* pResBlock, SLimit* pLimit, SLimit* pSlimit,
                                         SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4512
  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
L
Liu Jicong 已提交
4513
  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4514 4515 4516
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
4517

dengyihao's avatar
dengyihao 已提交
4518 4519 4520
  pInfo->limit = *pLimit;
  pInfo->slimit = *pSlimit;
  pInfo->curOffset = pLimit->offset;
H
Haojun Liao 已提交
4521 4522
  pInfo->curSOffset = pSlimit->offset;

H
Haojun Liao 已提交
4523
  pInfo->binfo.pRes = pResBlock;
H
Haojun Liao 已提交
4524 4525 4526

  int32_t numOfCols = num;
  int32_t numOfRows = 4096;
dengyihao's avatar
dengyihao 已提交
4527
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4528 4529 4530

  initResultSizeInfo(pOperator, numOfRows);
  initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4531
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, pTaskInfo);
H
Haojun Liao 已提交
4532
  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pInfo->binfo.pCtx, numOfCols);
4533

dengyihao's avatar
dengyihao 已提交
4534
  pOperator->name = "ProjectOperator";
H
Haojun Liao 已提交
4535
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
4536
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4537 4538 4539 4540
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
  pOperator->numOfOutput = num;
4541

L
Liu Jicong 已提交
4542 4543
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
                                         destroyProjectOperatorInfo, NULL, NULL, NULL);
L
Liu Jicong 已提交
4544 4545

  pOperator->pTaskInfo = pTaskInfo;
4546
  int32_t code = appendDownstream(pOperator, &downstream, 1);
H
Haojun Liao 已提交
4547
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4548 4549
    goto _error;
  }
4550 4551

  return pOperator;
H
Haojun Liao 已提交
4552

L
Liu Jicong 已提交
4553
_error:
H
Haojun Liao 已提交
4554 4555
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4556 4557
}

4558
static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SNodeListNode* pValNode,
L
Liu Jicong 已提交
4559
                            STimeWindow win, int32_t capacity, const char* id, SInterval* pInterval, int32_t fillType) {
4560
  SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pValNode);
H
Haojun Liao 已提交
4561 4562

  STimeWindow w = TSWINDOW_INITIALIZER;
4563
  getAlignQueryTimeWindow(pInterval, pInterval->precision, win.skey, &w);
H
Haojun Liao 已提交
4564 4565

  int32_t order = TSDB_ORDER_ASC;
4566
  pInfo->pFillInfo = taosCreateFillInfo(order, w.skey, 0, capacity, numOfCols, pInterval, fillType, pColInfo, id);
H
Haojun Liao 已提交
4567

wafwerar's avatar
wafwerar 已提交
4568
  pInfo->p = taosMemoryCalloc(numOfCols, POINTER_BYTES);
H
Haojun Liao 已提交
4569 4570 4571 4572 4573 4574 4575
  if (pInfo->pFillInfo == NULL || pInfo->p == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  } else {
    return TSDB_CODE_SUCCESS;
  }
}

L
Liu Jicong 已提交
4576
SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExpr, int32_t numOfCols,
4577
                                      SInterval* pInterval, STimeWindow* pWindow, SSDataBlock* pResBlock, int32_t fillType, SNodeListNode* pValueNode,
L
Liu Jicong 已提交
4578
                                      bool multigroupResult, SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4579
  SFillOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SFillOperatorInfo));
L
Liu Jicong 已提交
4580
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4581

L
Liu Jicong 已提交
4582
  pInfo->pRes = pResBlock;
4583 4584
  pInfo->multigroupResult = multigroupResult;

4585 4586
  int32_t type = TSDB_FILL_NONE;
  switch (fillType) {
dengyihao's avatar
dengyihao 已提交
4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604
    case FILL_MODE_PREV:
      type = TSDB_FILL_PREV;
      break;
    case FILL_MODE_NONE:
      type = TSDB_FILL_NONE;
      break;
    case FILL_MODE_NULL:
      type = TSDB_FILL_NULL;
      break;
    case FILL_MODE_NEXT:
      type = TSDB_FILL_NEXT;
      break;
    case FILL_MODE_VALUE:
      type = TSDB_FILL_SET_VALUE;
      break;
    case FILL_MODE_LINEAR:
      type = TSDB_FILL_LINEAR;
      break;
4605 4606 4607 4608
    default:
      type = TSDB_FILL_NONE;
  }

H
Haojun Liao 已提交
4609
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4610 4611
  initResultSizeInfo(pOperator, 4096);

4612
  int32_t code = initFillInfo(pInfo, pExpr, numOfCols, pValueNode, *pWindow, pResultInfo->capacity,
4613
                              pTaskInfo->id.str, pInterval, type);
4614 4615 4616
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4617

dengyihao's avatar
dengyihao 已提交
4618
  pOperator->name = "FillOperator";
4619
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4620
  pOperator->status = OP_NOT_OPENED;
4621
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_FILL;
dengyihao's avatar
dengyihao 已提交
4622 4623 4624
  pOperator->pExpr = pExpr;
  pOperator->numOfOutput = numOfCols;
  pOperator->info = pInfo;
H
Haojun Liao 已提交
4625

L
Liu Jicong 已提交
4626 4627
  pOperator->fpSet =
      createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, NULL, destroySFillOperatorInfo, NULL, NULL, NULL);
4628
  pOperator->pTaskInfo = pTaskInfo;
4629
  code = appendDownstream(pOperator, &downstream, 1);
4630
  return pOperator;
H
Haojun Liao 已提交
4631

L
Liu Jicong 已提交
4632
_error:
wafwerar's avatar
wafwerar 已提交
4633 4634
  taosMemoryFreeClear(pOperator);
  taosMemoryFreeClear(pInfo);
H
Haojun Liao 已提交
4635
  return NULL;
4636 4637
}

L
Liu Jicong 已提交
4638
static int32_t getColumnIndexInSource(SQueriedTableInfo* pTableInfo, SExprBasicInfo* pExpr, SColumnInfo* pTagCols) {
4639 4640
  int32_t j = 0;

H
Haojun Liao 已提交
4641 4642
  if (TSDB_COL_IS_TAG(pExpr->pParam[0].pCol->type)) {
    if (pExpr->pParam[0].pCol->colId == TSDB_TBNAME_COLUMN_INDEX) {
4643 4644 4645
      return TSDB_TBNAME_COLUMN_INDEX;
    }

L
Liu Jicong 已提交
4646
    while (j < pTableInfo->numOfTags) {
H
Haojun Liao 已提交
4647
      if (pExpr->pParam[0].pCol->colId == pTagCols[j].colId) {
4648 4649 4650 4651 4652 4653
        return j;
      }

      j += 1;
    }

4654
  } /*else if (TSDB_COL_IS_UD_COL(pExpr->colInfo.flag)) {  // user specified column data
4655 4656 4657 4658 4659 4660 4661 4662 4663
    return TSDB_UD_COLUMN_INDEX;
  } else {
    while (j < pTableInfo->numOfCols) {
      if (pExpr->colInfo.colId == pTableInfo->colList[j].colId) {
        return j;
      }

      j += 1;
    }
4664
  }*/
4665 4666 4667 4668

  return INT32_MIN;  // return a less than TSDB_TBNAME_COLUMN_INDEX value
}

L
Liu Jicong 已提交
4669
bool validateExprColumnInfo(SQueriedTableInfo* pTableInfo, SExprBasicInfo* pExpr, SColumnInfo* pTagCols) {
4670 4671 4672 4673
  int32_t j = getColumnIndexInSource(pTableInfo, pExpr, pTagCols);
  return j != INT32_MIN;
}

L
Liu Jicong 已提交
4674 4675
static SResSchema createResSchema(int32_t type, int32_t bytes, int32_t slotId, int32_t scale, int32_t precision,
                                  const char* name) {
H
Haojun Liao 已提交
4676
  SResSchema s = {0};
dengyihao's avatar
dengyihao 已提交
4677 4678 4679 4680
  s.scale = scale;
  s.type = type;
  s.bytes = bytes;
  s.slotId = slotId;
H
Haojun Liao 已提交
4681
  s.precision = precision;
H
Haojun Liao 已提交
4682 4683 4684 4685
  strncpy(s.name, name, tListLen(s.name));

  return s;
}
H
Haojun Liao 已提交
4686

H
Haojun Liao 已提交
4687 4688 4689 4690 4691 4692 4693
static SColumn* createColumn(int32_t blockId, int32_t slotId, SDataType* pType) {
  SColumn* pCol = taosMemoryCalloc(1, sizeof(SColumn));
  if (pCol == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

dengyihao's avatar
dengyihao 已提交
4694 4695 4696 4697 4698
  pCol->slotId = slotId;
  pCol->bytes = pType->bytes;
  pCol->type = pType->type;
  pCol->scale = pType->scale;
  pCol->precision = pType->precision;
H
Haojun Liao 已提交
4699 4700 4701 4702 4703
  pCol->dataBlockId = blockId;

  return pCol;
}

H
Haojun Liao 已提交
4704
SExprInfo* createExprInfo(SNodeList* pNodeList, SNodeList* pGroupKeys, int32_t* numOfExprs) {
H
Haojun Liao 已提交
4705
  int32_t numOfFuncs = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
4706 4707 4708 4709
  int32_t numOfGroupKeys = 0;
  if (pGroupKeys != NULL) {
    numOfGroupKeys = LIST_LENGTH(pGroupKeys);
  }
H
Haojun Liao 已提交
4710

H
Haojun Liao 已提交
4711
  *numOfExprs = numOfFuncs + numOfGroupKeys;
wafwerar's avatar
wafwerar 已提交
4712
  SExprInfo* pExprs = taosMemoryCalloc(*numOfExprs, sizeof(SExprInfo));
H
Haojun Liao 已提交
4713

L
Liu Jicong 已提交
4714
  for (int32_t i = 0; i < (*numOfExprs); ++i) {
H
Haojun Liao 已提交
4715 4716 4717 4718 4719 4720
    STargetNode* pTargetNode = NULL;
    if (i < numOfFuncs) {
      pTargetNode = (STargetNode*)nodesListGetNode(pNodeList, i);
    } else {
      pTargetNode = (STargetNode*)nodesListGetNode(pGroupKeys, i - numOfFuncs);
    }
H
Haojun Liao 已提交
4721

4722
    SExprInfo* pExp = &pExprs[i];
H
Haojun Liao 已提交
4723

wafwerar's avatar
wafwerar 已提交
4724
    pExp->pExpr = taosMemoryCalloc(1, sizeof(tExprNode));
H
Haojun Liao 已提交
4725
    pExp->pExpr->_function.num = 1;
H
Haojun Liao 已提交
4726
    pExp->pExpr->_function.functionId = -1;
H
Haojun Liao 已提交
4727

4728
    int32_t type = nodeType(pTargetNode->pExpr);
H
Haojun Liao 已提交
4729
    // it is a project query, or group by column
4730
    if (type == QUERY_NODE_COLUMN) {
H
Haojun Liao 已提交
4731
      pExp->pExpr->nodeType = QUERY_NODE_COLUMN;
L
Liu Jicong 已提交
4732
      SColumnNode* pColNode = (SColumnNode*)pTargetNode->pExpr;
H
Haojun Liao 已提交
4733

G
Ganlin Zhao 已提交
4734 4735 4736
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

H
Haojun Liao 已提交
4737
      SDataType* pType = &pColNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4738 4739
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pColNode->colName);
H
Haojun Liao 已提交
4740 4741
      pExp->base.pParam[0].pCol = createColumn(pColNode->dataBlockId, pColNode->slotId, pType);
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_COLUMN;
4742
    } else if (type == QUERY_NODE_VALUE) {
4743 4744 4745 4746 4747 4748 4749
      pExp->pExpr->nodeType = QUERY_NODE_VALUE;
      SValueNode* pValNode = (SValueNode*)pTargetNode->pExpr;

      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

      SDataType* pType = &pValNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4750 4751
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pValNode->node.aliasName);
4752 4753
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_VALUE;
      valueNodeToVariant(pValNode, &pExp->base.pParam[0].param);
4754
    } else if (type == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
4755
      pExp->pExpr->nodeType = QUERY_NODE_FUNCTION;
H
Haojun Liao 已提交
4756 4757 4758
      SFunctionNode* pFuncNode = (SFunctionNode*)pTargetNode->pExpr;

      SDataType* pType = &pFuncNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4759 4760
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pFuncNode->node.aliasName);
H
Haojun Liao 已提交
4761

H
Haojun Liao 已提交
4762
      pExp->pExpr->_function.functionId = pFuncNode->funcId;
H
Haojun Liao 已提交
4763
      pExp->pExpr->_function.pFunctNode = pFuncNode;
dengyihao's avatar
dengyihao 已提交
4764 4765
      strncpy(pExp->pExpr->_function.functionName, pFuncNode->functionName,
              tListLen(pExp->pExpr->_function.functionName));
H
Haojun Liao 已提交
4766 4767

      int32_t numOfParam = LIST_LENGTH(pFuncNode->pParameterList);
G
Ganlin Zhao 已提交
4768 4769 4770 4771

      pExp->base.pParam = taosMemoryCalloc(numOfParam, sizeof(SFunctParam));
      pExp->base.numOfParams = numOfParam;

H
Haojun Liao 已提交
4772
      for (int32_t j = 0; j < numOfParam; ++j) {
4773
        SNode* p1 = nodesListGetNode(pFuncNode->pParameterList, j);
G
Ganlin Zhao 已提交
4774
        if (p1->type == QUERY_NODE_COLUMN) {
dengyihao's avatar
dengyihao 已提交
4775
          SColumnNode* pcn = (SColumnNode*)p1;
G
Ganlin Zhao 已提交
4776 4777

          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_COLUMN;
H
Haojun Liao 已提交
4778
          pExp->base.pParam[j].pCol = createColumn(pcn->dataBlockId, pcn->slotId, &pcn->node.resType);
G
Ganlin Zhao 已提交
4779 4780 4781
        } else if (p1->type == QUERY_NODE_VALUE) {
          SValueNode* pvn = (SValueNode*)p1;
          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_VALUE;
4782
          valueNodeToVariant(pvn, &pExp->base.pParam[j].param);
G
Ganlin Zhao 已提交
4783
        }
H
Haojun Liao 已提交
4784
      }
4785
    } else if (type == QUERY_NODE_OPERATOR) {
H
Haojun Liao 已提交
4786
      pExp->pExpr->nodeType = QUERY_NODE_OPERATOR;
L
Liu Jicong 已提交
4787
      SOperatorNode* pNode = (SOperatorNode*)pTargetNode->pExpr;
4788

G
Ganlin Zhao 已提交
4789 4790 4791
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

4792
      SDataType* pType = &pNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4793 4794
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pNode->node.aliasName);
4795 4796 4797
      pExp->pExpr->_optrRoot.pRootNode = pTargetNode->pExpr;
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4798 4799 4800
    }
  }

H
Haojun Liao 已提交
4801
  return pExprs;
H
Haojun Liao 已提交
4802 4803
}

4804
static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model) {
wafwerar's avatar
wafwerar 已提交
4805
  SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
4806
  setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
4807

4808
  pTaskInfo->cost.created = taosGetTimestampMs();
H
Haojun Liao 已提交
4809
  pTaskInfo->id.queryId = queryId;
dengyihao's avatar
dengyihao 已提交
4810
  pTaskInfo->execModel = model;
H
Haojun Liao 已提交
4811

wafwerar's avatar
wafwerar 已提交
4812
  char* p = taosMemoryCalloc(1, 128);
L
Liu Jicong 已提交
4813
  snprintf(p, 128, "TID:0x%" PRIx64 " QID:0x%" PRIx64, taskId, queryId);
4814
  pTaskInfo->id.str = strdup(p);
H
Haojun Liao 已提交
4815

4816 4817
  return pTaskInfo;
}
H
Haojun Liao 已提交
4818

L
Liu Jicong 已提交
4819 4820
static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                                      STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId);
H
Haojun Liao 已提交
4821

L
Liu Jicong 已提交
4822 4823
static int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                                  uint64_t queryId, uint64_t taskId);
4824
static SArray* extractTableIdList(const STableGroupInfo* pTableGroupInfo);
H
Haojun Liao 已提交
4825
static SArray* extractColumnInfo(SNodeList* pNodeList);
H
Haojun Liao 已提交
4826
static SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols);
4827

4828
static SArray* createSortInfo(SNodeList* pNodeList, SNodeList* pNodeListTarget);
4829
static SArray* createIndexMap(SNodeList* pNodeList);
4830
static SArray* extractPartitionColInfo(SNodeList* pNodeList);
4831 4832 4833 4834
static int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode);

static SInterval extractIntervalInfo(const STableScanPhysiNode* pTableScanNode) {
  SInterval interval = {
L
Liu Jicong 已提交
4835 4836
      .interval = pTableScanNode->interval,
      .sliding = pTableScanNode->sliding,
4837
      .intervalUnit = pTableScanNode->intervalUnit,
L
Liu Jicong 已提交
4838 4839
      .slidingUnit = pTableScanNode->slidingUnit,
      .offset = pTableScanNode->offset,
4840 4841 4842 4843
  };

  return interval;
}
H
Haojun Liao 已提交
4844

H
Haojun Liao 已提交
4845
SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
4846
                                  uint64_t queryId, uint64_t taskId, STableGroupInfo* pTableGroupInfo) {
4847 4848
  int32_t type = nodeType(pPhyNode);

X
Xiaoyu Wang 已提交
4849
  if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
H
Haojun Liao 已提交
4850
    if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == type) {
dengyihao's avatar
dengyihao 已提交
4851 4852
      SScanPhysiNode*      pScanPhyNode = (SScanPhysiNode*)pPhyNode;
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4853

L
Liu Jicong 已提交
4854
      int32_t     numOfCols = 0;
H
Haojun Liao 已提交
4855
      tsdbReaderT pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableGroupInfo, (uint64_t)queryId, taskId);
4856
      if (pDataReader == NULL && terrno != 0) {
4857 4858
        return NULL;
      }
4859

dengyihao's avatar
dengyihao 已提交
4860 4861
      SArray* pColList =
          extractColMatchInfo(pScanPhyNode->pScanCols, pScanPhyNode->node.pOutputDataBlockDesc, &numOfCols);
4862
      SSDataBlock* pResBlock = createResDataBlock(pScanPhyNode->node.pOutputDataBlockDesc);
4863

4864
      SQueryTableDataCond cond = {0};
4865 4866

      int32_t code = initQueryTableDataCond(&cond, pTableScanNode);
4867 4868 4869
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
      }
4870

4871
      SInterval interval = extractIntervalInfo(pTableScanNode);
4872
      SOperatorInfo* pOperator = createTableScanOperatorInfo(pDataReader, &cond, numOfCols, pTableScanNode->dataRequired,
L
Liu Jicong 已提交
4873 4874
                                         pTableScanNode->scanSeq, pColList, pResBlock, pScanPhyNode->node.pConditions,
                                         &interval, pTableScanNode->ratio, pTaskInfo);
4875 4876 4877
      STableScanInfo* pScanInfo = pOperator->info;
      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
      return pOperator;
H
Haojun Liao 已提交
4878
    } else if (QUERY_NODE_PHYSICAL_PLAN_EXCHANGE == type) {
H
Haojun Liao 已提交
4879
      SExchangePhysiNode* pExchange = (SExchangePhysiNode*)pPhyNode;
4880
      SSDataBlock*        pResBlock = createResDataBlock(pExchange->node.pOutputDataBlockDesc);
H
Haojun Liao 已提交
4881
      return createExchangeOperatorInfo(pExchange->pSrcEndPoints, pResBlock, pTaskInfo);
H
Haojun Liao 已提交
4882
    } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN == type) {
H
Haojun Liao 已提交
4883
      SScanPhysiNode* pScanPhyNode = (SScanPhysiNode*)pPhyNode;  // simple child table.
4884

4885
      int32_t code = doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo, queryId, taskId);
H
Haojun Liao 已提交
4886
      SArray* tableIdList = extractTableIdList(pTableGroupInfo);
H
Haojun Liao 已提交
4887

4888
      SSDataBlock* pResBlock = createResDataBlock(pScanPhyNode->node.pOutputDataBlockDesc);
4889 4890

      int32_t numOfCols = 0;
4891
      SArray* pCols = extractColMatchInfo(pScanPhyNode->pScanCols, pScanPhyNode->node.pOutputDataBlockDesc, &numOfCols);
X
Xiaoyu Wang 已提交
4892 4893
      SOperatorInfo* pOperator = createStreamScanOperatorInfo(pHandle->reader, pResBlock, pCols, tableIdList, pTaskInfo,
                                                              pScanPhyNode->node.pConditions);
4894
      taosArrayDestroy(tableIdList);
H
Haojun Liao 已提交
4895
      return pOperator;
H
Haojun Liao 已提交
4896
    } else if (QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN == type) {
L
Liu Jicong 已提交
4897
      SSystemTableScanPhysiNode* pSysScanPhyNode = (SSystemTableScanPhysiNode*)pPhyNode;
L
Liu Jicong 已提交
4898
      SScanPhysiNode*            pScanNode = &pSysScanPhyNode->scan;
H
Haojun Liao 已提交
4899

4900
      SSDataBlock* pResBlock = createResDataBlock(pScanNode->node.pOutputDataBlockDesc);
4901

4902
      int32_t numOfOutputCols = 0;
L
Liu Jicong 已提交
4903 4904
      SArray* colList =
          extractColMatchInfo(pScanNode->pScanCols, pScanNode->node.pOutputDataBlockDesc, &numOfOutputCols);
L
Liu Jicong 已提交
4905
      SOperatorInfo* pOperator = createSysTableScanOperatorInfo(
L
Liu Jicong 已提交
4906 4907
          pHandle, pResBlock, &pScanNode->tableName, pScanNode->node.pConditions, pSysScanPhyNode->mgmtEpSet, colList,
          pTaskInfo, pSysScanPhyNode->showRewrite, pSysScanPhyNode->accountId);
H
Haojun Liao 已提交
4908 4909 4910
      return pOperator;
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4911 4912 4913
    }
  }

4914 4915
  int32_t num = 0;
  size_t  size = LIST_LENGTH(pPhyNode->pChildren);
H
Haojun Liao 已提交
4916

4917
  SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
dengyihao's avatar
dengyihao 已提交
4918
  for (int32_t i = 0; i < size; ++i) {
4919 4920
    SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableGroupInfo);
4921 4922 4923
    if (ops[i] == NULL) {
      return NULL;
    }
4924
  }
H
Haojun Liao 已提交
4925

4926
  SOperatorInfo* pOptr = NULL;
H
Haojun Liao 已提交
4927
  if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
dengyihao's avatar
dengyihao 已提交
4928 4929
    SProjectPhysiNode* pProjPhyNode = (SProjectPhysiNode*)pPhyNode;
    SExprInfo*         pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &num);
H
Haojun Liao 已提交
4930

4931
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
dengyihao's avatar
dengyihao 已提交
4932 4933
    SLimit       limit = {.limit = pProjPhyNode->limit, .offset = pProjPhyNode->offset};
    SLimit       slimit = {.limit = pProjPhyNode->slimit, .offset = pProjPhyNode->soffset};
4934
    pOptr = createProjectOperatorInfo(ops[0], pExprInfo, num, pResBlock, &limit, &slimit, pTaskInfo);
H
Haojun Liao 已提交
4935 4936 4937
  } else if (QUERY_NODE_PHYSICAL_PLAN_AGG == type) {
    SAggPhysiNode* pAggNode = (SAggPhysiNode*)pPhyNode;
    SExprInfo*     pExprInfo = createExprInfo(pAggNode->pAggFuncs, pAggNode->pGroupKeys, &num);
4938
    SSDataBlock*   pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4939

dengyihao's avatar
dengyihao 已提交
4940
    int32_t    numOfScalarExpr = 0;
4941 4942 4943 4944 4945
    SExprInfo* pScalarExprInfo = NULL;
    if (pAggNode->pExprs != NULL) {
      pScalarExprInfo = createExprInfo(pAggNode->pExprs, NULL, &numOfScalarExpr);
    }

H
Haojun Liao 已提交
4946 4947
    if (pAggNode->pGroupKeys != NULL) {
      SArray* pColList = extractColumnInfo(pAggNode->pGroupKeys);
dengyihao's avatar
dengyihao 已提交
4948 4949
      pOptr = createGroupOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pAggNode->node.pConditions,
                                      pScalarExprInfo, numOfScalarExpr, pTaskInfo, NULL);
H
Haojun Liao 已提交
4950
    } else {
dengyihao's avatar
dengyihao 已提交
4951 4952
      pOptr = createAggregateOperatorInfo(ops[0], pExprInfo, num, pResBlock, pScalarExprInfo, numOfScalarExpr,
                                          pTaskInfo, pTableGroupInfo);
H
Haojun Liao 已提交
4953
    }
H
Haojun Liao 已提交
4954 4955
  } else if (QUERY_NODE_PHYSICAL_PLAN_INTERVAL == type) {
    SIntervalPhysiNode* pIntervalPhyNode = (SIntervalPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4956

H
Haojun Liao 已提交
4957
    SExprInfo*   pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
4958
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4959

dengyihao's avatar
dengyihao 已提交
4960 4961 4962 4963 4964 4965
    SInterval interval = {.interval = pIntervalPhyNode->interval,
                          .sliding = pIntervalPhyNode->sliding,
                          .intervalUnit = pIntervalPhyNode->intervalUnit,
                          .slidingUnit = pIntervalPhyNode->slidingUnit,
                          .offset = pIntervalPhyNode->offset,
                          .precision = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->node.resType.precision};
H
Haojun Liao 已提交
4966

dengyihao's avatar
dengyihao 已提交
4967 4968
    STimeWindowAggSupp as = {.waterMark = pIntervalPhyNode->window.watermark,
                             .calTrigger = pIntervalPhyNode->window.triggerType};
4969

4970 4971 4972 4973
    int32_t tsSlotId = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
    pOptr = createIntervalOperatorInfo(ops[0], pExprInfo, num, pResBlock, &interval, tsSlotId, &as, pTableGroupInfo,
                                       pTaskInfo);

H
Haojun Liao 已提交
4974
  } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {
H
Haojun Liao 已提交
4975
    SSortPhysiNode* pSortPhyNode = (SSortPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4976

4977
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4978
    SArray*      info = createSortInfo(pSortPhyNode->pSortKeys, pSortPhyNode->pTargets);
4979
    SArray*      slotMap = createIndexMap(pSortPhyNode->pTargets);
4980
    pOptr = createSortOperatorInfo(ops[0], pResBlock, info, slotMap, pTaskInfo);
H
Haojun Liao 已提交
4981
  } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
H
Haojun Liao 已提交
4982 4983
    SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;

4984
    STimeWindowAggSupp as = {.waterMark = pSessionNode->window.watermark, .calTrigger = pSessionNode->window.triggerType};
4985

H
Haojun Liao 已提交
4986
    SExprInfo*   pExprInfo = createExprInfo(pSessionNode->window.pFuncs, NULL, &num);
4987
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4988 4989
    int32_t      tsSlotId = ((SColumnNode*)pSessionNode->window.pTspk)->slotId;

4990
    pOptr = createSessionAggOperatorInfo(ops[0], pExprInfo, num, pResBlock, pSessionNode->gap, tsSlotId, &as, pTaskInfo);
H
Haojun Liao 已提交
4991
  } else if (QUERY_NODE_PHYSICAL_PLAN_PARTITION == type) {
dengyihao's avatar
dengyihao 已提交
4992 4993 4994
    SPartitionPhysiNode* pPartNode = (SPartitionPhysiNode*)pPhyNode;
    SArray*              pColList = extractPartitionColInfo(pPartNode->pPartitionKeys);
    SSDataBlock*         pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4995 4996

    SExprInfo* pExprInfo = createExprInfo(pPartNode->pTargets, NULL, &num);
4997
    pOptr = createPartitionOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pTaskInfo, NULL);
4998
  } else if (QUERY_NODE_PHYSICAL_PLAN_STATE_WINDOW == type) {
dengyihao's avatar
dengyihao 已提交
4999
    SStateWinodwPhysiNode* pStateNode = (SStateWinodwPhysiNode*)pPhyNode;
5000

5001 5002
    STimeWindowAggSupp as = {.waterMark = pStateNode->window.watermark, .calTrigger = pStateNode->window.triggerType};

dengyihao's avatar
dengyihao 已提交
5003
    SExprInfo*   pExprInfo = createExprInfo(pStateNode->window.pFuncs, NULL, &num);
5004
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
5005 5006 5007
    int32_t      tsSlotId = ((SColumnNode*)pStateNode->window.pTspk)->slotId;

    pOptr = createStatewindowOperatorInfo(ops[0], pExprInfo, num, pResBlock, &as, tsSlotId, pTaskInfo);
5008
  } else if (QUERY_NODE_PHYSICAL_PLAN_JOIN == type) {
dengyihao's avatar
dengyihao 已提交
5009 5010
    SJoinPhysiNode* pJoinNode = (SJoinPhysiNode*)pPhyNode;
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
5011 5012

    SExprInfo* pExprInfo = createExprInfo(pJoinNode->pTargets, NULL, &num);
5013
    pOptr = createJoinOperatorInfo(ops, size, pExprInfo, num, pResBlock, pJoinNode->pOnConditions, pTaskInfo);
5014 5015 5016 5017 5018
  } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
    SFillPhysiNode* pFillNode = (SFillPhysiNode*)pPhyNode;
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
    SExprInfo* pExprInfo = createExprInfo(pFillNode->pTargets, NULL, &num);

5019
    SInterval* pInterval = &((SIntervalAggOperatorInfo*)ops[0]->info)->interval;
5020
    pOptr = createFillOperatorInfo(ops[0], pExprInfo, num, pInterval, &pFillNode->timeRange, pResBlock, pFillNode->mode, (SNodeListNode*)pFillNode->pValues, false, pTaskInfo);
H
Haojun Liao 已提交
5021 5022
  } else {
    ASSERT(0);
H
Haojun Liao 已提交
5023
  }
5024 5025 5026

  taosMemoryFree(ops);
  return pOptr;
5027
}
H
Haojun Liao 已提交
5028

L
Liu Jicong 已提交
5029
static int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode) {
5030 5031 5032 5033 5034 5035
  pCond->loadExternalRows = false;

  pCond->order = pTableScanNode->scanSeq[0] > 0 ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
  pCond->numOfCols = LIST_LENGTH(pTableScanNode->scan.pScanCols);
  pCond->colList = taosMemoryCalloc(pCond->numOfCols, sizeof(SColumnInfo));
  if (pCond->colList == NULL) {
H
Haojun Liao 已提交
5036
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
5037 5038 5039 5040 5041 5042
    return terrno;
  }

  pCond->twindow = pTableScanNode->scanRange;

#if 1
L
Liu Jicong 已提交
5043
  // todo work around a problem, remove it later
5044 5045
  if ((pCond->order == TSDB_ORDER_ASC && pCond->twindow.skey > pCond->twindow.ekey) ||
      (pCond->order == TSDB_ORDER_DESC && pCond->twindow.skey < pCond->twindow.ekey)) {
wafwerar's avatar
wafwerar 已提交
5046
    TSWAP(pCond->twindow.skey, pCond->twindow.ekey);
H
Haojun Liao 已提交
5047
  }
5048
#endif
H
Haojun Liao 已提交
5049

5050 5051
  pCond->type = BLOCK_LOAD_OFFSET_SEQ_ORDER;
  //  pCond->type = pTableScanNode->scanFlag;
H
Haojun Liao 已提交
5052

H
Haojun Liao 已提交
5053
  int32_t j = 0;
5054
  for (int32_t i = 0; i < pCond->numOfCols; ++i) {
H
Haojun Liao 已提交
5055 5056
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pTableScanNode->scan.pScanCols, i);
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;
H
Haojun Liao 已提交
5057 5058 5059
    if (pColNode->colType == COLUMN_TYPE_TAG) {
      continue;
    }
H
Haojun Liao 已提交
5060

5061 5062 5063
    pCond->colList[j].type = pColNode->node.resType.type;
    pCond->colList[j].bytes = pColNode->node.resType.bytes;
    pCond->colList[j].colId = pColNode->colId;
H
Haojun Liao 已提交
5064
    j += 1;
H
Haojun Liao 已提交
5065 5066
  }

5067 5068
  pCond->numOfCols = j;
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
5069
}
H
Haojun Liao 已提交
5070

H
Haojun Liao 已提交
5071
SArray* extractColumnInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
5072
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5073 5074 5075 5076 5077 5078
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
5079 5080
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
H
Haojun Liao 已提交
5081

5082 5083 5084 5085 5086
    if (nodeType(pNode->pExpr) == QUERY_NODE_COLUMN) {
      SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

      // todo extract method
      SColumn c = {0};
L
Liu Jicong 已提交
5087 5088 5089 5090 5091
      c.slotId = pColNode->slotId;
      c.colId = pColNode->colId;
      c.type = pColNode->node.resType.type;
      c.bytes = pColNode->node.resType.bytes;
      c.scale = pColNode->node.resType.scale;
5092 5093 5094 5095
      c.precision = pColNode->node.resType.precision;

      taosArrayPush(pList, &c);
    } else if (nodeType(pNode->pExpr) == QUERY_NODE_VALUE) {
L
Liu Jicong 已提交
5096 5097
      SValueNode* pValNode = (SValueNode*)pNode->pExpr;
      SColumn     c = {0};
5098
      c.slotId = pNode->slotId;
L
Liu Jicong 已提交
5099 5100 5101 5102
      c.colId = pNode->slotId;
      c.type = pValNode->node.type;
      c.bytes = pValNode->node.resType.bytes;
      c.scale = pValNode->node.resType.scale;
5103 5104 5105 5106
      c.precision = pValNode->node.resType.precision;

      taosArrayPush(pList, &c);
    }
H
Haojun Liao 已提交
5107 5108 5109 5110 5111
  }

  return pList;
}

5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125
SArray* extractPartitionColInfo(SNodeList* pNodeList) {
  size_t  numOfCols = LIST_LENGTH(pNodeList);
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

  for (int32_t i = 0; i < numOfCols; ++i) {
    SColumnNode* pColNode = (SColumnNode*)nodesListGetNode(pNodeList, i);

    // todo extract method
    SColumn c = {0};
    c.slotId = pColNode->slotId;
dengyihao's avatar
dengyihao 已提交
5126 5127 5128
    c.colId = pColNode->colId;
    c.type = pColNode->node.resType.type;
    c.bytes = pColNode->node.resType.bytes;
5129 5130 5131 5132 5133 5134 5135 5136 5137
    c.precision = pColNode->node.resType.precision;
    c.scale = pColNode->node.resType.scale;

    taosArrayPush(pList, &c);
  }

  return pList;
}

5138
SArray* createSortInfo(SNodeList* pNodeList, SNodeList* pNodeListTarget) {
L
Liu Jicong 已提交
5139
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5140 5141 5142 5143 5144 5145
  SArray* pList = taosArrayInit(numOfCols, sizeof(SBlockOrderInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return pList;
  }

L
Liu Jicong 已提交
5146
  for (int32_t i = 0; i < numOfCols; ++i) {
5147
    SOrderByExprNode* pSortKey = (SOrderByExprNode*)nodesListGetNode(pNodeList, i);
L
Liu Jicong 已提交
5148 5149
    SBlockOrderInfo   bi = {0};
    bi.order = (pSortKey->order == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
H
Haojun Liao 已提交
5150 5151 5152 5153
    bi.nullFirst = (pSortKey->nullOrder == NULL_ORDER_FIRST);

    SColumnNode* pColNode = (SColumnNode*)pSortKey->pExpr;

5154 5155 5156 5157 5158
    bool found = false;
    for (int32_t j = 0; j < LIST_LENGTH(pNodeListTarget); ++j) {
      STargetNode* pTarget = (STargetNode*)nodesListGetNode(pNodeListTarget, j);

      SColumnNode* pColNodeT = (SColumnNode*)pTarget->pExpr;
dengyihao's avatar
dengyihao 已提交
5159
      if (pColNode->slotId == pColNodeT->slotId) {  // to find slotId in PhysiSort OutputDataBlockDesc
5160 5161 5162 5163 5164 5165
        bi.slotId = pTarget->slotId;
        found = true;
        break;
      }
    }

dengyihao's avatar
dengyihao 已提交
5166
    if (!found) {
5167 5168
      qError("sort slot id does not found");
    }
H
Haojun Liao 已提交
5169 5170 5171 5172 5173 5174
    taosArrayPush(pList, &bi);
  }

  return pList;
}

5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192
SArray* createIndexMap(SNodeList* pNodeList) {
  size_t  numOfCols = LIST_LENGTH(pNodeList);
  SArray* pList = taosArrayInit(numOfCols, sizeof(int32_t));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return pList;
  }

  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pTarget = (STargetNode*)nodesListGetNode(pNodeList, i);

    SColumnNode* pColNode = (SColumnNode*)pTarget->pExpr;
    taosArrayPush(pList, &pColNode->slotId);
  }

  return pList;
}

H
Haojun Liao 已提交
5193
SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols) {
L
Liu Jicong 已提交
5194
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5195 5196 5197 5198 5199 5200
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColMatchInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
5201 5202 5203
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;
H
Haojun Liao 已提交
5204 5205

    SColMatchInfo c = {0};
5206
    c.output = true;
L
Liu Jicong 已提交
5207
    c.colId = pColNode->colId;
H
Haojun Liao 已提交
5208 5209 5210 5211
    c.targetSlotId = pNode->slotId;
    taosArrayPush(pList, &c);
  }

H
Haojun Liao 已提交
5212 5213
  *numOfOutputCols = 0;
  int32_t num = LIST_LENGTH(pOutputNodeList->pSlots);
L
Liu Jicong 已提交
5214 5215
  for (int32_t i = 0; i < num; ++i) {
    SSlotDescNode* pNode = (SSlotDescNode*)nodesListGetNode(pOutputNodeList->pSlots, i);
5216

5217
    // todo: add reserve flag check
5218 5219
    // it is a column reserved for the arithmetic expression calculation
    if (pNode->slotId >= numOfCols) {
5220 5221 5222 5223
      (*numOfOutputCols) += 1;
      continue;
    }

H
Haojun Liao 已提交
5224
    SColMatchInfo* info = taosArrayGet(pList, pNode->slotId);
H
Haojun Liao 已提交
5225 5226 5227 5228 5229
    if (pNode->output) {
      (*numOfOutputCols) += 1;
    } else {
      info->output = false;
    }
H
Haojun Liao 已提交
5230 5231
  }

H
Haojun Liao 已提交
5232 5233 5234
  return pList;
}

dengyihao's avatar
dengyihao 已提交
5235 5236
int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                           uint64_t queryId, uint64_t taskId) {
5237
  int32_t code = 0;
H
Haojun Liao 已提交
5238
  if (tableType == TSDB_SUPER_TABLE) {
H
Haojun Liao 已提交
5239
    code = tsdbQuerySTableByTagCond(metaHandle, tableUid, 0, NULL, 0, 0, NULL, pGroupInfo, NULL, 0, queryId, taskId);
H
Haojun Liao 已提交
5240
  } else {  // Create one table group.
H
Haojun Liao 已提交
5241
    code = tsdbGetOneTableGroup(metaHandle, tableUid, 0, pGroupInfo);
5242 5243 5244 5245
  }

  return code;
}
H
Haojun Liao 已提交
5246

5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264
SArray* extractTableIdList(const STableGroupInfo* pTableGroupInfo) {
  SArray* tableIdList = taosArrayInit(4, sizeof(uint64_t));

  if (pTableGroupInfo->numOfTables > 0) {
    SArray* pa = taosArrayGetP(pTableGroupInfo->pGroupList, 0);
    ASSERT(taosArrayGetSize(pTableGroupInfo->pGroupList) == 1);

    // Transfer the Array of STableKeyInfo into uid list.
    size_t numOfTables = taosArrayGetSize(pa);
    for (int32_t i = 0; i < numOfTables; ++i) {
      STableKeyInfo* pkeyInfo = taosArrayGet(pa, i);
      taosArrayPush(tableIdList, &pkeyInfo->uid);
    }
  }

  return tableIdList;
}

L
Liu Jicong 已提交
5265 5266
tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                               STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId) {
5267
  uint64_t uid = pTableScanNode->scan.uid;
L
Liu Jicong 已提交
5268 5269
  int32_t  code =
      doCreateTableGroup(pHandle->meta, pTableScanNode->scan.tableType, uid, pTableGroupInfo, queryId, taskId);
5270 5271
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
H
Haojun Liao 已提交
5272
  }
H
Haojun Liao 已提交
5273

H
Haojun Liao 已提交
5274
  if (pTableGroupInfo->numOfTables == 0) {
H
Haojun Liao 已提交
5275
    code = 0;
L
Liu Jicong 已提交
5276
    qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
5277 5278
    goto _error;
  }
H
Haojun Liao 已提交
5279

5280 5281 5282 5283
  SQueryTableDataCond cond = {0};
  code = initQueryTableDataCond(&cond, pTableScanNode);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
X
Xiaoyu Wang 已提交
5284
  }
5285
#if 0
5286
  return tsdbQueryTables(pHandle->reader, &cond, pTableGroupInfo, queryId, taskId);
5287 5288
#endif
  return tsdbQueryTables(pHandle->vnode, &cond, pTableGroupInfo, queryId, taskId);
H
Haojun Liao 已提交
5289

L
Liu Jicong 已提交
5290
_error:
H
Haojun Liao 已提交
5291 5292 5293 5294
  terrno = code;
  return NULL;
}

dengyihao's avatar
dengyihao 已提交
5295 5296
int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
                               EOPTR_EXEC_MODEL model) {
H
Haojun Liao 已提交
5297 5298
  uint64_t queryId = pPlan->id.queryId;

H
Haojun Liao 已提交
5299
  int32_t code = TSDB_CODE_SUCCESS;
5300
  *pTaskInfo = createExecTaskInfo(queryId, taskId, model);
H
Haojun Liao 已提交
5301 5302 5303 5304
  if (*pTaskInfo == NULL) {
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    goto _complete;
  }
H
Haojun Liao 已提交
5305

L
Liu Jicong 已提交
5306 5307
  (*pTaskInfo)->pRoot =
      createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId, &(*pTaskInfo)->tableqinfoGroupInfo);
D
dapan1121 已提交
5308
  if (NULL == (*pTaskInfo)->pRoot) {
L
Liu Jicong 已提交
5309
    code = terrno;
D
dapan1121 已提交
5310 5311
    goto _complete;
  }
H
Haojun Liao 已提交
5312

5313
  if ((*pTaskInfo)->pRoot == NULL) {
H
Haojun Liao 已提交
5314
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
H
Haojun Liao 已提交
5315
    goto _complete;
5316 5317
  }

H
Haojun Liao 已提交
5318 5319
  return code;

H
Haojun Liao 已提交
5320
_complete:
wafwerar's avatar
wafwerar 已提交
5321
  taosMemoryFreeClear(*pTaskInfo);
H
Haojun Liao 已提交
5322 5323 5324

  terrno = code;
  return code;
H
Haojun Liao 已提交
5325 5326
}

L
Liu Jicong 已提交
5327 5328
static int32_t updateOutputBufForTopBotQuery(SQueriedTableInfo* pTableInfo, SColumnInfo* pTagCols, SExprInfo* pExprs,
                                             int32_t numOfOutput, int32_t tagLen, bool superTable) {
5329 5330 5331 5332 5333 5334 5335 5336 5337
  for (int32_t i = 0; i < numOfOutput; ++i) {
    int16_t functId = getExprFunctionId(&pExprs[i]);

    if (functId == FUNCTION_TOP || functId == FUNCTION_BOTTOM) {
      int32_t j = getColumnIndexInSource(pTableInfo, &pExprs[i].base, pTagCols);
      if (j < 0 || j >= pTableInfo->numOfCols) {
        return TSDB_CODE_QRY_INVALID_MSG;
      } else {
        SColumnInfo* pCol = &pTableInfo->colList[j];
L
Liu Jicong 已提交
5338 5339 5340 5341
        //        int32_t ret = getResultDataInfo(pCol->type, pCol->bytes, functId, (int32_t)pExprs[i].base.param[0].i,
        //                                        &pExprs[i].base.resSchema.type, &pExprs[i].base.resSchema.bytes,
        //                                        &pExprs[i].base.interBytes, tagLen, superTable, NULL);
        //        assert(ret == TSDB_CODE_SUCCESS);
5342 5343 5344 5345 5346 5347 5348
      }
    }
  }

  return TSDB_CODE_SUCCESS;
}

H
Haojun Liao 已提交
5349
void setResultBufSize(STaskAttr* pQueryAttr, SResultInfo* pResultInfo) {
5350 5351 5352 5353 5354 5355 5356 5357
  const int32_t DEFAULT_RESULT_MSG_SIZE = 1024 * (1024 + 512);

  // the minimum number of rows for projection query
  const int32_t MIN_ROWS_FOR_PRJ_QUERY = 8192;
  const int32_t DEFAULT_MIN_ROWS = 4096;

  const float THRESHOLD_RATIO = 0.85f;

dengyihao's avatar
dengyihao 已提交
5358 5359 5360 5361 5362 5363 5364 5365 5366 5367
  //  if (isProjQuery(pQueryAttr)) {
  //    int32_t numOfRes = DEFAULT_RESULT_MSG_SIZE / pQueryAttr->resultRowSize;
  //    if (numOfRes < MIN_ROWS_FOR_PRJ_QUERY) {
  //      numOfRes = MIN_ROWS_FOR_PRJ_QUERY;
  //    }
  //
  //    pResultInfo->capacity = numOfRes;
  //  } else {  // in case of non-prj query, a smaller output buffer will be used.
  //    pResultInfo->capacity = DEFAULT_MIN_ROWS;
  //  }
5368 5369

  pResultInfo->threshold = (int32_t)(pResultInfo->capacity * THRESHOLD_RATIO);
H
Haojun Liao 已提交
5370
  pResultInfo->totalRows = 0;
5371 5372
}

L
Liu Jicong 已提交
5373
// TODO refactor
5374
void freeColumnFilterInfo(SColumnFilterInfo* pFilter, int32_t numOfFilters) {
L
Liu Jicong 已提交
5375 5376 5377
  if (pFilter == NULL || numOfFilters == 0) {
    return;
  }
5378

L
Liu Jicong 已提交
5379 5380 5381
  for (int32_t i = 0; i < numOfFilters; i++) {
    if (pFilter[i].filterstr && pFilter[i].pz) {
      taosMemoryFree((void*)(pFilter[i].pz));
5382
    }
L
Liu Jicong 已提交
5383
  }
5384

L
Liu Jicong 已提交
5385
  taosMemoryFree(pFilter);
5386 5387 5388 5389
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo) {
  if (pTableqinfoGroupInfo->pGroupList != NULL) {
L
Liu Jicong 已提交
5390
    int32_t numOfGroups = (int32_t)taosArrayGetSize(pTableqinfoGroupInfo->pGroupList);
5391
    for (int32_t i = 0; i < numOfGroups; ++i) {
L
Liu Jicong 已提交
5392
      SArray* p = taosArrayGetP(pTableqinfoGroupInfo->pGroupList, i);
5393 5394

      size_t num = taosArrayGetSize(p);
L
Liu Jicong 已提交
5395
      for (int32_t j = 0; j < num; ++j) {
5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411
        STableQueryInfo* item = taosArrayGetP(p, j);
        destroyTableQueryInfoImpl(item);
      }

      taosArrayDestroy(p);
    }
  }

  taosArrayDestroy(pTableqinfoGroupInfo->pGroupList);
  taosHashCleanup(pTableqinfoGroupInfo->map);

  pTableqinfoGroupInfo->pGroupList = NULL;
  pTableqinfoGroupInfo->map = NULL;
  pTableqinfoGroupInfo->numOfTables = 0;
}

L
Liu Jicong 已提交
5412
void doDestroyTask(SExecTaskInfo* pTaskInfo) {
H
Haojun Liao 已提交
5413 5414
  qDebug("%s execTask is freed", GET_TASKID(pTaskInfo));

H
Haojun Liao 已提交
5415
  doDestroyTableQueryInfo(&pTaskInfo->tableqinfoGroupInfo);
L
Liu Jicong 已提交
5416 5417
  //  taosArrayDestroy(pTaskInfo->summary.queryProfEvents);
  //  taosHashCleanup(pTaskInfo->summary.operatorProfResults);
5418

wafwerar's avatar
wafwerar 已提交
5419 5420 5421
  taosMemoryFreeClear(pTaskInfo->sql);
  taosMemoryFreeClear(pTaskInfo->id.str);
  taosMemoryFreeClear(pTaskInfo);
5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433
}

static void doSetTagValueToResultBuf(char* output, const char* val, int16_t type, int16_t bytes) {
  if (val == NULL) {
    setNull(output, type, bytes);
    return;
  }

  if (IS_VAR_DATA_TYPE(type)) {
    // Binary data overflows for sort of unknown reasons. Let trim the overflow data
    if (varDataTLen(val) > bytes) {
      int32_t maxLen = bytes - VARSTR_HEADER_SIZE;
L
Liu Jicong 已提交
5434
      int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446
      memcpy(varDataVal(output), varDataVal(val), len);
      varDataSetLen(output, len);
    } else {
      varDataCopy(output, val);
    }
  } else {
    memcpy(output, val, bytes);
  }
}

static int64_t getQuerySupportBufSize(size_t numOfTables) {
  size_t s1 = sizeof(STableQueryInfo);
L
Liu Jicong 已提交
5447 5448
  //  size_t s3 = sizeof(STableCheckInfo);  buffer consumption in tsdb
  return (int64_t)(s1 * 1.5 * numOfTables);
5449 5450 5451 5452 5453 5454 5455
}

int32_t checkForQueryBuf(size_t numOfTables) {
  int64_t t = getQuerySupportBufSize(numOfTables);
  if (tsQueryBufferSizeBytes < 0) {
    return TSDB_CODE_SUCCESS;
  } else if (tsQueryBufferSizeBytes > 0) {
L
Liu Jicong 已提交
5456
    while (1) {
5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482
      int64_t s = tsQueryBufferSizeBytes;
      int64_t remain = s - t;
      if (remain >= 0) {
        if (atomic_val_compare_exchange_64(&tsQueryBufferSizeBytes, s, remain) == s) {
          return TSDB_CODE_SUCCESS;
        }
      } else {
        return TSDB_CODE_QRY_NOT_ENOUGH_BUFFER;
      }
    }
  }

  // disable query processing if the value of tsQueryBufferSize is zero.
  return TSDB_CODE_QRY_NOT_ENOUGH_BUFFER;
}

void releaseQueryBuf(size_t numOfTables) {
  if (tsQueryBufferSizeBytes < 0) {
    return;
  }

  int64_t t = getQuerySupportBufSize(numOfTables);

  // restore value is not enough buffer available
  atomic_add_fetch_64(&tsQueryBufferSizeBytes, t);
}
D
dapan1121 已提交
5483

dengyihao's avatar
dengyihao 已提交
5484 5485
int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SExplainExecInfo** pRes, int32_t* capacity,
                                   int32_t* resNum) {
D
dapan1121 已提交
5486 5487
  if (*resNum >= *capacity) {
    *capacity += 10;
dengyihao's avatar
dengyihao 已提交
5488

D
dapan1121 已提交
5489 5490
    *pRes = taosMemoryRealloc(*pRes, (*capacity) * sizeof(SExplainExecInfo));
    if (NULL == *pRes) {
D
dapan1121 已提交
5491
      qError("malloc %d failed", (*capacity) * (int32_t)sizeof(SExplainExecInfo));
D
dapan1121 已提交
5492 5493 5494 5495
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

D
dapan1121 已提交
5496 5497 5498 5499
  (*pRes)[*resNum].numOfRows = operatorInfo->resultInfo.totalRows;
  (*pRes)[*resNum].startupCost = operatorInfo->cost.openCost;
  (*pRes)[*resNum].totalCost = operatorInfo->cost.totalCost;

5500 5501
  if (operatorInfo->fpSet.getExplainFn) {
    int32_t code = (*operatorInfo->fpSet.getExplainFn)(operatorInfo, &(*pRes)->verboseInfo);
D
dapan1121 已提交
5502 5503 5504 5505 5506
    if (code) {
      qError("operator getExplainFn failed, error:%s", tstrerror(code));
      return code;
    }
  }
dengyihao's avatar
dengyihao 已提交
5507

D
dapan1121 已提交
5508
  ++(*resNum);
dengyihao's avatar
dengyihao 已提交
5509

D
dapan1121 已提交
5510
  int32_t code = 0;
D
dapan1121 已提交
5511 5512
  for (int32_t i = 0; i < operatorInfo->numOfDownstream; ++i) {
    code = getOperatorExplainExecInfo(operatorInfo->pDownstream[i], pRes, capacity, resNum);
D
dapan1121 已提交
5513 5514 5515 5516 5517 5518 5519
    if (code) {
      taosMemoryFreeClear(*pRes);
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
D
dapan1121 已提交
5520 5521
}

5522
static SSDataBlock* doMergeJoin(struct SOperatorInfo* pOperator) {
5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534
  SJoinOperatorInfo* pJoinInfo = pOperator->info;

  SSDataBlock* pRes = pJoinInfo->pRes;
  blockDataCleanup(pRes);
  blockDataEnsureCapacity(pRes, 4096);

  int32_t nrows = 0;

  while (1) {
    if (pJoinInfo->pLeft == NULL || pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
      SOperatorInfo* ds1 = pOperator->pDownstream[0];
      publishOperatorProfEvent(ds1, QUERY_PROF_BEFORE_OPERATOR_EXEC);
5535
      pJoinInfo->pLeft = ds1->fpSet.getNextFn(ds1);
5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547
      publishOperatorProfEvent(ds1, QUERY_PROF_AFTER_OPERATOR_EXEC);

      pJoinInfo->leftPos = 0;
      if (pJoinInfo->pLeft == NULL) {
        setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
        break;
      }
    }

    if (pJoinInfo->pRight == NULL || pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
      SOperatorInfo* ds2 = pOperator->pDownstream[1];
      publishOperatorProfEvent(ds2, QUERY_PROF_BEFORE_OPERATOR_EXEC);
5548
      pJoinInfo->pRight = ds2->fpSet.getNextFn(ds2);
5549 5550 5551 5552 5553 5554 5555 5556 5557 5558
      publishOperatorProfEvent(ds2, QUERY_PROF_AFTER_OPERATOR_EXEC);

      pJoinInfo->rightPos = 0;
      if (pJoinInfo->pRight == NULL) {
        setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
        break;
      }
    }

    SColumnInfoData* pLeftCol = taosArrayGet(pJoinInfo->pLeft->pDataBlock, pJoinInfo->leftCol.slotId);
dengyihao's avatar
dengyihao 已提交
5559
    char*            pLeftVal = colDataGetData(pLeftCol, pJoinInfo->leftPos);
5560 5561

    SColumnInfoData* pRightCol = taosArrayGet(pJoinInfo->pRight->pDataBlock, pJoinInfo->rightCol.slotId);
dengyihao's avatar
dengyihao 已提交
5562
    char*            pRightVal = colDataGetData(pRightCol, pJoinInfo->rightPos);
5563 5564

    // only the timestamp match support for ordinary table
dengyihao's avatar
dengyihao 已提交
5565 5566 5567 5568
    ASSERT(pLeftCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
    if (*(int64_t*)pLeftVal == *(int64_t*)pRightVal) {
      for (int32_t i = 0; i < pOperator->numOfOutput; ++i) {
        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, i);
5569

dengyihao's avatar
dengyihao 已提交
5570
        SExprInfo* pExprInfo = &pOperator->pExpr[i];
5571

dengyihao's avatar
dengyihao 已提交
5572 5573
        int32_t blockId = pExprInfo->base.pParam[0].pCol->dataBlockId;
        int32_t slotId = pExprInfo->base.pParam[0].pCol->slotId;
5574

dengyihao's avatar
dengyihao 已提交
5575 5576 5577 5578 5579 5580
        SColumnInfoData* pSrc = NULL;
        if (pJoinInfo->pLeft->info.blockId == blockId) {
          pSrc = taosArrayGet(pJoinInfo->pLeft->pDataBlock, slotId);
        } else {
          pSrc = taosArrayGet(pJoinInfo->pRight->pDataBlock, slotId);
        }
5581

dengyihao's avatar
dengyihao 已提交
5582 5583 5584 5585 5586
        if (colDataIsNull_s(pSrc, pJoinInfo->leftPos)) {
          colDataAppendNULL(pDst, nrows);
        } else {
          char* p = colDataGetData(pSrc, pJoinInfo->leftPos);
          colDataAppend(pDst, nrows, p, false);
5587
        }
dengyihao's avatar
dengyihao 已提交
5588
      }
5589

dengyihao's avatar
dengyihao 已提交
5590 5591
      pJoinInfo->leftPos += 1;
      pJoinInfo->rightPos += 1;
5592

dengyihao's avatar
dengyihao 已提交
5593 5594 5595
      nrows += 1;
    } else if (*(int64_t*)pLeftVal < *(int64_t*)pRightVal) {
      pJoinInfo->leftPos += 1;
D
dapan1121 已提交
5596

dengyihao's avatar
dengyihao 已提交
5597 5598
      if (pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
        continue;
5599
      }
dengyihao's avatar
dengyihao 已提交
5600 5601 5602 5603
    } else if (*(int64_t*)pLeftVal > *(int64_t*)pRightVal) {
      pJoinInfo->rightPos += 1;
      if (pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
        continue;
5604
      }
dengyihao's avatar
dengyihao 已提交
5605
    }
5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616

    // the pDataBlock are always the same one, no need to call this again
    pRes->info.rows = nrows;
    if (pRes->info.rows >= pOperator->resultInfo.threshold) {
      break;
    }
  }

  return (pRes->info.rows > 0) ? pRes : NULL;
}

dengyihao's avatar
dengyihao 已提交
5617 5618 5619
SOperatorInfo* createJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
                                      int32_t numOfCols, SSDataBlock* pResBlock, SNode* pOnCondition,
                                      SExecTaskInfo* pTaskInfo) {
5620
  SJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SJoinOperatorInfo));
dengyihao's avatar
dengyihao 已提交
5621
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
5622 5623 5624 5625 5626 5627 5628
  if (pOperator == NULL || pInfo == NULL) {
    goto _error;
  }

  pOperator->resultInfo.capacity = 4096;
  pOperator->resultInfo.threshold = 4096 * 0.75;

dengyihao's avatar
dengyihao 已提交
5629 5630 5631
  //  initResultRowInf
  //  o(&pInfo->binfo.resultRowInfo, 8);
  pInfo->pRes = pResBlock;
5632

dengyihao's avatar
dengyihao 已提交
5633
  pOperator->name = "JoinOperator";
5634
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_JOIN;
5635
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
5636 5637 5638 5639 5640
  pOperator->status = OP_NOT_OPENED;
  pOperator->pExpr = pExprInfo;
  pOperator->numOfOutput = numOfCols;
  pOperator->info = pInfo;
  pOperator->pTaskInfo = pTaskInfo;
5641

L
Liu Jicong 已提交
5642 5643
  pOperator->fpSet =
      createOperatorFpSet(operatorDummyOpenFn, doMergeJoin, NULL, NULL, destroyBasicOperatorInfo, NULL, NULL, NULL);
5644 5645 5646
  int32_t code = appendDownstream(pOperator, pDownstream, numOfDownstream);
  return pOperator;

dengyihao's avatar
dengyihao 已提交
5647
_error:
5648 5649 5650 5651
  taosMemoryFree(pInfo);
  taosMemoryFree(pOperator);
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
dengyihao's avatar
dengyihao 已提交
5652
}