executorimpl.c 199.2 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

16
#include <libs/function/function.h>
H
Haojun Liao 已提交
17 18
#include "filter.h"
#include "function.h"
19 20
#include "functionMgt.h"
#include "os.h"
H
Haojun Liao 已提交
21
#include "querynodes.h"
22
#include "tfill.h"
dengyihao's avatar
dengyihao 已提交
23
#include "tname.h"
24

H
Haojun Liao 已提交
25
#include "tdatablock.h"
26
#include "tglobal.h"
H
Haojun Liao 已提交
27
#include "tmsg.h"
H
Haojun Liao 已提交
28
#include "tsort.h"
29
#include "ttime.h"
H
Haojun Liao 已提交
30

31
#include "executorimpl.h"
32
#include "query.h"
33 34
#include "tcompare.h"
#include "tcompression.h"
H
Haojun Liao 已提交
35
#include "thash.h"
36
#include "ttypes.h"
dengyihao's avatar
dengyihao 已提交
37
#include "vnode.h"
38

H
Haojun Liao 已提交
39
#define IS_MAIN_SCAN(runtime)          ((runtime)->scanFlag == MAIN_SCAN)
40 41
#define IS_REVERSE_SCAN(runtime)       ((runtime)->scanFlag == REVERSE_SCAN)
#define IS_REPEAT_SCAN(runtime)        ((runtime)->scanFlag == REPEAT_SCAN)
L
Liu Jicong 已提交
42
#define SET_MAIN_SCAN_FLAG(runtime)    ((runtime)->scanFlag = MAIN_SCAN)
43 44
#define SET_REVERSE_SCAN_FLAG(runtime) ((runtime)->scanFlag = REVERSE_SCAN)

L
Liu Jicong 已提交
45 46
#define SDATA_BLOCK_INITIALIZER \
  (SDataBlockInfo) { {0}, 0 }
47 48 49 50

#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
51 52
  TS_JOIN_TS_EQUAL = 0,
  TS_JOIN_TS_NOT_EQUALS = 1,
53 54 55 56 57
  TS_JOIN_TAG_NOT_EQUALS = 2,
};

#if 0
static UNUSED_FUNC void *u_malloc (size_t __size) {
wafwerar's avatar
wafwerar 已提交
58
  uint32_t v = taosRand();
59 60 61 62

  if (v % 1000 <= 0) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
63
    return taosMemoryMalloc(__size);
64 65 66 67
  }
}

static UNUSED_FUNC void* u_calloc(size_t num, size_t __size) {
wafwerar's avatar
wafwerar 已提交
68
  uint32_t v = taosRand();
69 70 71
  if (v % 1000 <= 0) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
72
    return taosMemoryCalloc(num, __size);
73 74 75 76
  }
}

static UNUSED_FUNC void* u_realloc(void* p, size_t __size) {
wafwerar's avatar
wafwerar 已提交
77
  uint32_t v = taosRand();
78 79 80
  if (v % 5 <= 1) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
81
    return taosMemoryRealloc(p, __size);
82 83 84 85 86 87 88 89 90 91 92 93
  }
}

#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
94 95 96
int32_t getMaximumIdleDurationSec() { return tsShellActivityTimer * 2; }

static int32_t getExprFunctionId(SExprInfo* pExprInfo) {
97
  assert(pExprInfo != NULL && pExprInfo->pExpr != NULL && pExprInfo->pExpr->nodeType == TEXPR_UNARYEXPR_NODE);
98
  return 0;
99 100 101
}

static void doSetTagValueToResultBuf(char* output, const char* val, int16_t type, int16_t bytes);
L
Liu Jicong 已提交
102
static bool functionNeedToExecute(SqlFunctionCtx* pCtx);
103

104
static void setBlockStatisInfo(SqlFunctionCtx* pCtx, SExprInfo* pExpr, SSDataBlock* pSDataBlock);
105

L
Liu Jicong 已提交
106
static void destroyTableQueryInfoImpl(STableQueryInfo* pTableQueryInfo);
107 108 109

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

L
Liu Jicong 已提交
110
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

158
static int32_t doCopyToSDataBlock(SExecTaskInfo *taskInfo, SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf,
dengyihao's avatar
dengyihao 已提交
159 160
                                  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 ??
5
54liuyao 已提交
201
  pBlock->info.type = STREAM_INVALID;
H
Haojun Liao 已提交
202

L
Liu Jicong 已提交
203
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
204
    SColumnInfoData idata = {{0}};
L
Liu Jicong 已提交
205
    SSlotDescNode*  pDescNode = nodesListGetNode(pNode->pSlots, i);
X
Xiaoyu Wang 已提交
206 207 208
    //    if (!pDescNode->output) {  // todo disable it temporarily
    //      continue;
    //    }
H
Haojun Liao 已提交
209

dengyihao's avatar
dengyihao 已提交
210
    idata.info.type = pDescNode->dataType.type;
L
Liu Jicong 已提交
211 212
    idata.info.bytes = pDescNode->dataType.bytes;
    idata.info.scale = pDescNode->dataType.scale;
H
Haojun Liao 已提交
213
    idata.info.slotId = pDescNode->slotId;
H
Haojun Liao 已提交
214 215
    idata.info.precision = pDescNode->dataType.precision;

H
Hongze Cheng 已提交
216 217 218 219
    if (IS_VAR_DATA_TYPE(idata.info.type)) {
      pBlock->info.hasVarCol = true;
    }

H
Haojun Liao 已提交
220
    taosArrayPush(pBlock->pDataBlock, &idata);
H
Haojun Liao 已提交
221
  }
H
Haojun Liao 已提交
222

223
  pBlock->info.numOfCols = taosArrayGetSize(pBlock->pDataBlock);
H
Haojun Liao 已提交
224
  return pBlock;
H
Haojun Liao 已提交
225 226
}

L
Liu Jicong 已提交
227
static bool hasNull(SColumn* pColumn, SColumnDataAgg* pStatis) {
dengyihao's avatar
dengyihao 已提交
228 229
  if (TSDB_COL_IS_TAG(pColumn->flag) || TSDB_COL_IS_UD_COL(pColumn->flag) ||
      pColumn->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
230 231 232 233 234 235 236 237 238 239
    return false;
  }

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

  return true;
}

H
Haojun Liao 已提交
240
static void prepareResultListBuffer(SResultRowInfo* pResultRowInfo, jmp_buf env) {
241 242
  int64_t newCapacity = 0;

243 244 245 246 247 248 249 250 251 252 253
  // 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);
  }

254
  if (newCapacity <= pResultRowInfo->capacity) {
H
Haojun Liao 已提交
255 256 257
    newCapacity += 4;
  }

258 259 260 261 262 263
  char* p = taosMemoryRealloc(pResultRowInfo->pPosition, newCapacity * sizeof(SResultRowPosition));
  if (p == NULL) {
    longjmp(env, TSDB_CODE_OUT_OF_MEMORY);
  }

  pResultRowInfo->pPosition = (SResultRowPosition*)p;
264 265

  int32_t inc = (int32_t)newCapacity - pResultRowInfo->capacity;
wmmhello's avatar
wmmhello 已提交
266
  memset(&pResultRowInfo->pPosition[pResultRowInfo->capacity], 0, sizeof(SResultRowPosition) * inc);
267 268 269
  pResultRowInfo->capacity = (int32_t)newCapacity;
}

L
Liu Jicong 已提交
270 271
static bool chkResultRowFromKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, char* pData,
                                int16_t bytes, bool masterscan, uint64_t uid) {
272 273 274
  bool existed = false;
  SET_RES_WINDOW_KEY(pRuntimeEnv->keyBuf, pData, bytes, uid);

L
Liu Jicong 已提交
275 276
  SResultRow** p1 =
      (SResultRow**)taosHashGet(pRuntimeEnv->pResultRowHashTable, pRuntimeEnv->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
277 278 279 280 281 282 283 284 285 286 287

  // 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
288
        //        existed = (pResultRowInfo->pResult[0] == (*p1));
289 290
      } 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
291 292
        int64_t* index =
            taosHashGet(pRuntimeEnv->pResultRowListSet, pRuntimeEnv->keyBuf, GET_RES_EXT_WINDOW_KEY_LEN(bytes));
293 294 295 296 297 298 299 300 301 302 303 304 305 306
        if (index != NULL) {
          existed = true;
        } else {
          existed = false;
        }
      }
    }

    return existed;
  }

  return p1 != NULL;
}

307
SResultRow* getNewResultRow_rv(SDiskbasedBuf* pResultBuf, int64_t tableGroupId, int32_t interBufSize) {
L
Liu Jicong 已提交
308
  SFilePage* pData = NULL;
309 310 311 312 313 314 315 316 317 318 319 320 321

  // 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
322
    if (pData->num + interBufSize > getBufPageSize(pResultBuf)) {
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341
      // 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
342
  pData->num += interBufSize;
343 344 345 346

  return pResultRow;
}

347 348 349 350 351 352 353
/**
 * the struct of key in hash table
 * +----------+---------------+
 * | group id |   key data    |
 * | 8 bytes  | actual length |
 * +----------+---------------+
 */
354 355 356
SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pResultRowInfo, char* pData,
                                   int16_t bytes, bool masterscan, uint64_t groupId, SExecTaskInfo* pTaskInfo,
                                   bool isIntervalQuery, SAggSupporter* pSup) {
357
  SET_RES_WINDOW_KEY(pSup->keyBuf, pData, bytes, groupId);
H
Haojun Liao 已提交
358

dengyihao's avatar
dengyihao 已提交
359 360
  SResultRowPosition* p1 =
      (SResultRowPosition*)taosHashGet(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
H
Haojun Liao 已提交
361

362 363
  SResultRow* pResult = NULL;

H
Haojun Liao 已提交
364 365
  // in case of repeat scan/reverse scan, no new time window added.
  if (isIntervalQuery) {
366 367
    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
368 369
    }
  } else {
dengyihao's avatar
dengyihao 已提交
370 371
    // In case of group by column query, the required SResultRow object must be existInCurrentResusltRowInfo in the
    // pResultRowInfo object.
H
Haojun Liao 已提交
372
    if (p1 != NULL) {
373
      pResult = getResultRowByPos(pResultBuf, p1);
H
Haojun Liao 已提交
374 375 376
    }
  }

L
Liu Jicong 已提交
377
  // 1. close current opened time window
378
  if (pResultRowInfo->cur.pageId != -1 && ((pResult == NULL) || (pResult->pageId != pResultRowInfo->cur.pageId &&
L
Liu Jicong 已提交
379
                                                                 pResult->offset != pResultRowInfo->cur.offset))) {
380 381 382 383 384 385 386 387 388 389 390
    // 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) {
H
Haojun Liao 已提交
391
    ASSERT(pSup->resultRowSize > 0);
392 393
    pResult = getNewResultRow_rv(pResultBuf, groupId, pSup->resultRowSize);
    initResultRow(pResult);
H
Haojun Liao 已提交
394

395 396
    // add a new result set for a new group
    SResultRowPosition pos = {.pageId = pResult->pageId, .offset = pResult->offset};
L
Liu Jicong 已提交
397 398
    taosHashPut(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes), &pos,
                sizeof(SResultRowPosition));
H
Haojun Liao 已提交
399 400
  }

401 402 403 404 405
  // 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
406 407 408 409 410
  // 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
411
  return pResult;
H
Haojun Liao 已提交
412 413
}

414
// get the correct time window according to the handled timestamp
L
Liu Jicong 已提交
415
static STimeWindow getCurrentActiveTimeWindow(SResultRowInfo* pResultRowInfo, int64_t ts, STaskAttr* pQueryAttr) {
416
  STimeWindow w = {0};
H
Haojun Liao 已提交
417
#if 0
L
Liu Jicong 已提交
418 419
  if (pResultRowInfo->curPos == -1) {  // the first window, from the previous stored value
                                       //    getInitialStartTimeWindow(pQueryAttr, ts, &w);
420 421

    if (pQueryAttr->interval.intervalUnit == 'n' || pQueryAttr->interval.intervalUnit == 'y') {
L
Liu Jicong 已提交
422 423 424
      w.ekey =
          taosTimeAdd(w.skey, pQueryAttr->interval.interval, pQueryAttr->interval.intervalUnit, pQueryAttr->precision) -
          1;
425 426 427 428
    } else {
      w.ekey = w.skey + pQueryAttr->interval.interval - 1;
    }
  } else {
H
Haojun Liao 已提交
429
    w = pRow->win;
430 431 432 433 434 435 436 437 438
  }

  /*
   * 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
439
#endif
440 441 442 443 444

  return w;
}

// a new buffer page for each table. Needs to opt this design
L
Liu Jicong 已提交
445
static int32_t addNewWindowResultBuf(SResultRow* pWindowRes, SDiskbasedBuf* pResultBuf, int32_t tid, uint32_t size) {
446 447 448 449
  if (pWindowRes->pageId != -1) {
    return 0;
  }

L
Liu Jicong 已提交
450
  SFilePage* pData = NULL;
451 452 453 454 455 456

  // 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
457
    pData = getNewBufPage(pResultBuf, tid, &pageId);
458
    pData->num = sizeof(SFilePage);
459 460
  } else {
    SPageInfo* pi = getLastPageInfo(list);
461
    pData = getBufPage(pResultBuf, getPageId(pi));
462
    pageId = getPageId(pi);
463

464
    if (pData->num + size > getBufPageSize(pResultBuf)) {
465
      // release current page first, and prepare the next one
466
      releaseBufPageInfo(pResultBuf, pi);
467

H
Haojun Liao 已提交
468
      pData = getNewBufPage(pResultBuf, tid, &pageId);
469
      if (pData != NULL) {
470
        pData->num = sizeof(SFilePage);
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490
      }
    }
  }

  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
491 492 493
static bool chkWindowOutputBufByKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, STimeWindow* win,
                                    bool masterscan, SResultRow** pResult, int64_t groupId, SqlFunctionCtx* pCtx,
                                    int32_t numOfOutput, int32_t* rowCellInfoOffset) {
494
  assert(win->skey <= win->ekey);
L
Liu Jicong 已提交
495
  return chkResultRowFromKey(pRuntimeEnv, pResultRowInfo, (char*)&win->skey, TSDB_KEYSIZE, masterscan, groupId);
496 497
}

dengyihao's avatar
dengyihao 已提交
498 499
static void doUpdateResultRowIndex(SResultRowInfo* pResultRowInfo, TSKEY lastKey, bool ascQuery,
                                   bool timeWindowInterpo) {
500
  int64_t skey = TSKEY_INITIAL_VAL;
H
Haojun Liao 已提交
501
#if 0
502 503
  int32_t i = 0;
  for (i = pResultRowInfo->size - 1; i >= 0; --i) {
L
Liu Jicong 已提交
504
    SResultRow* pResult = pResultRowInfo->pResult[i];
505 506 507 508 509 510
    if (pResult->closed) {
      break;
    }

    // new closed result rows
    if (timeWindowInterpo) {
L
Liu Jicong 已提交
511 512 513
      if (pResult->endInterp &&
          ((pResult->win.skey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery))) {
        if (i > 0) {  // the first time window, the startInterp is false.
514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532
          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
533
      //      assert(pResultRowInfo->current == NULL);
534 535 536 537 538 539 540
      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
541
      SResultRow* pResult = pResultRowInfo->pResult[i];
542 543 544 545 546 547 548 549 550 551 552
      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
553
#endif
554
}
555
//
dengyihao's avatar
dengyihao 已提交
556
// static void updateResultRowInfoActiveIndex(SResultRowInfo* pResultRowInfo, const STimeWindow* pWin, TSKEY lastKey,
557 558 559 560 561 562 563 564 565
//                                           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);
//  }
//}
566

567
//  query_range_start, query_range_end, window_duration, window_start, window_end
568
void initExecTimeWindowInfo(SColumnInfoData* pColData, STimeWindow* pQueryWindow) {
569 570 571
  pColData->info.type = TSDB_DATA_TYPE_TIMESTAMP;
  pColData->info.bytes = sizeof(int64_t);

572
  colInfoDataEnsureCapacity(pColData, 0, 5);
573 574 575 576 577 578 579 580 581
  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);
}

582
void doApplyFunctions(SExecTaskInfo* taskInfo, SqlFunctionCtx* pCtx, STimeWindow* pWin, SColumnInfoData* pTimeWindowData, int32_t offset,
dengyihao's avatar
dengyihao 已提交
583
                      int32_t forwardStep, TSKEY* tsCol, int32_t numOfTotal, int32_t numOfOutput, int32_t order) {
584 585 586
  for (int32_t k = 0; k < numOfOutput; ++k) {
    pCtx[k].startTs = pWin->skey;

H
Haojun Liao 已提交
587
    // keep it temporarily
dengyihao's avatar
dengyihao 已提交
588 589
    bool    hasAgg = pCtx[k].input.colDataAggIsSet;
    int32_t numOfRows = pCtx[k].input.numOfRows;
H
Haojun Liao 已提交
590
    int32_t startOffset = pCtx[k].input.startRowIndex;
591

H
Haojun Liao 已提交
592
    int32_t pos = (order == TSDB_ORDER_ASC) ? offset : offset - (forwardStep - 1);
593 594
    pCtx[k].input.startRowIndex = pos;
    pCtx[k].input.numOfRows = forwardStep;
595 596

    if (tsCol != NULL) {
H
Haojun Liao 已提交
597
      pCtx[k].ptsList = tsCol;
598 599 600 601 602 603 604 605
    }

    // 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;
    }

606 607
    if (fmIsWindowPseudoColumnFunc(pCtx[k].functionId)) {
      SResultRowEntryInfo* pEntryInfo = GET_RES_INFO(&pCtx[k]);
dengyihao's avatar
dengyihao 已提交
608
      char*                p = GET_ROWCELL_INTERBUF(pEntryInfo);
609

610
      SColumnInfoData idata = {0};
dengyihao's avatar
dengyihao 已提交
611
      idata.info.type = TSDB_DATA_TYPE_BIGINT;
612
      idata.info.bytes = tDataTypes[TSDB_DATA_TYPE_BIGINT].bytes;
dengyihao's avatar
dengyihao 已提交
613
      idata.pData = p;
614 615 616 617

      SScalarParam out = {.columnData = &idata};
      SScalarParam tw = {.numOfRows = 5, .columnData = pTimeWindowData};
      pCtx[k].sfp.process(&tw, 1, &out);
618 619 620
      pEntryInfo->numOfRes = 1;
      continue;
    }
621
    int32_t code = TSDB_CODE_SUCCESS;
622
    if (functionNeedToExecute(&pCtx[k]) && pCtx[k].fpSet.process != NULL) {
623 624 625 626 627 628
      code = pCtx[k].fpSet.process(&pCtx[k]);
      if (code != TSDB_CODE_SUCCESS) {
        qError("%s apply functions error, code: %s", GET_TASKID(taskInfo), tstrerror(code));
        taskInfo->code = code;
        longjmp(taskInfo->env, code);
      }
629 630 631
    }

    // restore it
632 633 634
    pCtx[k].input.colDataAggIsSet = hasAgg;
    pCtx[k].input.startRowIndex = startOffset;
    pCtx[k].input.numOfRows = numOfRows;
635 636 637
  }
}

L
Liu Jicong 已提交
638
static FORCE_INLINE TSKEY reviseWindowEkey(STaskAttr* pQueryAttr, STimeWindow* pWindow) {
dengyihao's avatar
dengyihao 已提交
639
  TSKEY   ekey = -1;
640 641
  int32_t order = TSDB_ORDER_ASC;
  if (order == TSDB_ORDER_ASC) {
642 643 644 645 646 647 648 649 650 651 652 653 654 655
    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
656 657
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                                   bool createDummyCol);
658

dengyihao's avatar
dengyihao 已提交
659 660
static void doSetInputDataBlockInfo(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock,
                                    int32_t order) {
661
  for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
662
    pCtx[i].order = order;
L
Liu Jicong 已提交
663
    pCtx[i].size = pBlock->info.rows;
664
    setBlockStatisInfo(&pCtx[i], &pOperator->pExpr[i], pBlock);
665 666 667
  }
}

dengyihao's avatar
dengyihao 已提交
668 669
void setInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                       bool createDummyCol) {
670
  if (pBlock->pBlockAgg != NULL) {
H
Haojun Liao 已提交
671
    doSetInputDataBlockInfo(pOperator, pCtx, pBlock, order);
672
  } else {
673
    doSetInputDataBlock(pOperator, pCtx, pBlock, order, createDummyCol);
H
Haojun Liao 已提交
674
  }
675 676
}

L
Liu Jicong 已提交
677 678
static int32_t doCreateConstantValColumnInfo(SInputColumnInfoData* pInput, SFunctParam* pFuncParam, int32_t paramIndex,
                                             int32_t numOfRows) {
679 680 681 682 683 684 685 686
  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)
687 688
    pColInfo->info.type = pFuncParam->param.nType;
    pColInfo->info.bytes = pFuncParam->param.nLen;
689 690

    pInput->pData[paramIndex] = pColInfo;
691 692
  } else {
    pColInfo = pInput->pData[paramIndex];
693 694
  }

695
  colInfoDataEnsureCapacity(pColInfo, 0, numOfRows);
696

697
  int8_t type = pFuncParam->param.nType;
698 699
  if (type == TSDB_DATA_TYPE_BIGINT || type == TSDB_DATA_TYPE_UBIGINT) {
    int64_t v = pFuncParam->param.i;
dengyihao's avatar
dengyihao 已提交
700
    for (int32_t i = 0; i < numOfRows; ++i) {
701 702 703 704
      colDataAppendInt64(pColInfo, i, &v);
    }
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
dengyihao's avatar
dengyihao 已提交
705
    for (int32_t i = 0; i < numOfRows; ++i) {
706 707
      colDataAppendDouble(pColInfo, i, &v);
    }
708
  } else if (type == TSDB_DATA_TYPE_VARCHAR) {
L
Liu Jicong 已提交
709
    char* tmp = taosMemoryMalloc(pFuncParam->param.nLen + VARSTR_HEADER_SIZE);
710
    STR_WITH_SIZE_TO_VARSTR(tmp, pFuncParam->param.pz, pFuncParam->param.nLen);
L
Liu Jicong 已提交
711
    for (int32_t i = 0; i < numOfRows; ++i) {
712 713
      colDataAppend(pColInfo, i, tmp, false);
    }
714 715 716 717 718
  }

  return TSDB_CODE_SUCCESS;
}

dengyihao's avatar
dengyihao 已提交
719 720
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                                   bool createDummyCol) {
721 722
  int32_t code = TSDB_CODE_SUCCESS;

723
  for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
L
Liu Jicong 已提交
724 725 726
    pCtx[i].order = order;
    pCtx[i].size = pBlock->info.rows;
    pCtx[i].pSrcBlock = pBlock;
H
Haojun Liao 已提交
727 728
    pCtx[i].currentStage = MAIN_SCAN;

729
    SInputColumnInfoData* pInput = &pCtx[i].input;
730
    pInput->uid = pBlock->info.uid;
C
Cary Xu 已提交
731
    pInput->colDataAggIsSet = false;
732

733 734
    SExprInfo* pOneExpr = &pOperator->pExpr[i];
    for (int32_t j = 0; j < pOneExpr->base.numOfParams; ++j) {
dengyihao's avatar
dengyihao 已提交
735
      SFunctParam* pFuncParam = &pOneExpr->base.pParam[j];
G
Ganlin Zhao 已提交
736 737
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
dengyihao's avatar
dengyihao 已提交
738
        pInput->pData[j] = taosArrayGet(pBlock->pDataBlock, slotId);
739 740 741
        pInput->totalRows = pBlock->info.rows;
        pInput->numOfRows = pBlock->info.rows;
        pInput->startRowIndex = 0;
742

743 744 745 746
        // the last parameter is the timestamp column
        if (fmIsTimelineFunc(pCtx[i].functionId) && (j == pOneExpr->base.numOfParams - 1)) {
          pInput->pPTS = pInput->pData[j];
        }
747 748
        ASSERT(pInput->pData[j] != NULL);
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
749 750 751
        // 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) {
752 753 754 755
          pInput->totalRows = pBlock->info.rows;
          pInput->numOfRows = pBlock->info.rows;
          pInput->startRowIndex = 0;

756
          code = doCreateConstantValColumnInfo(pInput, pFuncParam, j, pBlock->info.rows);
757 758 759
          if (code != TSDB_CODE_SUCCESS) {
            return code;
          }
760
        }
G
Ganlin Zhao 已提交
761 762
      }
    }
H
Haojun Liao 已提交
763

H
Haojun Liao 已提交
764
    //    setBlockStatisInfo(&pCtx[i], pBlock, pOperator->pExpr[i].base.pColumns);
H
Haojun Liao 已提交
765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802
    //      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);
    //        }
    //      }
  }
803 804

  return code;
H
Haojun Liao 已提交
805 806 807
}

static void doAggregateImpl(SOperatorInfo* pOperator, TSKEY startTs, SqlFunctionCtx* pCtx) {
808
  for (int32_t k = 0; k < pOperator->numOfExprs; ++k) {
H
Haojun Liao 已提交
809
    if (functionNeedToExecute(&pCtx[k])) {
810 811 812 813
      pCtx[k].startTs = startTs;
      // this can be set during create the struct
      // todo add a dummy funtion to avoid process check
      if (pCtx[k].fpSet.process != NULL) {
814 815 816 817 818 819 820
        int32_t code = pCtx[k].fpSet.process(&pCtx[k]);
        if (code != TSDB_CODE_SUCCESS) {
          qError("%s call aggregate function error happens, code : %s",
                 GET_TASKID(pOperator->pTaskInfo), tstrerror(code));
          pOperator->pTaskInfo->code = code;
          longjmp(pOperator->pTaskInfo->env, code);
        }
821
      }
822 823 824 825
    }
  }
}

H
Haojun Liao 已提交
826
static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, SArray* pPseudoList) {
dengyihao's avatar
dengyihao 已提交
827
  size_t num = (pPseudoList != NULL) ? taosArrayGetSize(pPseudoList) : 0;
H
Haojun Liao 已提交
828 829 830 831 832
  for (int32_t i = 0; i < num; ++i) {
    pCtx[i].pOutput = taosArrayGet(pResult->pDataBlock, i);
  }
}

833
int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
X
Xiaoyu Wang 已提交
834
                              int32_t numOfOutput, SArray* pPseudoList) {
H
Haojun Liao 已提交
835
  setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
H
Haojun Liao 已提交
836
  pResult->info.groupId = pSrcBlock->info.groupId;
H
Haojun Liao 已提交
837

dengyihao's avatar
dengyihao 已提交
838 839
  // if the source equals to the destination, it is to create a new column as the result of scalar function or some
  // operators.
840 841
  bool createNewColModel = (pResult == pSrcBlock);

842 843
  int32_t numOfRows = 0;

844
  for (int32_t k = 0; k < numOfOutput; ++k) {
dengyihao's avatar
dengyihao 已提交
845
    int32_t         outputSlotId = pExpr[k].base.resSchema.slotId;
846 847
    SqlFunctionCtx* pfCtx = &pCtx[k];

L
Liu Jicong 已提交
848
    if (pExpr[k].pExpr->nodeType == QUERY_NODE_COLUMN) {  // it is a project query
849
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
850
      if (pResult->info.rows > 0 && !createNewColModel) {
X
Xiaoyu Wang 已提交
851 852
        colDataMergeCol(pColInfoData, pResult->info.rows, &pResult->info.capacity, pfCtx->input.pData[0],
                        pfCtx->input.numOfRows);
853 854 855
      } else {
        colDataAssign(pColInfoData, pfCtx->input.pData[0], pfCtx->input.numOfRows);
      }
856

857
      numOfRows = pfCtx->input.numOfRows;
858
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE) {
859
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
860

dengyihao's avatar
dengyihao 已提交
861
      int32_t offset = createNewColModel ? 0 : pResult->info.rows;
862
      for (int32_t i = 0; i < pSrcBlock->info.rows; ++i) {
dengyihao's avatar
dengyihao 已提交
863 864 865
        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);
866
      }
867 868

      numOfRows = pSrcBlock->info.rows;
H
Haojun Liao 已提交
869
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_OPERATOR) {
870 871 872
      SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
      taosArrayPush(pBlockList, &pSrcBlock);

873
      SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
874
      SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
875

876
      SScalarParam dest = {.columnData = &idata};
877 878
      scalarCalculate(pExpr[k].pExpr->_optrRoot.pRootNode, pBlockList, &dest);

dengyihao's avatar
dengyihao 已提交
879
      int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
880
      colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
881 882

      numOfRows = dest.numOfRows;
883 884
      taosArrayDestroy(pBlockList);
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_FUNCTION) {
885
      ASSERT(!fmIsAggFunc(pfCtx->functionId));
886

887
      if (fmIsPseudoColumnFunc(pfCtx->functionId)) {
H
Haojun Liao 已提交
888
        // do nothing
889
      } else if (fmIsNonstandardSQLFunc(pfCtx->functionId)) {
dengyihao's avatar
dengyihao 已提交
890
        SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[k]);
891
        pfCtx->fpSet.init(&pCtx[k], pResInfo);
892

893
        pfCtx->pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
dengyihao's avatar
dengyihao 已提交
894
        pfCtx->offset = createNewColModel ? 0 : pResult->info.rows;  // set the start offset
H
Haojun Liao 已提交
895

896
        // set the timestamp(_rowts) output buffer
897 898
        if (taosArrayGetSize(pPseudoList) > 0) {
          int32_t* outputColIndex = taosArrayGet(pPseudoList, 0);
899
          pfCtx->pTsOutput = (SColumnInfoData*)pCtx[*outputColIndex].pOutput;
900
        }
H
Haojun Liao 已提交
901

902
        numOfRows = pfCtx->fpSet.process(pfCtx);
H
Haojun Liao 已提交
903 904 905
      } else {
        SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
        taosArrayPush(pBlockList, &pSrcBlock);
G
Ganlin Zhao 已提交
906

907
        SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
908
        SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
H
Haojun Liao 已提交
909

910
        SScalarParam dest = {.columnData = &idata};
X
Xiaoyu Wang 已提交
911
        int32_t      code = scalarCalculate((SNode*)pExpr[k].pExpr->_function.pFunctNode, pBlockList, &dest);
912 913 914 915
        if (code != TSDB_CODE_SUCCESS) {
          taosArrayDestroy(pBlockList);
          return code;
        }
916

dengyihao's avatar
dengyihao 已提交
917
        int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
918
        colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
919 920

        numOfRows = dest.numOfRows;
H
Haojun Liao 已提交
921 922
        taosArrayDestroy(pBlockList);
      }
923
    } else {
924
      ASSERT(0);
925 926
    }
  }
927

928 929 930
  if (!createNewColModel) {
    pResult->info.rows += numOfRows;
  }
931 932

  return TSDB_CODE_SUCCESS;
933 934
}

935 936 937
static void setResultRowKey(SResultRow* pResultRow, char* pData, int16_t type) {
  if (IS_VAR_DATA_TYPE(type)) {
    // todo disable this
938

939 940 941 942 943 944 945 946 947
    //    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);
948

949 950
    pResultRow->win.skey = v;
    pResultRow->win.ekey = v;
951 952 953
  }
}

954 955 956 957 958
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;
959

960 961 962
  SResultRow* pResultRow =
      doSetResultOutBufByKey(pBuf, pResultRowInfo, (char*)pData, bytes, true, groupId, pTaskInfo, false, pAggSup);
  assert(pResultRow != NULL);
963

964
  setResultRowKey(pResultRow, pData, type);
965
  setResultRowInitCtx(pResultRow, pCtx, numOfCols, binfo->rowCellInfoOffset);
966
  return TSDB_CODE_SUCCESS;
967 968
}

969 970
static bool functionNeedToExecute(SqlFunctionCtx* pCtx) {
  struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
971

972 973 974 975 976
  // in case of timestamp column, always generated results.
  int32_t functionId = pCtx->functionId;
  if (functionId == -1) {
    return false;
  }
977

978
  if (isRowEntryCompleted(pResInfo)) {
979 980 981
    return false;
  }

982 983 984 985 986 987 988
  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;
989 990
  }

991 992 993 994 995
  // 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;
  //  }
996 997 998 999

  return true;
}

1000 1001 1002 1003 1004 1005 1006
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;
    }
1007

1008 1009 1010
    // Set the correct column info (data type and bytes)
    pInput->pData[paramIndex]->info.type = type;
    pInput->pData[paramIndex]->info.bytes = tDataTypes[type].bytes;
1011
  }
H
Haojun Liao 已提交
1012

1013 1014 1015 1016 1017 1018
  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;
1019 1020
    }
  } else {
1021
    da = pInput->pColumnDataAgg[paramIndex];
1022 1023
  }

1024
  ASSERT(!IS_VAR_DATA_TYPE(type));
1025

1026 1027 1028 1029 1030 1031
  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};
1032

1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
    *(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
1045
  } else {
1046
    ASSERT(0);
1047 1048
  }

1049 1050
  return TSDB_CODE_SUCCESS;
}
1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061

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;

1062 1063
    for (int32_t j = 0; j < pExprInfo->base.numOfParams; ++j) {
      SFunctParam* pFuncParam = &pExprInfo->base.pParam[j];
1064

1065 1066
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
1067 1068 1069 1070
        pInput->pColumnDataAgg[j] = pBlock->pBlockAgg[slotId];
        if (pInput->pColumnDataAgg[j] == NULL) {
          pInput->colDataAggIsSet = false;
        }
1071 1072 1073 1074

        // 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);
1075 1076
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
        doCreateConstantValColumnAggInfo(pInput, pFuncParam, pFuncParam->param.nType, j, pBlock->info.rows);
1077 1078
      }
    }
1079
  } else {
1080
    pInput->colDataAggIsSet = false;
1081 1082 1083
  }

  // set the statistics data for primary time stamp column
1084 1085 1086 1087 1088
  //  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;
  //  }
1089 1090 1091
}

// set the output buffer for the selectivity + tag query
1092
static int32_t setSelectValueColumnInfo(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1093 1094
  int32_t num = 0;

H
Haojun Liao 已提交
1095
  SqlFunctionCtx*  p = NULL;
1096 1097
  SqlFunctionCtx** pValCtx = taosMemoryCalloc(numOfOutput, POINTER_BYTES);
  if (pValCtx == NULL) {
1098 1099 1100 1101
    return TSDB_CODE_QRY_OUT_OF_MEMORY;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
1102 1103
    if (strcmp(pCtx[i].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
      pValCtx[num++] = &pCtx[i];
1104
    } else if (fmIsAggFunc(pCtx[i].functionId)) {
1105
      p = &pCtx[i];
1106
    }
1107 1108 1109 1110 1111 1112 1113 1114 1115 1116
//    if (functionId == FUNCTION_TAG_DUMMY || functionId == FUNCTION_TS_DUMMY) {
//      tagLen += pCtx[i].resDataInfo.bytes;
//      pTagCtx[num++] = &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
//    }
1117
  }
1118

1119
  if (p != NULL) {
1120
    p->subsidiaries.pCtx = pValCtx;
1121
    p->subsidiaries.num = num;
1122
  } else {
1123
    taosMemoryFreeClear(pValCtx);
1124 1125 1126 1127 1128
  }

  return TSDB_CODE_SUCCESS;
}

1129
SqlFunctionCtx* createSqlFunctionCtx(SExprInfo* pExprInfo, int32_t numOfOutput, int32_t** rowCellInfoOffset) {
L
Liu Jicong 已提交
1130
  SqlFunctionCtx* pFuncCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfOutput, sizeof(SqlFunctionCtx));
H
Haojun Liao 已提交
1131 1132 1133 1134
  if (pFuncCtx == NULL) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
1135
  *rowCellInfoOffset = taosMemoryCalloc(numOfOutput, sizeof(int32_t));
H
Haojun Liao 已提交
1136
  if (*rowCellInfoOffset == 0) {
wafwerar's avatar
wafwerar 已提交
1137
    taosMemoryFreeClear(pFuncCtx);
H
Haojun Liao 已提交
1138 1139 1140 1141
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
1142
    SExprInfo* pExpr = &pExprInfo[i];
H
Haojun Liao 已提交
1143

L
Liu Jicong 已提交
1144
    SExprBasicInfo* pFunct = &pExpr->base;
H
Haojun Liao 已提交
1145
    SqlFunctionCtx* pCtx = &pFuncCtx[i];
H
Haojun Liao 已提交
1146

1147
    pCtx->functionId = -1;
1148
    pCtx->curBufPage = -1;
L
Liu Jicong 已提交
1149
    pCtx->pExpr = pExpr;
1150

H
Haojun Liao 已提交
1151
    if (pExpr->pExpr->nodeType == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
1152
      SFuncExecEnv env = {0};
H
Haojun Liao 已提交
1153 1154
      pCtx->functionId = pExpr->pExpr->_function.pFunctNode->funcId;

H
Haojun Liao 已提交
1155
      if (fmIsAggFunc(pCtx->functionId) || fmIsNonstandardSQLFunc(pCtx->functionId)) {
1156 1157 1158 1159
        bool isUdaf = fmIsUserDefinedFunc(pCtx->functionId);
        if (!isUdaf) {
          fmGetFuncExecFuncs(pCtx->functionId, &pCtx->fpSet);
        } else {
1160
          char* udfName = pExpr->pExpr->_function.pFunctNode->functionName;
1161 1162 1163
          strncpy(pCtx->udfName, udfName, strlen(udfName));
          fmGetUdafExecFuncs(pCtx->functionId, &pCtx->fpSet);
        }
1164 1165 1166
        pCtx->fpSet.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
      } else {
        fmGetScalarFuncExecFuncs(pCtx->functionId, &pCtx->sfp);
1167 1168 1169
        if (pCtx->sfp.getEnv != NULL) {
          pCtx->sfp.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
        }
1170
      }
H
Haojun Liao 已提交
1171
      pCtx->resDataInfo.interBufSize = env.calcMemSize;
1172 1173
    } else if (pExpr->pExpr->nodeType == QUERY_NODE_COLUMN || pExpr->pExpr->nodeType == QUERY_NODE_OPERATOR ||
               pExpr->pExpr->nodeType == QUERY_NODE_VALUE) {
H
Haojun Liao 已提交
1174
      // for simple column, the result buffer needs to hold at least one element.
1175
      pCtx->resDataInfo.interBufSize = pFunct->resSchema.bytes;
H
Haojun Liao 已提交
1176
    }
H
Haojun Liao 已提交
1177

H
Haojun Liao 已提交
1178
    pCtx->input.numOfInputCols = pFunct->numOfParams;
wafwerar's avatar
wafwerar 已提交
1179 1180
    pCtx->input.pData = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
    pCtx->input.pColumnDataAgg = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
H
Haojun Liao 已提交
1181

1182
    pCtx->pTsOutput = NULL;
L
Liu Jicong 已提交
1183
    pCtx->resDataInfo.bytes = pFunct->resSchema.bytes;
1184
    pCtx->resDataInfo.type = pFunct->resSchema.type;
L
Liu Jicong 已提交
1185
    pCtx->order = TSDB_ORDER_ASC;
1186
    pCtx->start.key = INT64_MIN;
L
Liu Jicong 已提交
1187
    pCtx->end.key = INT64_MIN;
1188
    pCtx->numOfParams = pExpr->base.numOfParams;
H
Haojun Liao 已提交
1189

1190
    pCtx->param = pFunct->pParam;
dengyihao's avatar
dengyihao 已提交
1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224
    //    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
1225 1226
  }

L
Liu Jicong 已提交
1227 1228 1229
  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
1230
  }
H
Haojun Liao 已提交
1231

1232
  setSelectValueColumnInfo(pFuncCtx, numOfOutput);
H
Haojun Liao 已提交
1233 1234 1235
  return pFuncCtx;
}

1236
static void* destroySqlFunctionCtx(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1237 1238 1239 1240 1241 1242
  if (pCtx == NULL) {
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
    for (int32_t j = 0; j < pCtx[i].numOfParams; ++j) {
1243
      taosVariantDestroy(&pCtx[i].param[j].param);
1244 1245 1246
    }

    taosVariantDestroy(&pCtx[i].tag);
1247
    taosMemoryFreeClear(pCtx[i].subsidiaries.pCtx);
H
Haojun Liao 已提交
1248 1249
    taosMemoryFree(pCtx[i].input.pData);
    taosMemoryFree(pCtx[i].input.pColumnDataAgg);
1250 1251
  }

wafwerar's avatar
wafwerar 已提交
1252
  taosMemoryFreeClear(pCtx);
1253 1254 1255
  return NULL;
}

L
Liu Jicong 已提交
1256
bool isTaskKilled(SExecTaskInfo* pTaskInfo) {
1257 1258
  // query has been executed more than tsShellActivityTimer, and the retrieve has not arrived
  // abort current query execution.
L
Liu Jicong 已提交
1259 1260
  if (pTaskInfo->owner != 0 &&
      ((taosGetTimestampSec() - pTaskInfo->cost.start / 1000) > 10 * getMaximumIdleDurationSec())
1261 1262
      /*(!needBuildResAfterQueryComplete(pTaskInfo))*/) {
    assert(pTaskInfo->cost.start != 0);
L
Liu Jicong 已提交
1263 1264 1265
    //    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;
1266 1267 1268 1269 1270
  }

  return false;
}

L
Liu Jicong 已提交
1271
void setTaskKilled(SExecTaskInfo* pTaskInfo) { pTaskInfo->code = TSDB_CODE_TSC_QUERY_CANCELLED; }
1272

L
Liu Jicong 已提交
1273
static bool isCachedLastQuery(STaskAttr* pQueryAttr) {
1274 1275 1276 1277 1278 1279 1280 1281 1282
  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;
  }

1283 1284
  int32_t order = TSDB_ORDER_ASC;
  if (order != TSDB_ORDER_DESC || !TSWINDOW_IS_EQUAL(pQueryAttr->window, TSWINDOW_DESC_INITIALIZER)) {
1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303
    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
1304
// todo refactor : return window
1305
void getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int64_t key, STimeWindow* win) {
H
Haojun Liao 已提交
1306
  win->skey = taosTimeTruncate(key, pInterval, precision);
1307 1308

  /*
H
Haojun Liao 已提交
1309
   * if the realSkey > INT64_MAX - pInterval->interval, the query duration between
1310 1311
   * realSkey and realEkey must be less than one interval.Therefore, no need to adjust the query ranges.
   */
1312 1313
  win->ekey = taosTimeAdd(win->skey, pInterval->interval, pInterval->intervalUnit, precision) - 1;
  if (win->ekey < win->skey) {
1314 1315 1316 1317
    win->ekey = INT64_MAX;
  }
}

L
Liu Jicong 已提交
1318
static int32_t updateBlockLoadStatus(STaskAttr* pQuery, int32_t status) {
1319 1320 1321
  bool hasFirstLastFunc = false;
  bool hasOtherFunc = false;

1322
  if (status == BLK_DATA_DATA_LOAD || status == BLK_DATA_FILTEROUT) {
1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
    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;
    }
  }

1341
  if (hasFirstLastFunc && status == BLK_DATA_NOT_LOAD) {
L
Liu Jicong 已提交
1342
    if (!hasOtherFunc) {
1343
      return BLK_DATA_FILTEROUT;
1344
    } else {
1345
      return BLK_DATA_DATA_LOAD;
1346 1347 1348 1349 1350 1351
    }
  }

  return status;
}

L
Liu Jicong 已提交
1352 1353
// static void updateDataCheckOrder(SQInfo *pQInfo, SQueryTableReq* pQueryMsg, bool stableQuery) {
//   STaskAttr* pQueryAttr = pQInfo->runtimeEnv.pQueryAttr;
H
Haojun Liao 已提交
1354
//
L
Liu Jicong 已提交
1355 1356 1357 1358
//   // 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
1359
//
L
Liu Jicong 已提交
1360 1361 1362 1363 1364
//   // 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
1365
//
L
Liu Jicong 已提交
1366 1367
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     if (pQueryAttr->window.skey > pQueryAttr->window.ekey) {
wafwerar's avatar
wafwerar 已提交
1368
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1369
//     }
H
Haojun Liao 已提交
1370
//
L
Liu Jicong 已提交
1371 1372 1373
//     pQueryAttr->needReverseScan = false;
//     return;
//   }
H
Haojun Liao 已提交
1374
//
L
Liu Jicong 已提交
1375 1376 1377
//   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
1378
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1379
//     }
H
Haojun Liao 已提交
1380
//
L
Liu Jicong 已提交
1381 1382 1383 1384
//     pQueryAttr->needReverseScan = false;
//     doUpdateLastKey(pQueryAttr);
//     return;
//   }
H
Haojun Liao 已提交
1385
//
L
Liu Jicong 已提交
1386 1387 1388 1389 1390 1391
//   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
1392
//
L
Liu Jicong 已提交
1393 1394 1395
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     return;
//   }
H
Haojun Liao 已提交
1396
//
L
Liu Jicong 已提交
1397 1398 1399 1400
//   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
1401 1402
////               pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey);
//
wafwerar's avatar
wafwerar 已提交
1403
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1404 1405 1406 1407 1408 1409 1410 1411 1412 1413
//        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
1414
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426
//        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
1427 1428
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1429
//
wafwerar's avatar
wafwerar 已提交
1430
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1431 1432 1433 1434 1435 1436 1437 1438
//          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
1439 1440
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1441
//
wafwerar's avatar
wafwerar 已提交
1442
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1443 1444 1445 1446 1447 1448 1449 1450 1451
//          doUpdateLastKey(pQueryAttr);
//        }
//
//        pQueryAttr->order.order = TSDB_ORDER_DESC;
//        pQueryAttr->needReverseScan = false;
//      }
//    }
//  }
//}
1452

H
Haojun Liao 已提交
1453 1454
static void getIntermediateBufInfo(STaskRuntimeEnv* pRuntimeEnv, int32_t* ps, int32_t* rowsize) {
  STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
L
Liu Jicong 已提交
1455
  int32_t    MIN_ROWS_PER_PAGE = 4;
1456

L
Liu Jicong 已提交
1457 1458
  *rowsize = (int32_t)(pQueryAttr->resultRowSize *
                       getRowNumForMultioutput(pQueryAttr, pQueryAttr->topBotQuery, pQueryAttr->stableQuery));
1459 1460 1461 1462
  int32_t overhead = sizeof(SFilePage);

  // one page contains at least two rows
  *ps = DEFAULT_INTERN_BUF_PAGE_SIZE;
L
Liu Jicong 已提交
1463
  while (((*rowsize) * MIN_ROWS_PER_PAGE) > (*ps) - overhead) {
1464 1465 1466 1467
    *ps = ((*ps) << 1u);
  }
}

L
Liu Jicong 已提交
1468 1469 1470
// static FORCE_INLINE bool doFilterByBlockStatistics(STaskRuntimeEnv* pRuntimeEnv, SDataStatis *pDataStatis,
// SqlFunctionCtx *pCtx, int32_t numOfRows) {
//   STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
1471
//
L
Liu Jicong 已提交
1472 1473 1474
//   if (pDataStatis == NULL || pQueryAttr->pFilters == NULL) {
//     return true;
//   }
1475
//
L
Liu Jicong 已提交
1476 1477
//   return filterRangeExecute(pQueryAttr->pFilters, pDataStatis, pQueryAttr->numOfCols, numOfRows);
// }
1478

H
Haojun Liao 已提交
1479
static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
1480 1481
  STimeWindow w = {0};

dengyihao's avatar
dengyihao 已提交
1482 1483
  TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
  TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
1484

1485
  if (true) {
L
Liu Jicong 已提交
1486
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
1487 1488 1489 1490 1491 1492
    assert(w.ekey >= pBlockInfo->window.skey);

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

L
Liu Jicong 已提交
1493 1494
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1495 1496 1497 1498 1499 1500 1501 1502 1503 1504
      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
1505
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
1506 1507 1508 1509 1510 1511
    assert(w.skey <= pBlockInfo->window.ekey);

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

L
Liu Jicong 已提交
1512 1513
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
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 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577
      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
1578
static void         doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes);
1579 1580

static uint32_t doFilterByBlockTimeWindow(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock) {
H
Haojun Liao 已提交
1581
  SqlFunctionCtx* pCtx = pTableScanInfo->pCtx;
1582
  uint32_t        status = BLK_DATA_NOT_LOAD;
1583 1584 1585 1586

  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
1587
    int32_t colId = pTableScanInfo->pExpr[i].base.pParam[0].pCol->colId;
1588 1589 1590

    // group by + first/last should not apply the first/last block filter
    if (functionId < 0) {
1591
      status |= BLK_DATA_DATA_LOAD;
1592 1593
      return status;
    } else {
L
Liu Jicong 已提交
1594
      //      status |= aAggs[functionId].dataReqFunc(&pTableScanInfo->pCtx[i], &pBlock->info.window, colId);
1595
      //      if ((status & BLK_DATA_DATA_LOAD) == BLK_DATA_DATA_LOAD) {
L
Liu Jicong 已提交
1596 1597
      //        return status;
      //      }
1598 1599 1600 1601 1602 1603
    }
  }

  return status;
}

L
Liu Jicong 已提交
1604 1605
int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                              uint32_t* status) {
1606
  *status = BLK_DATA_NOT_LOAD;
1607

H
Haojun Liao 已提交
1608
  pBlock->pDataBlock = NULL;
L
Liu Jicong 已提交
1609
  pBlock->pBlockAgg = NULL;
H
Haojun Liao 已提交
1610

L
Liu Jicong 已提交
1611 1612
  //  int64_t groupId = pRuntimeEnv->current->groupIndex;
  //  bool    ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
1613

H
Haojun Liao 已提交
1614
  STaskCostInfo* pCost = &pTaskInfo->cost;
1615

1616 1617
//  pCost->totalBlocks += 1;
//  pCost->totalRows += pBlock->info.rows;
H
Haojun Liao 已提交
1618
#if 0
1619 1620 1621
  // 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
1622
      (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && overlapWithTimeWindow(pTaskInfo, &pBlock->info))) {
1623
    (*status) = BLK_DATA_DATA_LOAD;
1624 1625 1626
  }

  // check if this data block is required to load
1627
  if ((*status) != BLK_DATA_DATA_LOAD) {
1628 1629 1630 1631 1632 1633 1634
    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
1635
      bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658
      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 {
1659
      (*status) = BLK_DATA_DATA_LOAD;
1660 1661 1662 1663
    }
  }

  SDataBlockInfo* pBlockInfo = &pBlock->info;
H
Haojun Liao 已提交
1664
//  *status = updateBlockLoadStatus(pRuntimeEnv->pQueryAttr, *status);
1665

1666
  if ((*status) == BLK_DATA_NOT_LOAD || (*status) == BLK_DATA_FILTEROUT) {
1667 1668
    //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//           pBlockInfo->window.ekey, pBlockInfo->rows);
1669
    pCost->skipBlocks += 1;
1670
  } else if ((*status) == BLK_DATA_SMA_LOAD) {
1671 1672
    // this function never returns error?
    pCost->loadBlockStatis += 1;
1673
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1674 1675

    if (pBlock->pBlockAgg == NULL) {  // data block statistics does not exist, load data block
1676
//      pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1677 1678 1679
      pCost->totalCheckedRows += pBlock->info.rows;
    }
  } else {
1680
    assert((*status) == BLK_DATA_DATA_LOAD);
1681 1682 1683

    // load the data block statistics to perform further filter
    pCost->loadBlockStatis += 1;
1684
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1685 1686 1687 1688 1689 1690

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

H
Haojun Liao 已提交
1691
          bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708
          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
1709
            pCost->skipBlocks += 1;
1710 1711
            //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId,
//                   pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
1712
            (*status) = BLK_DATA_FILTEROUT;
1713 1714 1715 1716 1717 1718 1719 1720
            return TSDB_CODE_SUCCESS;
          }
        }
      }
    }

    // current block has been discard due to filter applied
//    if (!doFilterByBlockStatistics(pRuntimeEnv, pBlock->pBlockAgg, pTableScanInfo->pCtx, pBlockInfo->rows)) {
1721
//      pCost->skipBlocks += 1;
1722 1723
//      qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//             pBlockInfo->window.ekey, pBlockInfo->rows);
1724
//      (*status) = BLK_DATA_FILTEROUT;
1725 1726 1727 1728 1729
//      return TSDB_CODE_SUCCESS;
//    }

    pCost->totalCheckedRows += pBlockInfo->rows;
    pCost->loadBlocks += 1;
1730
//    pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1731 1732 1733 1734 1735 1736 1737
//    if (pBlock->pDataBlock == NULL) {
//      return terrno;
//    }

//    if (pQueryAttr->pFilters != NULL) {
//      filterSetColFieldData(pQueryAttr->pFilters, pBlock->info.numOfCols, pBlock->pDataBlock);
//    }
1738

1739 1740 1741 1742
//    if (pQueryAttr->pFilters != NULL || pRuntimeEnv->pTsBuf != NULL) {
//      filterColRowsInDataBlock(pRuntimeEnv, pBlock, ascQuery);
//    }
  }
H
Haojun Liao 已提交
1743
#endif
1744 1745 1746 1747
  return TSDB_CODE_SUCCESS;
}

/*
H
Haojun Liao 已提交
1748
 * set tag value in SqlFunctionCtx
1749 1750
 * e.g.,tag information into input buffer
 */
L
Liu Jicong 已提交
1751
static void doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes) {
1752 1753 1754
  taosVariantDestroy(tag);

  char* val = NULL;
L
Liu Jicong 已提交
1755 1756 1757 1758 1759 1760
  //  if (tagColId == TSDB_TBNAME_COLUMN_INDEX) {
  //    val = tsdbGetTableName(pTable);
  //    assert(val != NULL);
  //  } else {
  //    val = tsdbGetTableTagVal(pTable, tagColId, type, bytes);
  //  }
1761 1762 1763 1764 1765 1766 1767 1768

  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
1769
    int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
1770
    taosVariantCreateFromBinary(tag, varDataVal(val), len, type);
L
Liu Jicong 已提交
1771
    // taosVariantCreateFromBinary(tag, varDataVal(val), varDataLen(val), type);
1772 1773 1774 1775 1776 1777 1778 1779
  } 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
1780
  for (int32_t i = 0; i < numOfTags; ++i) {
1781 1782 1783 1784 1785 1786 1787 1788
    if (pTagColList[i].colId == colId) {
      return &pTagColList[i];
    }
  }

  return NULL;
}

L
Liu Jicong 已提交
1789 1790
void setTagValue(SOperatorInfo* pOperatorInfo, void* pTable, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
  SExprInfo* pExpr = pOperatorInfo->pExpr;
1791
  SExprInfo* pExprInfo = &pExpr[0];
L
Liu Jicong 已提交
1792
  int32_t    functionId = getExprFunctionId(pExprInfo);
1793
#if 0
1794 1795 1796
  if (pQueryAttr->numOfOutput == 1 && functionId == FUNCTION_TS_COMP && pQueryAttr->stableQuery) {
    assert(pExprInfo->base.numOfParams == 1);

L
Liu Jicong 已提交
1797
    //    int16_t      tagColId = (int16_t)pExprInfo->base.param[0].i;
H
Haojun Liao 已提交
1798
    int16_t      tagColId = -1;
1799 1800 1801
    SColumnInfo* pColInfo = doGetTagColumnInfoById(pQueryAttr->tagColList, pQueryAttr->numOfTags, tagColId);

    doSetTagValueInParam(pTable, tagColId, &pCtx[0].tag, pColInfo->type, pColInfo->bytes);
H
Haojun Liao 已提交
1802

1803 1804 1805 1806 1807 1808 1809 1810 1811
  } 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
1812
      if (!TSDB_COL_IS_TAG(pLocalExprInfo->base.pParam[0].pCol->flag)) {
1813 1814 1815 1816
        continue;
      }

      // todo use tag column index to optimize performance
L
Liu Jicong 已提交
1817 1818
      doSetTagValueInParam(pTable, pLocalExprInfo->base.pParam[0].pCol->colId, &pCtx[idx].tag,
                           pLocalExprInfo->base.resSchema.type, pLocalExprInfo->base.resSchema.bytes);
1819

L
Liu Jicong 已提交
1820 1821 1822
      if (IS_NUMERIC_TYPE(pLocalExprInfo->base.resSchema.type) ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_BOOL ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_TIMESTAMP) {
1823 1824 1825 1826
        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
1827
        }
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837
      }

      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);
  }
1838
#endif
1839 1840
}

H
Haojun Liao 已提交
1841
void copyToSDataBlock(SSDataBlock* pBlock, int32_t* offset, SGroupResInfo* pGroupResInfo, SDiskbasedBuf* pResBuf) {
1842 1843 1844 1845 1846 1847 1848
  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
1849 1850 1851
      //      if ((code = mergeIntoGroupResult(&pGroupResInfo, pRuntimeEnv, offset)) != TSDB_CODE_SUCCESS) {
      return;
      //      }
1852 1853
    }

L
Liu Jicong 已提交
1854
    //    doCopyToSDataBlock(pResBuf, pGroupResInfo, TSDB_ORDER_ASC, pBlock, );
1855 1856 1857 1858 1859 1860 1861 1862 1863 1864

    // 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
1865 1866 1867
    //    if (pBlock->info.rows >= threshold) {
    //      break;
    //    }
1868 1869 1870
  }
}

L
Liu Jicong 已提交
1871
static void updateTableQueryInfoForReverseScan(STableQueryInfo* pTableQueryInfo) {
1872 1873 1874 1875
  if (pTableQueryInfo == NULL) {
    return;
  }

wafwerar's avatar
wafwerar 已提交
1876
  //  TSWAP(pTableQueryInfo->win.skey, pTableQueryInfo->win.ekey);
L
Liu Jicong 已提交
1877
  //  pTableQueryInfo->lastKey = pTableQueryInfo->win.skey;
1878

L
Liu Jicong 已提交
1879 1880
  //  SWITCH_ORDER(pTableQueryInfo->cur.order);
  //  pTableQueryInfo->cur.vgroupIndex = -1;
1881 1882

  // set the index to be the end slot of result rows array
dengyihao's avatar
dengyihao 已提交
1883 1884 1885 1886 1887 1888
  //  SResultRowInfo* pResultRowInfo = &pTableQueryInfo->resInfo;
  //  if (pResultRowInfo->size > 0) {
  //    pResultRowInfo->curPos = pResultRowInfo->size - 1;
  //  } else {
  //    pResultRowInfo->curPos = -1;
  //  }
1889 1890
}

H
Haojun Liao 已提交
1891
void initResultRow(SResultRow* pResultRow) {
X
Xiaoyu Wang 已提交
1892
  //  pResultRow->pEntryInfo = (struct SResultRowEntryInfo*)((char*)pResultRow + sizeof(SResultRow));
1893 1894 1895 1896 1897
}

/*
 * 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
1898 1899 1900
 * +------------+-----------------result column 1------------+------------------result column 2-----------+
 * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
 * +------------+--------------------------------------------+--------------------------------------------+
1901 1902
 *           offset[0]                                  offset[1]                                   offset[2]
 */
1903
// TODO refactor: some function move away
X
Xiaoyu Wang 已提交
1904 1905
void setFunctionResultOutput(SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage, int32_t numOfExprs,
                             SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
1906 1907 1908
  SqlFunctionCtx* pCtx = pInfo->pCtx;
  SSDataBlock*    pDataBlock = pInfo->pRes;
  int32_t*        rowCellInfoOffset = pInfo->rowCellInfoOffset;
H
Haojun Liao 已提交
1909

H
Haojun Liao 已提交
1910
  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
H
Haojun Liao 已提交
1911
  initResultRowInfo(pResultRowInfo, 16);
H
Haojun Liao 已提交
1912

L
Liu Jicong 已提交
1913 1914
  int64_t     tid = 0;
  int64_t     groupId = 0;
1915 1916
  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
                                            pTaskInfo, false, pSup);
H
Haojun Liao 已提交
1917

H
Haojun Liao 已提交
1918
  ASSERT(pDataBlock->info.numOfCols == numOfExprs);
H
Haojun Liao 已提交
1919 1920 1921 1922
  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
1923
    pCtx[i].resultInfo = pEntry;
H
Haojun Liao 已提交
1924 1925 1926
    pCtx[i].currentStage = stage;

    // set the timestamp output buffer for top/bottom/diff query
L
Liu Jicong 已提交
1927 1928
    //    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
1929
    //      if (i > 0) pCtx[i].pTsOutput = pCtx[i-1].pOutput;
L
Liu Jicong 已提交
1930
    //    }
H
Haojun Liao 已提交
1931 1932 1933 1934 1935
  }

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

L
Liu Jicong 已提交
1936
void updateOutputBuf(SOptrBasicInfo* pBInfo, int32_t* bufCapacity, int32_t numOfInputRows) {
1937 1938
  SSDataBlock* pDataBlock = pBInfo->pRes;

L
Liu Jicong 已提交
1939
  int32_t newSize = pDataBlock->info.rows + numOfInputRows + 5;  // extra output buffer
1940
  if ((*bufCapacity) < newSize) {
L
Liu Jicong 已提交
1941 1942
    for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
      SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1943

wafwerar's avatar
wafwerar 已提交
1944
      char* p = taosMemoryRealloc(pColInfo->pData, newSize * pColInfo->info.bytes);
1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957
      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
1958
    SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1959 1960 1961 1962 1963
    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
1964 1965
    if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF ||
        functionId == FUNCTION_DERIVATIVE) {
dengyihao's avatar
dengyihao 已提交
1966
      //      if (i > 0) pBInfo->pCtx[i].pTsOutput = pBInfo->pCtx[i - 1].pOutput;
1967 1968 1969 1970
    }
  }
}

H
Haojun Liao 已提交
1971
void copyTsColoum(SSDataBlock* pRes, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1972 1973
  bool    needCopyTs = false;
  int32_t tsNum = 0;
L
Liu Jicong 已提交
1974
  char*   src = NULL;
1975 1976 1977 1978
  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
1979 1980
      if (i > 0 && pCtx[i - 1].functionId == FUNCTION_TS_DUMMY) {
        SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i - 1);  // find ts data
1981 1982
        src = pColRes->pData;
      }
L
Liu Jicong 已提交
1983
    } else if (functionId == FUNCTION_TS_DUMMY) {
1984 1985 1986 1987 1988 1989 1990 1991 1992 1993
      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
1994
    if (functionId == FUNCTION_TS_DUMMY) {
1995 1996 1997 1998 1999 2000
      SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i);
      memcpy(pColRes->pData, src, pColRes->info.bytes * pRes->info.rows);
    }
  }
}

H
Haojun Liao 已提交
2001
void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
2002 2003
  for (int32_t j = 0; j < size; ++j) {
    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
dengyihao's avatar
dengyihao 已提交
2004 2005
    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
        fmIsScalarFunc(pCtx[j].functionId)) {
2006 2007 2008
      continue;
    }

H
Haojun Liao 已提交
2009
    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
2010 2011 2012
  }
}

L
Liu Jicong 已提交
2013
void setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status) {
2014
  if (status == TASK_NOT_COMPLETED) {
H
Haojun Liao 已提交
2015
    pTaskInfo->status = status;
2016 2017
  } else {
    // QUERY_NOT_COMPLETED is not compatible with any other status, so clear its position first
2018
    CLEAR_QUERY_STATUS(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
2019
    pTaskInfo->status |= status;
2020 2021 2022
  }
}

2023
// todo merged with the build group result.
2024 2025
void finalizeMultiTupleQueryResult(int32_t numOfOutput, SDiskbasedBuf* pBuf, SResultRowInfo* pResultRowInfo,
                                   int32_t* rowCellInfoOffset) {
2026 2027 2028
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
    SResultRowPosition* pPos = &pResultRowInfo->pPosition[i];

L
Liu Jicong 已提交
2029
    SFilePage*  bufPage = getBufPage(pBuf, pPos->pageId);
2030
    SResultRow* pRow = (SResultRow*)((char*)bufPage + pPos->offset);
H
Haojun Liao 已提交
2031 2032

    // TODO ignore the close status anyway.
dengyihao's avatar
dengyihao 已提交
2033 2034 2035
    //    if (!isResultRowClosed(pRow)) {
    //      continue;
    //    }
2036

2037
    for (int32_t j = 0; j < numOfOutput; ++j) {
2038
      struct SResultRowEntryInfo* pResInfo = getResultCell(pRow, j, rowCellInfoOffset);
H
Haojun Liao 已提交
2039
      if (!isRowEntryInitialized(pResInfo)) {
2040 2041 2042 2043 2044 2045
        continue;
      }

      if (pRow->numOfRows < pResInfo->numOfRes) {
        pRow->numOfRows = pResInfo->numOfRes;
      }
2046
    }
2047 2048 2049

    releaseBufPage(pBuf, bufPage);
  }
2050 2051
}

2052
STableQueryInfo* createTableQueryInfo(void* buf, STimeWindow win) {
L
Liu Jicong 已提交
2053
  STableQueryInfo* pTableQueryInfo = buf;
2054 2055 2056 2057
  pTableQueryInfo->lastKey = win.skey;
  return pTableQueryInfo;
}

L
Liu Jicong 已提交
2058
void destroyTableQueryInfoImpl(STableQueryInfo* pTableQueryInfo) {
2059 2060 2061 2062
  if (pTableQueryInfo == NULL) {
    return;
  }

L
Liu Jicong 已提交
2063
  //  taosVariantDestroy(&pTableQueryInfo->tag);
dengyihao's avatar
dengyihao 已提交
2064
  //  cleanupResultRowInfo(&pTableQueryInfo->resInfo);
2065 2066
}

2067
void setResultRowInitCtx(SResultRow* pResult, SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t* rowCellInfoOffset) {
2068 2069 2070 2071 2072 2073 2074
  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;
    }
2075 2076 2077 2078 2079

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

2080 2081 2082 2083 2084 2085
    if (!pResInfo->initialized) {
      if (pCtx[i].functionId != -1) {
        pCtx[i].fpSet.init(&pCtx[i], pResInfo);
      } else {
        pResInfo->initialized = true;
      }
2086 2087 2088 2089
    }
  }
}

2090 2091 2092 2093 2094 2095
void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock) {
  if (pFilterNode == NULL) {
    return;
  }

  SFilterInfo* filter = NULL;
H
Haojun Liao 已提交
2096

H
Haojun Liao 已提交
2097
  // todo move to the initialization function
H
Haojun Liao 已提交
2098
  int32_t code = filterInitFromNode((SNode*)pFilterNode, &filter, 0);
2099 2100 2101 2102 2103 2104

  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
2105
  filterFreeInfo(filter);
2106

2107
  SSDataBlock* px = createOneDataBlock(pBlock, false);
2108 2109
  blockDataEnsureCapacity(px, pBlock->info.rows);

H
Haojun Liao 已提交
2110
  // todo extract method
2111 2112 2113 2114
  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
2115 2116 2117 2118 2119 2120 2121 2122 2123
    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;
        }
2124

D
dapan1121 已提交
2125 2126 2127 2128 2129 2130
        if (colDataIsNull_s(pSrc, j)) {
          colDataAppendNULL(pDst, numOfRow);
        } else {
          colDataAppend(pDst, numOfRow, colDataGetData(pSrc, j), false);
        }
        numOfRow += 1;
H
Haojun Liao 已提交
2131
      }
D
dapan1121 已提交
2132 2133
    } else {
      numOfRow = 0;
2134 2135 2136 2137 2138 2139
    }

    *pSrc = *pDst;
  }

  pBlock->info.rows = numOfRow;
2140
  blockDataUpdateTsWindow(pBlock);
2141 2142
}

dengyihao's avatar
dengyihao 已提交
2143 2144
void doSetTableGroupOutputBuf(SAggOperatorInfo* pAggInfo, int32_t numOfOutput, uint64_t groupId,
                              SExecTaskInfo* pTaskInfo) {
2145 2146 2147
  // for simple group by query without interval, all the tables belong to one group result.
  int64_t uid = 0;

2148 2149
  SResultRowInfo* pResultRowInfo = &pAggInfo->binfo.resultRowInfo;
  SqlFunctionCtx* pCtx = pAggInfo->binfo.pCtx;
L
Liu Jicong 已提交
2150
  int32_t*        rowCellInfoOffset = pAggInfo->binfo.rowCellInfoOffset;
2151

2152
  SResultRow* pResultRow = doSetResultOutBufByKey(pAggInfo->aggSup.pResultBuf, pResultRowInfo, (char*)&groupId,
L
Liu Jicong 已提交
2153
                                                  sizeof(groupId), true, groupId, pTaskInfo, false, &pAggInfo->aggSup);
L
Liu Jicong 已提交
2154
  assert(pResultRow != NULL);
2155 2156 2157 2158 2159 2160

  /*
   * 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
2161 2162
    int32_t ret =
        addNewWindowResultBuf(pResultRow, pAggInfo->aggSup.pResultBuf, groupId, pAggInfo->binfo.pRes->info.rowSize);
2163 2164 2165 2166 2167
    if (ret != TSDB_CODE_SUCCESS) {
      return;
    }
  }

2168
  setResultRowInitCtx(pResultRow, pCtx, numOfOutput, rowCellInfoOffset);
2169 2170
}

H
Haojun Liao 已提交
2171 2172
void setExecutionContext(int32_t numOfOutput, uint64_t groupId, SExecTaskInfo* pTaskInfo, SAggOperatorInfo* pAggInfo) {
  if (pAggInfo->groupId != INT32_MIN && pAggInfo->groupId == groupId) {
2173 2174 2175
    return;
  }

H
Haojun Liao 已提交
2176
  doSetTableGroupOutputBuf(pAggInfo, numOfOutput, groupId, pTaskInfo);
2177 2178

  // record the current active group id
H
Haojun Liao 已提交
2179
  pAggInfo->groupId = groupId;
2180 2181 2182 2183 2184 2185 2186 2187 2188 2189
}

/**
 * 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
 */
2190
int32_t doCopyToSDataBlock(SExecTaskInfo* taskInfo, SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf, SGroupResInfo* pGroupResInfo,
2191
                           int32_t orderType, int32_t* rowCellOffset, SqlFunctionCtx* pCtx) {
2192
  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
L
Liu Jicong 已提交
2193
  int32_t numOfResult = pBlock->info.rows;  // there are already exists result rows
2194 2195

  int32_t start = 0;
2196
  int32_t step = 1;
2197

L
Liu Jicong 已提交
2198
  // qDebug("QInfo:0x%"PRIx64" start to copy data from windowResInfo to output buf", GET_TASKID(pRuntimeEnv));
2199 2200 2201 2202 2203 2204 2205 2206 2207
  assert(orderType == TSDB_ORDER_ASC || orderType == TSDB_ORDER_DESC);

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

  for (int32_t i = start; (i < numOfRows) && (i >= 0); i += step) {
L
Liu Jicong 已提交
2208 2209
    SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
    SFilePage*  page = getBufPage(pBuf, pPos->pos.pageId);
2210

2211
    SResultRow* pRow = (SResultRow*)((char*)page + pPos->pos.offset);
2212 2213 2214 2215 2216
    if (pRow->numOfRows == 0) {
      pGroupResInfo->index += 1;
      continue;
    }

2217
    // TODO copy multiple rows?
2218
    int32_t numOfRowsToCopy = pRow->numOfRows;
2219
    if (numOfResult + numOfRowsToCopy >= pBlock->info.capacity) {
2220 2221 2222 2223 2224 2225
      break;
    }

    pGroupResInfo->index += 1;

    for (int32_t j = 0; j < pBlock->info.numOfCols; ++j) {
2226 2227
      int32_t slotId = pExprInfo[j].base.resSchema.slotId;

2228
      pCtx[j].resultInfo = getResultCell(pRow, j, rowCellOffset);
2229 2230 2231
      if (pCtx[j].fpSet.finalize) {
        int32_t code = TSDB_CODE_SUCCESS;
        code = pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
2232
        if (TAOS_FAILED(code)) {
2233 2234 2235 2236
          qError("%s build result data block error, code %s", GET_TASKID(taskInfo), tstrerror(code));
          taskInfo->code = code;
          longjmp(taskInfo->env, code);
        }
2237 2238
      } else if (strcmp(pCtx[j].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
        // do nothing, todo refactor
2239
      } else {
2240 2241 2242 2243 2244 2245 2246
        // expand the result into multiple rows. E.g., _wstartts, top(k, 20)
        // the _wstartts needs to copy to 20 following rows, since the results of top-k expands to 20 different rows.
          SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, slotId);
          char* in = GET_ROWCELL_INTERBUF(pCtx[j].resultInfo);
          for(int32_t k = 0; k < pRow->numOfRows; ++k) {
            colDataAppend(pColInfoData, pBlock->info.rows + k, in, pCtx[j].resultInfo->isNullRes);
          }
2247
      }
2248 2249
    }

2250 2251
    releaseBufPage(pBuf, page);

2252 2253
    pBlock->info.rows += pRow->numOfRows;
    if (pBlock->info.rows >= pBlock->info.capacity) {  // output buffer is full
2254 2255 2256 2257
      break;
    }
  }

L
Liu Jicong 已提交
2258
  // qDebug("QInfo:0x%"PRIx64" copy data to query buf completed", GET_TASKID(pRuntimeEnv));
2259
  blockDataUpdateTsWindow(pBlock);
2260 2261 2262
  return 0;
}

2263
void doBuildResultDatablock(SExecTaskInfo *taskInfo, SOptrBasicInfo* pbInfo, SGroupResInfo* pGroupResInfo, SExprInfo* pExprInfo,
X
Xiaoyu Wang 已提交
2264
                            SDiskbasedBuf* pBuf) {
2265 2266
  assert(pGroupResInfo->currentGroup <= pGroupResInfo->totalGroup);

X
Xiaoyu Wang 已提交
2267 2268 2269
  int32_t*        rowCellOffset = pbInfo->rowCellInfoOffset;
  SSDataBlock*    pBlock = pbInfo->pRes;
  SqlFunctionCtx* pCtx = pbInfo->pCtx;
2270

2271
  blockDataCleanup(pBlock);
2272 2273 2274 2275
  if (!hasRemainDataInCurrentGroup(pGroupResInfo)) {
    return;
  }

2276
  int32_t orderType = TSDB_ORDER_ASC;
2277
  doCopyToSDataBlock(taskInfo, pBlock, pExprInfo, pBuf, pGroupResInfo, orderType, rowCellOffset, pCtx);
2278

H
Haojun Liao 已提交
2279 2280
  // add condition (pBlock->info.rows >= 1) just to runtime happy
  blockDataUpdateTsWindow(pBlock);
2281 2282
}

L
Liu Jicong 已提交
2283 2284
static void updateNumOfRowsInResultRows(SqlFunctionCtx* pCtx, int32_t numOfOutput, SResultRowInfo* pResultRowInfo,
                                        int32_t* rowCellInfoOffset) {
2285
  // update the number of result for each, only update the number of rows for the corresponding window result.
L
Liu Jicong 已提交
2286 2287 2288
  //  if (QUERY_IS_INTERVAL_QUERY(pQueryAttr)) {
  //    return;
  //  }
H
Haojun Liao 已提交
2289
#if 0
2290
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
L
Liu Jicong 已提交
2291
    SResultRow* pResult = pResultRowInfo->pResult[i];
2292 2293 2294 2295 2296 2297 2298

    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;
      }

2299 2300
      SResultRowEntryInfo* pCell = getResultCell(pResult, j, rowCellInfoOffset);
      pResult->numOfRows = (uint16_t)(TMAX(pResult->numOfRows, pCell->numOfRes));
2301 2302
    }
  }
H
Haojun Liao 已提交
2303
#endif
2304 2305
}

L
Liu Jicong 已提交
2306
static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
2307 2308
  int32_t colSize = pColRes->info.bytes * numOfRows;
  return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
L
Liu Jicong 已提交
2309
                                                      colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
2310 2311
}

2312
int32_t doFillTimeIntervalGapsInResults(struct SFillInfo* pFillInfo, SSDataBlock* pBlock, int32_t capacity) {
L
Liu Jicong 已提交
2313 2314 2315 2316
  //  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);
  //  }
2317

2318 2319
  int32_t numOfRows = (int32_t)taosFillResultDataBlock(pFillInfo, pBlock, capacity - pBlock->info.rows);
  pBlock->info.rows += numOfRows;
2320

2321
  return pBlock->info.rows;
2322 2323
}

2324
void publishOperatorProfEvent(SOperatorInfo* pOperator, EQueryProfEventType eventType) {
2325 2326
  SQueryProfEvent event = {0};

L
Liu Jicong 已提交
2327 2328
  event.eventType = eventType;
  event.eventTime = taosGetTimestampUs();
2329
  event.operatorType = pOperator->operatorType;
dengyihao's avatar
dengyihao 已提交
2330 2331 2332
  //    if (pQInfo->summary.queryProfEvents) {
  //      taosArrayPush(pQInfo->summary.queryProfEvents, &event);
  //    }
2333 2334
}

L
Liu Jicong 已提交
2335
void publishQueryAbortEvent(SExecTaskInfo* pTaskInfo, int32_t code) {
2336 2337 2338 2339 2340
  SQueryProfEvent event;
  event.eventType = QUERY_PROF_QUERY_ABORT;
  event.eventTime = taosGetTimestampUs();
  event.abortCode = code;

2341 2342
  if (pTaskInfo->cost.queryProfEvents) {
    taosArrayPush(pTaskInfo->cost.queryProfEvents, &event);
2343 2344 2345
  }
}

L
Liu Jicong 已提交
2346
typedef struct {
2347 2348 2349 2350 2351 2352 2353
  uint8_t operatorType;
  int64_t beginTime;
  int64_t endTime;
  int64_t selfTime;
  int64_t descendantsTime;
} SOperatorStackItem;

L
Liu Jicong 已提交
2354 2355
static void doOperatorExecProfOnce(SOperatorStackItem* item, SQueryProfEvent* event, SArray* opStack,
                                   SHashObj* profResults) {
2356 2357 2358 2359 2360 2361 2362 2363
  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
2364
  uint8_t              operatorType = item->operatorType;
2365 2366 2367 2368 2369 2370 2371 2372 2373
  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
2374
    taosHashPut(profResults, &(operatorType), sizeof(operatorType), &opResult, sizeof(opResult));
2375 2376 2377
  }
}

2378
void calculateOperatorProfResults(void) {
dengyihao's avatar
dengyihao 已提交
2379 2380 2381 2382 2383 2384 2385 2386 2387
  //  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;
  //  }
2388 2389 2390 2391 2392

  SArray* opStack = taosArrayInit(32, sizeof(SOperatorStackItem));
  if (opStack == NULL) {
    return;
  }
2393
#if 0
L
Liu Jicong 已提交
2394
  size_t    size = taosArrayGetSize(pQInfo->summary.queryProfEvents);
2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415
  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);
      }
    }
  }
2416
#endif
2417 2418 2419
  taosArrayDestroy(opStack);
}

L
Liu Jicong 已提交
2420 2421
void queryCostStatis(SExecTaskInfo* pTaskInfo) {
  STaskCostInfo* pSummary = &pTaskInfo->cost;
2422

L
Liu Jicong 已提交
2423 2424 2425
  //  uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
  //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
  //  pSummary->hashSize = hashSize;
2426 2427 2428 2429

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

L
Liu Jicong 已提交
2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440
  //  SResultRowPool* p = pTaskInfo->pool;
  //  if (p != NULL) {
  //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
  //    pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
  //  } else {
  //    pSummary->winInfoSize = 0;
  //    pSummary->numOfTimeWindows = 0;
  //  }
  //
  //  calculateOperatorProfResults(pQInfo);

2441 2442
  SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
  if (pSummary->pRecoder != NULL) {
X
Xiaoyu Wang 已提交
2443 2444
    qDebug("%s :cost summary: elapsed time:%" PRId64 " us, first merge:%" PRId64
           " us, total blocks:%d, "
2445 2446 2447 2448
           "load block statis:%d, load data block:%d, total rows:%" PRId64 ", check rows:%" PRId64,
           GET_TASKID(pTaskInfo), pSummary->elapsedTime, pSummary->firstStageMergeTime, pRecorder->totalBlocks,
           pRecorder->loadBlockStatis, pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
  }
L
Liu Jicong 已提交
2449 2450 2451
  // 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);
2452 2453 2454 2455

  if (pSummary->operatorProfResults) {
    SOperatorProfResult* opRes = taosHashIterate(pSummary->operatorProfResults, NULL);
    while (opRes != NULL) {
L
Liu Jicong 已提交
2456 2457
      // qDebug("QInfo:0x%" PRIx64 " :cost summary: operator : %d, exec times: %" PRId64 ", self time: %" PRId64,
      //             pQInfo->qId, opRes->operatorType, opRes->sumRunTimes, opRes->sumSelfTime);
2458 2459 2460 2461 2462
      opRes = taosHashIterate(pSummary->operatorProfResults, opRes);
    }
  }
}

L
Liu Jicong 已提交
2463 2464 2465
// static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
2466
//
L
Liu Jicong 已提交
2467
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2468
//
L
Liu Jicong 已提交
2469 2470 2471 2472
//   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;
//   }
2473
//
L
Liu Jicong 已提交
2474 2475 2476 2477 2478
//   if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//     pQueryAttr->pos = (int32_t)pQueryAttr->limit.offset;
//   } else {
//     pQueryAttr->pos = pBlockInfo->rows - (int32_t)pQueryAttr->limit.offset - 1;
//   }
2479
//
L
Liu Jicong 已提交
2480
//   assert(pQueryAttr->pos >= 0 && pQueryAttr->pos <= pBlockInfo->rows - 1);
2481
//
L
Liu Jicong 已提交
2482 2483
//   SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//   SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2484
//
L
Liu Jicong 已提交
2485 2486
//   // update the pQueryAttr->limit.offset value, and pQueryAttr->pos value
//   TSKEY *keys = (TSKEY *) pColInfoData->pData;
2487
//
L
Liu Jicong 已提交
2488 2489 2490
//   // update the offset value
//   pTableQueryInfo->lastKey = keys[pQueryAttr->pos];
//   pQueryAttr->limit.offset = 0;
2491
//
L
Liu Jicong 已提交
2492
//   int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
2493
//
L
Liu Jicong 已提交
2494 2495 2496 2497
//   //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);
// }
2498

L
Liu Jicong 已提交
2499 2500
// void skipBlocks(STaskRuntimeEnv *pRuntimeEnv) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
2501
//
L
Liu Jicong 已提交
2502 2503 2504
//   if (pQueryAttr->limit.offset <= 0 || pQueryAttr->numOfFilterCols > 0) {
//     return;
//   }
2505
//
L
Liu Jicong 已提交
2506 2507
//   pQueryAttr->pos = 0;
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2508
//
L
Liu Jicong 已提交
2509 2510
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
//   TsdbQueryHandleT pTsdbReadHandle = pRuntimeEnv->pTsdbReadHandle;
2511
//
L
Liu Jicong 已提交
2512 2513 2514 2515 2516
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pTsdbReadHandle)) {
//     if (isTaskKilled(pRuntimeEnv->qinfo)) {
//       longjmp(pRuntimeEnv->env, TSDB_CODE_TSC_QUERY_CANCELLED);
//     }
2517
//
L
Liu Jicong 已提交
2518
//     tsdbRetrieveDataBlockInfo(pTsdbReadHandle, &blockInfo);
2519
//
L
Liu Jicong 已提交
2520 2521 2522 2523
//     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;
2524
//
L
Liu Jicong 已提交
2525 2526 2527 2528 2529 2530 2531
//       //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;
//     }
//   }
2532
//
L
Liu Jicong 已提交
2533 2534 2535 2536 2537 2538 2539 2540 2541
//   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;
2542
//
L
Liu Jicong 已提交
2543 2544 2545
//   assert(pQueryAttr->limit.offset == 0);
//   STimeWindow tw = *win;
//   getNextTimeWindow(pQueryAttr, &tw);
2546
//
L
Liu Jicong 已提交
2547 2548
//   if ((tw.skey <= pBlockInfo->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) ||
//       (tw.ekey >= pBlockInfo->window.skey && !QUERY_IS_ASC_QUERY(pQueryAttr))) {
2549
//
L
Liu Jicong 已提交
2550 2551 2552 2553
//     // 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);
2554
//
L
Liu Jicong 已提交
2555 2556 2557 2558
//     tw = *win;
//     int32_t startPos =
//         getNextQualifiedWindow(pQueryAttr, &tw, pBlockInfo, pColInfoData->pData, binarySearchForKey, -1);
//     assert(startPos >= 0);
2559
//
L
Liu Jicong 已提交
2560 2561
//     // set the abort info
//     pQueryAttr->pos = startPos;
2562
//
L
Liu Jicong 已提交
2563 2564 2565 2566
//     // reset the query start timestamp
//     pTableQueryInfo->win.skey = ((TSKEY *)pColInfoData->pData)[startPos];
//     pQueryAttr->window.skey = pTableQueryInfo->win.skey;
//     TSKEY key = pTableQueryInfo->win.skey;
2567
//
L
Liu Jicong 已提交
2568 2569
//     pWindowResInfo->prevSKey = tw.skey;
//     int32_t index = pRuntimeEnv->resultRowInfo.curIndex;
2570
//
L
Liu Jicong 已提交
2571 2572
//     int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
//     pRuntimeEnv->resultRowInfo.curIndex = index;  // restore the window index
2573
//
L
Liu Jicong 已提交
2574 2575 2576 2577
//     //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);
2578
//
L
Liu Jicong 已提交
2579 2580 2581 2582 2583
//     return key;
//   } else {  // do nothing
//     pQueryAttr->window.skey      = tw.skey;
//     pWindowResInfo->prevSKey = tw.skey;
//     pTableQueryInfo->lastKey = tw.skey;
2584
//
L
Liu Jicong 已提交
2585 2586
//     return tw.skey;
//   }
2587
//
L
Liu Jicong 已提交
2588 2589 2590 2591 2592 2593 2594 2595 2596 2597
//   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);
//   }
2598
//
L
Liu Jicong 已提交
2599 2600 2601 2602 2603
//   // 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;
//   }
2604
//
L
Liu Jicong 已提交
2605 2606 2607 2608 2609 2610 2611
//   /*
//    * 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);
2612
//
L
Liu Jicong 已提交
2613 2614
//   STimeWindow w = TSWINDOW_INITIALIZER;
//   bool ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
2615
//
L
Liu Jicong 已提交
2616 2617
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
//   STableQueryInfo *pTableQueryInfo = pRuntimeEnv->current;
2618
//
L
Liu Jicong 已提交
2619 2620 2621
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pRuntimeEnv->pTsdbReadHandle)) {
//     tsdbRetrieveDataBlockInfo(pRuntimeEnv->pTsdbReadHandle, &blockInfo);
2622
//
L
Liu Jicong 已提交
2623 2624 2625 2626 2627 2628 2629 2630 2631
//     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;
//     }
2632
//
L
Liu Jicong 已提交
2633 2634
//     // the first time window
//     STimeWindow win = getActiveTimeWindow(pWindowResInfo, pWindowResInfo->prevSKey, pQueryAttr);
2635
//
L
Liu Jicong 已提交
2636 2637
//     while (pQueryAttr->limit.offset > 0) {
//       STimeWindow tw = win;
2638
//
L
Liu Jicong 已提交
2639 2640 2641
//       if ((win.ekey <= blockInfo.window.ekey && ascQuery) || (win.ekey >= blockInfo.window.skey && !ascQuery)) {
//         pQueryAttr->limit.offset -= 1;
//         pWindowResInfo->prevSKey = win.skey;
2642
//
L
Liu Jicong 已提交
2643 2644 2645 2646 2647 2648
//         // 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;
//         }
//       }
2649
//
L
Liu Jicong 已提交
2650 2651 2652 2653
//       if (pQueryAttr->limit.offset == 0) {
//         *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
//         return true;
//       }
2654
//
L
Liu Jicong 已提交
2655 2656
//       // current window does not ended in current data block, try next data block
//       getNextTimeWindow(pQueryAttr, &tw);
2657
//
L
Liu Jicong 已提交
2658 2659 2660 2661 2662 2663 2664 2665 2666
//       /*
//        * 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)) {
2667
//
L
Liu Jicong 已提交
2668 2669
//         SArray *pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//         SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2670
//
L
Liu Jicong 已提交
2671 2672 2673
//         if ((win.ekey > blockInfo.window.ekey && ascQuery) || (win.ekey < blockInfo.window.skey && !ascQuery)) {
//           pQueryAttr->limit.offset -= 1;
//         }
2674
//
L
Liu Jicong 已提交
2675 2676 2677 2678 2679 2680 2681 2682
//         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);
2683
//
L
Liu Jicong 已提交
2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694
//           // 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.
//       }
//     }
//   }
2695
//
L
Liu Jicong 已提交
2696 2697 2698 2699
//   // check for error
//   if (terrno != TSDB_CODE_SUCCESS) {
//     longjmp(pRuntimeEnv->env, terrno);
//   }
2700
//
L
Liu Jicong 已提交
2701 2702
//   return true;
// }
2703

2704
int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t num) {
H
Haojun Liao 已提交
2705
  if (p->pDownstream == NULL) {
H
Haojun Liao 已提交
2706
    assert(p->numOfDownstream == 0);
2707 2708
  }

wafwerar's avatar
wafwerar 已提交
2709
  p->pDownstream = taosMemoryCalloc(1, num * POINTER_BYTES);
2710 2711 2712 2713 2714 2715 2716
  if (p->pDownstream == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

  memcpy(p->pDownstream, pDownstream, num * POINTER_BYTES);
  p->numOfDownstream = num;
  return TSDB_CODE_SUCCESS;
2717 2718 2719 2720
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo);

2721
static void doTableQueryInfoTimeWindowCheck(SExecTaskInfo* pTaskInfo, STableQueryInfo* pTableQueryInfo, int32_t order) {
H
Haojun Liao 已提交
2722 2723
#if 0
    if (order == TSDB_ORDER_ASC) {
2724 2725
    assert(
        (pTableQueryInfo->win.skey <= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2726 2727
        (pTableQueryInfo->lastKey >= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey >= pTaskInfo->window.skey && pTableQueryInfo->win.ekey <= pTaskInfo->window.ekey));
2728 2729 2730
  } else {
    assert(
        (pTableQueryInfo->win.skey >= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2731 2732
        (pTableQueryInfo->lastKey <= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey <= pTaskInfo->window.skey && pTableQueryInfo->win.ekey >= pTaskInfo->window.ekey));
2733
  }
H
Haojun Liao 已提交
2734
#endif
2735 2736
}

L
Liu Jicong 已提交
2737 2738 2739 2740
// 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;
2741
//
L
Liu Jicong 已提交
2742 2743 2744
//   if (pTableQueryInfo->pTable == NULL) {
//     return;
//   }
2745
//
L
Liu Jicong 已提交
2746 2747 2748 2749 2750 2751 2752 2753 2754
//   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));
//   }
// }
2755

2756
int32_t loadRemoteDataCallback(void* param, const SDataBuf* pMsg, int32_t code) {
L
Liu Jicong 已提交
2757
  SSourceDataInfo* pSourceDataInfo = (SSourceDataInfo*)param;
H
Haojun Liao 已提交
2758 2759
  if (code == TSDB_CODE_SUCCESS) {
    pSourceDataInfo->pRsp = pMsg->pData;
2760

H
Haojun Liao 已提交
2761 2762
    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
    pRsp->numOfRows = htonl(pRsp->numOfRows);
dengyihao's avatar
dengyihao 已提交
2763
    pRsp->compLen = htonl(pRsp->compLen);
2764
    pRsp->numOfCols = htonl(pRsp->numOfCols);
dengyihao's avatar
dengyihao 已提交
2765
    pRsp->useconds = htobe64(pRsp->useconds);
H
Haojun Liao 已提交
2766 2767 2768
  } else {
    pSourceDataInfo->code = code;
  }
H
Haojun Liao 已提交
2769

H
Haojun Liao 已提交
2770
  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
2771
  tsem_post(&pSourceDataInfo->pEx->ready);
wmmhello's avatar
wmmhello 已提交
2772
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
2773 2774 2775 2776
}

static void destroySendMsgInfo(SMsgSendInfo* pMsgBody) {
  assert(pMsgBody != NULL);
wafwerar's avatar
wafwerar 已提交
2777 2778
  taosMemoryFreeClear(pMsgBody->msgInfo.pData);
  taosMemoryFreeClear(pMsgBody);
H
Haojun Liao 已提交
2779 2780
}

S
Shengliang Guan 已提交
2781
void qProcessFetchRsp(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
L
Liu Jicong 已提交
2782
  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->ahandle;
H
Haojun Liao 已提交
2783 2784 2785 2786 2787
  assert(pMsg->ahandle != NULL);

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

  if (pMsg->contLen > 0) {
wafwerar's avatar
wafwerar 已提交
2788
    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
H
Haojun Liao 已提交
2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799
    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);
2800 2801
}

L
Liu Jicong 已提交
2802
static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
2803
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
2804

wafwerar's avatar
wafwerar 已提交
2805
  SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
2806 2807 2808 2809
  if (NULL == pMsg) {
    pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return pTaskInfo->code;
  }
2810

L
Liu Jicong 已提交
2811 2812
  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
2813

L
Liu Jicong 已提交
2814 2815
  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);
2816 2817 2818 2819 2820 2821 2822

  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
2823
  SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
2824
  if (NULL == pMsgSendInfo) {
wafwerar's avatar
wafwerar 已提交
2825
    taosMemoryFreeClear(pMsg);
2826 2827 2828
    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
2829 2830
  }

2831 2832 2833 2834 2835
  pMsgSendInfo->param = pDataInfo;
  pMsgSendInfo->msgInfo.pData = pMsg;
  pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
  pMsgSendInfo->msgType = TDMT_VND_FETCH;
  pMsgSendInfo->fp = loadRemoteDataCallback;
2836

2837
  int64_t transporterId = 0;
L
Liu Jicong 已提交
2838
  int32_t code = asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
2839 2840 2841
  return TSDB_CODE_SUCCESS;
}

2842
// NOTE: sources columns are more than the destination SSDatablock columns.
2843
void relocateColumnData(SSDataBlock* pBlock, const SArray* pColMatchInfo, SArray* pCols) {
2844 2845 2846
  size_t numOfSrcCols = taosArrayGetSize(pCols);

  int32_t i = 0, j = 0;
L
Liu Jicong 已提交
2847
  while (i < numOfSrcCols && j < taosArrayGetSize(pColMatchInfo)) {
2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866
    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);
    }
  }
}

2867
int32_t setSDataBlockFromFetchRsp(SSDataBlock* pRes, SLoadRemoteDataInfo* pLoadInfo, int32_t numOfRows, char* pData,
dengyihao's avatar
dengyihao 已提交
2868 2869
                                  int32_t compLen, int32_t numOfOutput, int64_t startTs, uint64_t* total,
                                  SArray* pColList) {
H
Haojun Liao 已提交
2870
  if (pColList == NULL) {  // data from other sources
H
Haojun Liao 已提交
2871 2872
    blockDataEnsureCapacity(pRes, numOfRows);

L
Liu Jicong 已提交
2873
    int32_t dataLen = *(int32_t*)pData;
2874 2875
    pData += sizeof(int32_t);

L
Liu Jicong 已提交
2876
    pRes->info.groupId = *(uint64_t*)pData;
2877 2878
    pData += sizeof(uint64_t);

H
Haojun Liao 已提交
2879 2880
    int32_t* colLen = (int32_t*)pData;

2881
    char* pStart = pData + sizeof(int32_t) * numOfOutput;
2882
    for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
2883
      colLen[i] = htonl(colLen[i]);
2884
      ASSERT(colLen[i] >= 0);
H
Haojun Liao 已提交
2885

2886
      SColumnInfoData* pColInfoData = taosArrayGet(pRes->pDataBlock, i);
H
Haojun Liao 已提交
2887 2888 2889
      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
        pColInfoData->varmeta.length = colLen[i];
        pColInfoData->varmeta.allocLen = colLen[i];
2890

L
Liu Jicong 已提交
2891 2892
        memcpy(pColInfoData->varmeta.offset, pStart, sizeof(int32_t) * numOfRows);
        pStart += sizeof(int32_t) * numOfRows;
H
Haojun Liao 已提交
2893

2894 2895 2896
        if (colLen[i] > 0) {
          pColInfoData->pData = taosMemoryMalloc(colLen[i]);
        }
H
Haojun Liao 已提交
2897 2898 2899
      } else {
        memcpy(pColInfoData->nullbitmap, pStart, BitmapLen(numOfRows));
        pStart += BitmapLen(numOfRows);
2900
      }
H
Haojun Liao 已提交
2901

2902 2903 2904 2905
      if (colLen[i] > 0) {
        memcpy(pColInfoData->pData, pStart, colLen[i]);
      }

L
Liu Jicong 已提交
2906 2907
      // TODO setting this flag to true temporarily so aggregate function on stable will
      // examine NULL value for non-primary key column
2908
      pColInfoData->hasNull = true;
H
Haojun Liao 已提交
2909
      pStart += colLen[i];
2910
    }
H
Haojun Liao 已提交
2911
  } else {  // extract data according to pColList
2912
    ASSERT(numOfOutput == taosArrayGetSize(pColList));
2913 2914 2915 2916 2917
    char* pStart = pData;

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

2918
    // todo refactor:extract method
2919
    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
dengyihao's avatar
dengyihao 已提交
2920
    for (int32_t i = 0; i < numOfCols; ++i) {
2921 2922 2923 2924 2925 2926 2927
      SSysTableSchema* p = (SSysTableSchema*)pStart;

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

H
Haojun Liao 已提交
2928 2929 2930 2931
    SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
    pBlock->pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData));
    pBlock->info.numOfCols = numOfCols;

dengyihao's avatar
dengyihao 已提交
2932
    for (int32_t i = 0; i < numOfCols; ++i) {
2933 2934 2935 2936 2937
      SColumnInfoData idata = {0};
      idata.info.type = pSchema[i].type;
      idata.info.bytes = pSchema[i].bytes;
      idata.info.colId = pSchema[i].colId;

H
Haojun Liao 已提交
2938
      taosArrayPush(pBlock->pDataBlock, &idata);
2939
      if (IS_VAR_DATA_TYPE(idata.info.type)) {
H
Haojun Liao 已提交
2940
        pBlock->info.hasVarCol = true;
2941 2942 2943
      }
    }

H
Haojun Liao 已提交
2944
    blockDataEnsureCapacity(pBlock, numOfRows);
2945

L
Liu Jicong 已提交
2946 2947
    int32_t  dataLen = *(int32_t*)pStart;
    uint64_t groupId = *(uint64_t*)(pStart + sizeof(int32_t));
2948 2949
    pStart += sizeof(int32_t) + sizeof(uint64_t);

L
Liu Jicong 已提交
2950
    int32_t* colLen = (int32_t*)(pStart);
2951 2952 2953 2954
    pStart += sizeof(int32_t) * numOfCols;

    for (int32_t i = 0; i < numOfCols; ++i) {
      colLen[i] = htonl(colLen[i]);
2955
      ASSERT(colLen[i] >= 0);
2956

H
Haojun Liao 已提交
2957
      SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
      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
2974 2975

    // data from mnode
H
Haojun Liao 已提交
2976 2977 2978
    relocateColumnData(pRes, pColList, pBlock->pDataBlock);
    taosArrayDestroy(pBlock->pDataBlock);
    taosMemoryFree(pBlock);
X
Xiaoyu Wang 已提交
2979
    //    blockDataDestroy(pBlock);
2980
  }
2981

H
Haojun Liao 已提交
2982
  pRes->info.rows = numOfRows;
2983
  blockDataUpdateTsWindow(pRes);
2984

2985
  int64_t el = taosGetTimestampUs() - startTs;
2986

H
Haojun Liao 已提交
2987 2988
  pLoadInfo->totalRows += numOfRows;
  pLoadInfo->totalSize += compLen;
2989

H
Haojun Liao 已提交
2990 2991 2992
  if (total != NULL) {
    *total += numOfRows;
  }
2993

H
Haojun Liao 已提交
2994
  pLoadInfo->totalElapsed += el;
2995 2996
  return TSDB_CODE_SUCCESS;
}
2997

L
Liu Jicong 已提交
2998 2999
static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
3000
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3001

L
Liu Jicong 已提交
3002
  int64_t              el = taosGetTimestampUs() - startTs;
H
Haojun Liao 已提交
3003 3004
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
  pLoadInfo->totalElapsed += el;
H
Haojun Liao 已提交
3005

3006
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
L
Liu Jicong 已提交
3007 3008 3009
  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);
3010 3011 3012 3013 3014

  doSetOperatorCompleted(pOperator);
  return NULL;
}

L
Liu Jicong 已提交
3015 3016
static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
                                                   SExecTaskInfo* pTaskInfo) {
3017 3018 3019 3020 3021 3022 3023 3024 3025
  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);

3026
      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
3027
        completed += 1;
H
Haojun Liao 已提交
3028 3029
        continue;
      }
3030

3031
      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
3032 3033 3034
        continue;
      }

L
Liu Jicong 已提交
3035
      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3036
      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
3037

L
Liu Jicong 已提交
3038
      SSDataBlock*         pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3039
      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3040
      if (pRsp->numOfRows == 0) {
L
Liu Jicong 已提交
3041 3042
        qDebug("%s vgId:%d, taskID:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
               " try next",
3043
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i + 1, pDataInfo->totalRows,
H
Haojun Liao 已提交
3044
               pExchangeInfo->loadInfo.totalRows);
3045
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3046 3047 3048
        completed += 1;
        continue;
      }
H
Haojun Liao 已提交
3049

H
Haojun Liao 已提交
3050
      SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3051 3052
      code = setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
                                       pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3053
      if (code != 0) {
3054 3055 3056
        goto _error;
      }

3057 3058 3059
      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
3060 3061
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
               pLoadInfo->totalRows, pLoadInfo->totalSize, i + 1, totalSources);
3062
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3063
      } else {
dengyihao's avatar
dengyihao 已提交
3064 3065
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
               ", totalBytes:%" PRIu64,
H
Haojun Liao 已提交
3066 3067
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
               pLoadInfo->totalSize);
3068 3069
      }

3070 3071
      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090
        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
3091 3092 3093
static SSDataBlock* concurrentlyLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3094

H
Haojun Liao 已提交
3095 3096 3097
  if (pOperator->status == OP_RES_TO_RETURN) {
    return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
  }
3098

L
Liu Jicong 已提交
3099
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3100 3101 3102
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3103
  for (int32_t i = 0; i < totalSources; ++i) {
3104 3105
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3106
      return NULL;
3107 3108 3109 3110
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3111 3112
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3113 3114

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3115 3116
  pOperator->status = OP_RES_TO_RETURN;
  return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
3117 3118
}

L
Liu Jicong 已提交
3119 3120 3121
static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3122

L
Liu Jicong 已提交
3123
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3124 3125 3126
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3127
  for (int32_t i = 0; i < totalSources; ++i) {
3128 3129
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3130 3131
      pTaskInfo->code = code;
      return code;
3132 3133 3134 3135
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3136 3137
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3138 3139

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3140
  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
3141

H
Haojun Liao 已提交
3142
  return TSDB_CODE_SUCCESS;
3143 3144
}

L
Liu Jicong 已提交
3145 3146 3147
static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3148

L
Liu Jicong 已提交
3149
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3150
  int64_t startTs = taosGetTimestampUs();
3151

L
Liu Jicong 已提交
3152
  while (1) {
3153 3154
    if (pExchangeInfo->current >= totalSources) {
      return setAllSourcesCompleted(pOperator, startTs);
3155
    }
3156

3157 3158 3159
    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
    tsem_wait(&pExchangeInfo->ready);

dengyihao's avatar
dengyihao 已提交
3160
    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
X
Xiaoyu Wang 已提交
3161
    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
3162

H
Haojun Liao 已提交
3163
    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
dengyihao's avatar
dengyihao 已提交
3164 3165
      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
3166 3167 3168 3169
      pOperator->pTaskInfo->code = pDataInfo->code;
      return NULL;
    }

L
Liu Jicong 已提交
3170
    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
H
Haojun Liao 已提交
3171
    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3172
    if (pRsp->numOfRows == 0) {
dengyihao's avatar
dengyihao 已提交
3173 3174
      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " %d of total completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
             " try next",
3175
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pExchangeInfo->current + 1,
H
Haojun Liao 已提交
3176
             pDataInfo->totalRows, pLoadInfo->totalRows);
H
Haojun Liao 已提交
3177

3178
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3179 3180 3181
      pExchangeInfo->current += 1;
      continue;
    }
H
Haojun Liao 已提交
3182

L
Liu Jicong 已提交
3183
    SSDataBlock*       pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3184
    SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
L
Liu Jicong 已提交
3185 3186
    int32_t            code =
        setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
3187
                                  pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3188 3189

    if (pRsp->completed == 1) {
H
Haojun Liao 已提交
3190
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
L
Liu Jicong 已提交
3191 3192 3193
             ", 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);
3194

3195
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3196 3197
      pExchangeInfo->current += 1;
    } else {
L
Liu Jicong 已提交
3198 3199 3200 3201
      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);
3202 3203 3204 3205
    }

    return pExchangeInfo->pResult;
  }
3206 3207
}

L
Liu Jicong 已提交
3208
static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
3209
  if (OPTR_IS_OPENED(pOperator)) {
H
Haojun Liao 已提交
3210 3211 3212
    return TSDB_CODE_SUCCESS;
  }

L
Liu Jicong 已提交
3213
  SExchangeInfo* pExchangeInfo = pOperator->info;
H
Haojun Liao 已提交
3214 3215 3216 3217 3218 3219 3220 3221 3222
  if (pExchangeInfo->seqLoadData) {
    // do nothing for sequentially load data
  } else {
    int32_t code = prepareConcurrentlyLoad(pOperator);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
    }
  }

3223
  OPTR_SET_OPENED(pOperator);
H
Haojun Liao 已提交
3224 3225 3226
  return TSDB_CODE_SUCCESS;
}

3227
static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3228 3229
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3230

3231
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
3232
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3233 3234
    return NULL;
  }
3235

L
Liu Jicong 已提交
3236
  size_t               totalSources = taosArrayGetSize(pExchangeInfo->pSources);
H
Haojun Liao 已提交
3237
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
H
Haojun Liao 已提交
3238

3239
  if (pOperator->status == OP_EXEC_DONE) {
L
Liu Jicong 已提交
3240 3241 3242
    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);
3243 3244 3245 3246 3247 3248
    return NULL;
  }

  if (pExchangeInfo->seqLoadData) {
    return seqLoadRemoteData(pOperator);
  } else {
H
Haojun Liao 已提交
3249
    return concurrentlyLoadRemoteData(pOperator);
3250
  }
H
Haojun Liao 已提交
3251

3252
#if 0
H
Haojun Liao 已提交
3253
  _error:
wafwerar's avatar
wafwerar 已提交
3254 3255
  taosMemoryFreeClear(pMsg);
  taosMemoryFreeClear(pMsgSendInfo);
H
Haojun Liao 已提交
3256 3257 3258

  terrno = pTaskInfo->code;
  return NULL;
3259
#endif
H
Haojun Liao 已提交
3260
}
3261

H
Haojun Liao 已提交
3262
static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo) {
3263
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
H
Haojun Liao 已提交
3264 3265
  if (pInfo->pSourceDataInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
3266 3267
  }

L
Liu Jicong 已提交
3268
  for (int32_t i = 0; i < numOfSources; ++i) {
3269
    SSourceDataInfo dataInfo = {0};
H
Haojun Liao 已提交
3270
    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
L
Liu Jicong 已提交
3271 3272
    dataInfo.pEx = pInfo;
    dataInfo.index = i;
3273

H
Haojun Liao 已提交
3274 3275 3276 3277 3278 3279 3280 3281 3282 3283
    void* ret = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
    if (ret == NULL) {
      taosArrayDestroy(pInfo->pSourceDataInfo);
      return TSDB_CODE_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
}

3284
SOperatorInfo* createExchangeOperatorInfo(void *pTransporter, const SNodeList* pSources, SSDataBlock* pBlock,
3285
                                          SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
3286
  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
wafwerar's avatar
wafwerar 已提交
3287
  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
3288 3289

  if (pInfo == NULL || pOperator == NULL) {
wafwerar's avatar
wafwerar 已提交
3290 3291
    taosMemoryFreeClear(pInfo);
    taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3292 3293
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return NULL;
H
Haojun Liao 已提交
3294 3295
  }

H
Haojun Liao 已提交
3296
  size_t numOfSources = LIST_LENGTH(pSources);
H
Haojun Liao 已提交
3297
  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
H
Haojun Liao 已提交
3298 3299 3300
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
  if (pInfo->pSourceDataInfo == NULL || pInfo->pSources == NULL) {
    goto _error;
H
Haojun Liao 已提交
3301 3302
  }

L
Liu Jicong 已提交
3303 3304
  for (int32_t i = 0; i < numOfSources; ++i) {
    SNodeListNode* pNode = nodesListGetNode((SNodeList*)pSources, i);
H
Haojun Liao 已提交
3305 3306
    taosArrayPush(pInfo->pSources, pNode);
  }
3307

H
Haojun Liao 已提交
3308 3309 3310
  int32_t code = initDataSource(numOfSources, pInfo);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3311
  }
H
Haojun Liao 已提交
3312

dengyihao's avatar
dengyihao 已提交
3313
  pInfo->pResult = pBlock;
3314 3315 3316
  pInfo->seqLoadData = true;

  tsem_init(&pInfo->ready, 0, 0);
3317

dengyihao's avatar
dengyihao 已提交
3318
  pOperator->name = "ExchangeOperator";
X
Xiaoyu Wang 已提交
3319
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
3320
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
3321 3322
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3323
  pOperator->numOfExprs = pBlock->info.numOfCols;
dengyihao's avatar
dengyihao 已提交
3324
  pOperator->pTaskInfo = pTaskInfo;
3325

L
Liu Jicong 已提交
3326 3327
  pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                         destroyExchangeOperatorInfo, NULL, NULL, NULL);
3328
  pInfo->pTransporter = pTransporter;
3329

3330
  return pOperator;
H
Haojun Liao 已提交
3331

L
Liu Jicong 已提交
3332
_error:
H
Haojun Liao 已提交
3333
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3334
    destroyExchangeOperatorInfo(pInfo, numOfSources);
H
Haojun Liao 已提交
3335 3336
  }

wafwerar's avatar
wafwerar 已提交
3337 3338
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3339
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
H
Haojun Liao 已提交
3340
  return NULL;
3341 3342
}

dengyihao's avatar
dengyihao 已提交
3343 3344
static int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                                const char* pKey);
3345

3346
static void destroySortedMergeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
3347
  SSortedMergeOperatorInfo* pInfo = (SSortedMergeOperatorInfo*)param;
H
Haojun Liao 已提交
3348
  taosArrayDestroy(pInfo->pSortInfo);
3349 3350 3351
  taosArrayDestroy(pInfo->groupInfo);

  if (pInfo->pSortHandle != NULL) {
H
Haojun Liao 已提交
3352
    tsortDestroySortHandle(pInfo->pSortHandle);
3353 3354
  }

H
Haojun Liao 已提交
3355
  blockDataDestroy(pInfo->binfo.pRes);
H
Haojun Liao 已提交
3356
  cleanupAggSup(&pInfo->aggSup);
3357
}
H
Haojun Liao 已提交
3358

L
Liu Jicong 已提交
3359
static bool needToMerge(SSDataBlock* pBlock, SArray* groupInfo, char** buf, int32_t rowIndex) {
3360 3361 3362 3363
  size_t size = taosArrayGetSize(groupInfo);
  if (size == 0) {
    return true;
  }
3364

3365 3366
  for (int32_t i = 0; i < size; ++i) {
    int32_t* index = taosArrayGet(groupInfo, i);
3367

3368
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
L
Liu Jicong 已提交
3369
    bool             isNull = colDataIsNull(pColInfo, rowIndex, pBlock->info.rows, NULL);
3370

3371 3372 3373
    if ((isNull && buf[i] != NULL) || (!isNull && buf[i] == NULL)) {
      return false;
    }
3374

3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387
    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;
      }
3388 3389 3390
    }
  }

3391
  return 0;
3392 3393
}

L
Liu Jicong 已提交
3394 3395 3396
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
3397
    pCtx[j].startRow = rowIndex;
3398 3399
  }

3400 3401
  for (int32_t j = 0; j < numOfExpr; ++j) {
    int32_t functionId = pCtx[j].functionId;
L
Liu Jicong 已提交
3402 3403 3404 3405 3406 3407 3408 3409 3410
    //    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]);
    //    }
3411
  }
3412
}
3413

L
Liu Jicong 已提交
3414 3415
static void doFinalizeResultImpl(SqlFunctionCtx* pCtx, int32_t numOfExpr) {
  for (int32_t j = 0; j < numOfExpr; ++j) {
3416 3417 3418 3419
    int32_t functionId = pCtx[j].functionId;
    //    if (functionId == FUNC_TAG_DUMMY || functionId == FUNC_TS_DUMMY) {
    //      continue;
    //    }
3420

3421 3422 3423 3424
    //    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
3425
    //    pCtx[j].fpSet.finalize(&pCtx[j]);
3426 3427
  }
}
3428

3429
static bool saveCurrentTuple(char** rowColData, SArray* pColumnList, SSDataBlock* pBlock, int32_t rowIndex) {
L
Liu Jicong 已提交
3430
  int32_t size = (int32_t)taosArrayGetSize(pColumnList);
3431

L
Liu Jicong 已提交
3432 3433
  for (int32_t i = 0; i < size; ++i) {
    int32_t*         index = taosArrayGet(pColumnList, i);
3434
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
H
Haojun Liao 已提交
3435

3436 3437 3438
    char* data = colDataGetData(pColInfo, rowIndex);
    memcpy(rowColData[i], data, colDataGetLength(pColInfo, rowIndex));
  }
3439

3440 3441
  return true;
}
3442

3443 3444
static void doMergeImpl(SOperatorInfo* pOperator, int32_t numOfExpr, SSDataBlock* pBlock) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
3445

3446
  SqlFunctionCtx* pCtx = pInfo->binfo.pCtx;
L
Liu Jicong 已提交
3447
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
3448 3449
    pCtx[i].size = 1;
  }
3450

L
Liu Jicong 已提交
3451
  for (int32_t i = 0; i < pBlock->info.rows; ++i) {
3452 3453 3454 3455 3456 3457 3458 3459 3460
    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);
3461 3462
        int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfExprs, NULL);
        //        setTagValueForMultipleRows(pCtx, pOperator->numOfExprs, numOfRows);
3463

3464
        // TODO check for available buffer;
H
Haojun Liao 已提交
3465

3466 3467 3468 3469 3470
        // next group info data
        pInfo->binfo.pRes->info.rows += numOfRows;
        for (int32_t j = 0; j < numOfExpr; ++j) {
          if (pCtx[j].functionId < 0) {
            continue;
3471
          }
3472

H
Haojun Liao 已提交
3473
          pCtx[j].fpSet.process(&pCtx[j]);
3474
        }
3475 3476 3477

        doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
        pInfo->hasGroupVal = saveCurrentTuple(pInfo->groupVal, pInfo->groupInfo, pBlock, i);
H
Haojun Liao 已提交
3478
      }
3479 3480 3481 3482
    }
  }
}

3483 3484
static SSDataBlock* doMerge(SOperatorInfo* pOperator) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
L
Liu Jicong 已提交
3485
  SSortHandle*              pHandle = pInfo->pSortHandle;
3486

3487
  SSDataBlock* pDataBlock = createOneDataBlock(pInfo->binfo.pRes, false);
3488
  blockDataEnsureCapacity(pDataBlock, pOperator->resultInfo.capacity);
3489

L
Liu Jicong 已提交
3490
  while (1) {
3491
    blockDataCleanup(pDataBlock);
3492
    while (1) {
H
Haojun Liao 已提交
3493
      STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
3494 3495
      if (pTupleHandle == NULL) {
        break;
3496
      }
3497

3498 3499
      // build datablock for merge for one group
      appendOneRowToDataBlock(pDataBlock, pTupleHandle);
3500
      if (pDataBlock->info.rows >= pOperator->resultInfo.capacity) {
3501 3502
        break;
      }
3503
    }
3504

3505 3506 3507
    if (pDataBlock->info.rows == 0) {
      break;
    }
3508

3509
    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pDataBlock, TSDB_ORDER_ASC, true);
L
Liu Jicong 已提交
3510 3511
    //  updateOutputBuf(&pInfo->binfo, &pAggInfo->bufCapacity, pBlock->info.rows * pAggInfo->resultRowFactor,
    //  pOperator->pRuntimeEnv, true);
3512
    doMergeImpl(pOperator, pOperator->numOfExprs, pDataBlock);
3513 3514
    // flush to tuple store, and after all data have been handled, return to upstream node or sink node
  }
3515

3516 3517 3518
  doFinalizeResultImpl(pInfo->binfo.pCtx, pOperator->numOfExprs);
  int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfExprs, NULL);
  //        setTagValueForMultipleRows(pCtx, pOperator->numOfExprs, numOfRows);
3519

3520
  // TODO check for available buffer;
3521

3522 3523
  // next group info data
  pInfo->binfo.pRes->info.rows += numOfRows;
L
Liu Jicong 已提交
3524
  return (pInfo->binfo.pRes->info.rows > 0) ? pInfo->binfo.pRes : NULL;
3525
}
3526

3527
static SSDataBlock* doSortedMerge(SOperatorInfo* pOperator) {
3528 3529
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
3530 3531
  }

L
Liu Jicong 已提交
3532
  SExecTaskInfo*            pTaskInfo = pOperator->pTaskInfo;
3533
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
H
Haojun Liao 已提交
3534
  if (pOperator->status == OP_RES_TO_RETURN) {
3535
    return getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity, NULL);
3536 3537
  }

3538
  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
dengyihao's avatar
dengyihao 已提交
3539 3540
  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, NULL, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize,
                                             numOfBufPage, pInfo->binfo.pRes, "GET_TASKID(pTaskInfo)");
H
Haojun Liao 已提交
3541

3542
  tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock, NULL, NULL);
3543

L
Liu Jicong 已提交
3544
  for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
wmmhello's avatar
wmmhello 已提交
3545
    SSortSource* ps = taosMemoryCalloc(1, sizeof(SSortSource));
H
Haojun Liao 已提交
3546
    ps->param = pOperator->pDownstream[i];
H
Haojun Liao 已提交
3547
    tsortAddSource(pInfo->pSortHandle, ps);
3548 3549
  }

H
Haojun Liao 已提交
3550
  int32_t code = tsortOpen(pInfo->pSortHandle);
3551
  if (code != TSDB_CODE_SUCCESS) {
3552
    longjmp(pTaskInfo->env, terrno);
3553 3554
  }

H
Haojun Liao 已提交
3555
  pOperator->status = OP_RES_TO_RETURN;
3556
  return doMerge(pOperator);
3557
}
3558

L
Liu Jicong 已提交
3559 3560
static int32_t initGroupCol(SExprInfo* pExprInfo, int32_t numOfCols, SArray* pGroupInfo,
                            SSortedMergeOperatorInfo* pInfo) {
3561 3562
  if (pGroupInfo == NULL || taosArrayGetSize(pGroupInfo) == 0) {
    return 0;
H
Haojun Liao 已提交
3563 3564
  }

3565 3566 3567 3568 3569 3570 3571 3572
  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
3573 3574
  size_t numOfGroupCol = taosArrayGetSize(pInfo->groupInfo);
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3575
    SColumn* pCol = taosArrayGet(pGroupInfo, i);
L
Liu Jicong 已提交
3576
    for (int32_t j = 0; j < numOfCols; ++j) {
H
Haojun Liao 已提交
3577
      SExprInfo* pe = &pExprInfo[j];
3578
      if (pe->base.resSchema.slotId == pCol->colId) {
3579 3580
        taosArrayPush(plist, pCol);
        taosArrayPush(pInfo->groupInfo, &j);
H
Haojun Liao 已提交
3581
        len += pCol->bytes;
3582 3583
        break;
      }
H
Haojun Liao 已提交
3584 3585 3586
    }
  }

3587
  ASSERT(taosArrayGetSize(pGroupInfo) == taosArrayGetSize(plist));
H
Haojun Liao 已提交
3588

wafwerar's avatar
wafwerar 已提交
3589
  pInfo->groupVal = taosMemoryCalloc(1, (POINTER_BYTES * numOfGroupCol + len));
3590 3591 3592 3593
  if (pInfo->groupVal == NULL) {
    taosArrayDestroy(plist);
    return TSDB_CODE_OUT_OF_MEMORY;
  }
H
Haojun Liao 已提交
3594

3595
  int32_t offset = 0;
L
Liu Jicong 已提交
3596 3597
  char*   start = (char*)(pInfo->groupVal + (POINTER_BYTES * numOfGroupCol));
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3598 3599
    pInfo->groupVal[i] = start + offset;
    SColumn* pCol = taosArrayGet(plist, i);
H
Haojun Liao 已提交
3600
    offset += pCol->bytes;
3601
  }
H
Haojun Liao 已提交
3602

3603
  taosArrayDestroy(plist);
H
Haojun Liao 已提交
3604

3605 3606
  return TSDB_CODE_SUCCESS;
}
H
Haojun Liao 已提交
3607

L
Liu Jicong 已提交
3608 3609 3610
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
3611
  SSortedMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortedMergeOperatorInfo));
L
Liu Jicong 已提交
3612
  SOperatorInfo*            pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
3613
  if (pInfo == NULL || pOperator == NULL) {
3614
    goto _error;
3615
  }
H
Haojun Liao 已提交
3616

3617
  pInfo->binfo.pCtx = createSqlFunctionCtx(pExprInfo, num, &pInfo->binfo.rowCellInfoOffset);
3618
  initResultRowInfo(&pInfo->binfo.resultRowInfo, (int32_t)1);
H
Haojun Liao 已提交
3619

3620 3621 3622
  if (pInfo->binfo.pCtx == NULL || pInfo->binfo.pRes == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
3623

dengyihao's avatar
dengyihao 已提交
3624
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
3625
  int32_t code = doInitAggInfoSup(&pInfo->aggSup, pInfo->binfo.pCtx, num, keyBufSize, pTaskInfo->id.str);
3626 3627 3628
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3629

H
Haojun Liao 已提交
3630
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, num, pTaskInfo);
H
Haojun Liao 已提交
3631
  code = initGroupCol(pExprInfo, num, pGroupInfo, pInfo);
3632 3633 3634
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3635

L
Liu Jicong 已提交
3636 3637 3638 3639 3640
  //  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
3641

3642
  pOperator->resultInfo.capacity = blockDataGetCapacityInRow(pInfo->binfo.pRes, pInfo->bufPageSize);
H
Haojun Liao 已提交
3643

L
Liu Jicong 已提交
3644
  pOperator->name = "SortedMerge";
X
Xiaoyu Wang 已提交
3645
  // pOperator->operatorType = OP_SortedMerge;
3646
  pOperator->blocking = true;
L
Liu Jicong 已提交
3647 3648
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3649
  pOperator->numOfExprs = num;
L
Liu Jicong 已提交
3650
  pOperator->pExpr = pExprInfo;
H
Haojun Liao 已提交
3651

L
Liu Jicong 已提交
3652
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
3653

3654 3655
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSortedMerge, NULL, NULL, destroySortedMergeOperatorInfo,
                                         NULL, NULL, NULL);
3656 3657 3658
  code = appendDownstream(pOperator, downstream, numOfDownstream);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3659
  }
H
Haojun Liao 已提交
3660

3661
  return pOperator;
H
Haojun Liao 已提交
3662

L
Liu Jicong 已提交
3663
_error:
3664
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3665
    destroySortedMergeOperatorInfo(pInfo, num);
H
Haojun Liao 已提交
3666 3667
  }

wafwerar's avatar
wafwerar 已提交
3668 3669
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
3670 3671
  terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
  return NULL;
H
Haojun Liao 已提交
3672 3673
}

3674 3675
int32_t getTableScanOrder(SOperatorInfo* pOperator) {
  if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
H
Haojun Liao 已提交
3676
    if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
3677
      return TSDB_ORDER_ASC;
H
Haojun Liao 已提交
3678 3679
    } else {
      return getTableScanOrder(pOperator->pDownstream[0]);
3680 3681 3682 3683 3684 3685
    }
  }

  STableScanInfo* pTableScanInfo = pOperator->info;
  return pTableScanInfo->cond.order;
}
3686 3687

// this is a blocking operator
L
Liu Jicong 已提交
3688
static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
H
Haojun Liao 已提交
3689 3690
  if (OPTR_IS_OPENED(pOperator)) {
    return TSDB_CODE_SUCCESS;
3691 3692
  }

H
Haojun Liao 已提交
3693
  SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
3694
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3695

dengyihao's avatar
dengyihao 已提交
3696
  SOptrBasicInfo* pInfo = &pAggInfo->binfo;
3697

H
Haojun Liao 已提交
3698
  SOperatorInfo* downstream = pOperator->pDownstream[0];
3699

H
Haojun Liao 已提交
3700
  while (1) {
H
Haojun Liao 已提交
3701
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
3702
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
3703
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
3704 3705 3706 3707

    if (pBlock == NULL) {
      break;
    }
H
Haojun Liao 已提交
3708
    //    if (pAggInfo->current != NULL) {
3709
    //      setTagValue(pOperator, pAggInfo->current->pTable, pInfo->pCtx, pOperator->numOfExprs);
H
Haojun Liao 已提交
3710
    //    }
3711

3712 3713
    int32_t order = getTableScanOrder(pOperator);

3714 3715
    // 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
3716 3717
      int32_t code = projectApplyFunctions(pAggInfo->pScalarExprInfo, pBlock, pBlock, pAggInfo->pScalarCtx,
                                           pAggInfo->numOfScalarExpr, NULL);
3718 3719 3720 3721
      if (code != TSDB_CODE_SUCCESS) {
        pTaskInfo->code = code;
        longjmp(pTaskInfo->env, pTaskInfo->code);
      }
3722 3723
    }

3724
    // the pDataBlock are always the same one, no need to call this again
3725
    setExecutionContext(pOperator->numOfExprs, pBlock->info.groupId, pTaskInfo, pAggInfo);
3726
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, true);
H
Haojun Liao 已提交
3727
    doAggregateImpl(pOperator, 0, pInfo->pCtx);
3728

dengyihao's avatar
dengyihao 已提交
3729
#if 0  // test for encode/decode result info
3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740
    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);
      }
3741
    }
3742
#endif
3743 3744
  }

H
Haojun Liao 已提交
3745
  closeAllResultRows(&pAggInfo->binfo.resultRowInfo);
3746 3747
  finalizeMultiTupleQueryResult(pOperator->numOfExprs, pAggInfo->aggSup.pResultBuf, &pAggInfo->binfo.resultRowInfo,
                                pAggInfo->binfo.rowCellInfoOffset);
H
Haojun Liao 已提交
3748

3749
  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, false);
H
Haojun Liao 已提交
3750 3751 3752 3753
  OPTR_SET_OPENED(pOperator);
  return TSDB_CODE_SUCCESS;
}

3754
static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3755
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3756 3757 3758 3759 3760 3761
  SOptrBasicInfo*   pInfo = &pAggInfo->binfo;

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

L
Liu Jicong 已提交
3762
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3763
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
H
Haojun Liao 已提交
3764 3765 3766 3767
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
    return NULL;
  }

H
Haojun Liao 已提交
3768
  blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
3769
  doBuildResultDatablock(pTaskInfo, pInfo, &pAggInfo->groupResInfo, pOperator->pExpr, pAggInfo->aggSup.pResultBuf);
H
Haojun Liao 已提交
3770 3771 3772
  if (pInfo->pRes->info.rows == 0 || !hasRemainDataInCurrentGroup(&pAggInfo->groupResInfo)) {
    doSetOperatorCompleted(pOperator);
  }
3773

H
Haojun Liao 已提交
3774
  doSetOperatorCompleted(pOperator);
L
Liu Jicong 已提交
3775
  return (blockDataGetNumOfRows(pInfo->pRes) != 0) ? pInfo->pRes : NULL;
3776 3777
}

dengyihao's avatar
dengyihao 已提交
3778 3779
void aggEncodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char** result,
                        int32_t* length) {
wmmhello's avatar
wmmhello 已提交
3780
  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
3781
  size_t  keyLen = sizeof(uint64_t) * 2;  // estimate the key length
wmmhello's avatar
wmmhello 已提交
3782
  int32_t totalSize = sizeof(int32_t) + size * (sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize);
wafwerar's avatar
wafwerar 已提交
3783
  *result = taosMemoryCalloc(1, totalSize);
L
Liu Jicong 已提交
3784
  if (*result == NULL) {
3785
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
wmmhello's avatar
wmmhello 已提交
3786 3787 3788
  }
  *(int32_t*)(*result) = size;
  int32_t offset = sizeof(int32_t);
3789 3790

  // prepare memory
3791
  SResultRowPosition* pos = &pInfo->resultRowInfo.cur;
dengyihao's avatar
dengyihao 已提交
3792 3793
  void*               pPage = getBufPage(pSup->pResultBuf, pos->pageId);
  SResultRow*         pRow = (SResultRow*)((char*)pPage + pos->offset);
3794 3795 3796
  setBufPageDirty(pPage, true);
  releaseBufPage(pSup->pResultBuf, pPage);

dengyihao's avatar
dengyihao 已提交
3797
  void* pIter = taosHashIterate(pSup->pResultRowHashTable, NULL);
wmmhello's avatar
wmmhello 已提交
3798
  while (pIter) {
dengyihao's avatar
dengyihao 已提交
3799
    void*               key = taosHashGetKey(pIter, &keyLen);
3800
    SResultRowPosition* p1 = (SResultRowPosition*)pIter;
3801

dengyihao's avatar
dengyihao 已提交
3802
    pPage = (SFilePage*)getBufPage(pSup->pResultBuf, p1->pageId);
3803
    pRow = (SResultRow*)((char*)pPage + p1->offset);
3804 3805
    setBufPageDirty(pPage, true);
    releaseBufPage(pSup->pResultBuf, pPage);
wmmhello's avatar
wmmhello 已提交
3806 3807 3808

    // recalculate the result size
    int32_t realTotalSize = offset + sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize;
L
Liu Jicong 已提交
3809 3810 3811
    if (realTotalSize > totalSize) {
      char* tmp = taosMemoryRealloc(*result, realTotalSize);
      if (tmp == NULL) {
wmmhello's avatar
wmmhello 已提交
3812
        terrno = TSDB_CODE_OUT_OF_MEMORY;
wafwerar's avatar
wafwerar 已提交
3813
        taosMemoryFree(*result);
wmmhello's avatar
wmmhello 已提交
3814
        *result = NULL;
3815
        longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
L
Liu Jicong 已提交
3816
      } else {
wmmhello's avatar
wmmhello 已提交
3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828
        *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);
3829
    memcpy(*result + offset, pRow, pSup->resultRowSize);
wmmhello's avatar
wmmhello 已提交
3830 3831 3832 3833 3834
    offset += pSup->resultRowSize;

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

L
Liu Jicong 已提交
3835
  if (length) {
wmmhello's avatar
wmmhello 已提交
3836 3837 3838 3839 3840
    *length = offset;
  }
  return;
}

dengyihao's avatar
dengyihao 已提交
3841 3842
bool aggDecodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char* result,
                        int32_t length) {
L
Liu Jicong 已提交
3843
  if (!result || length <= 0) {
wmmhello's avatar
wmmhello 已提交
3844 3845 3846 3847 3848 3849 3850
    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
3851
  while (count-- > 0 && length > offset) {
wmmhello's avatar
wmmhello 已提交
3852 3853 3854
    int32_t keyLen = *(int32_t*)(result + offset);
    offset += sizeof(int32_t);

L
Liu Jicong 已提交
3855 3856 3857
    uint64_t    tableGroupId = *(uint64_t*)(result + offset);
    SResultRow* resultRow = getNewResultRow_rv(pSup->pResultBuf, tableGroupId, pSup->resultRowSize);
    if (!resultRow) {
3858
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3859
    }
3860

wmmhello's avatar
wmmhello 已提交
3861
    // add a new result set for a new group
3862 3863
    SResultRowPosition pos = {.pageId = resultRow->pageId, .offset = resultRow->offset};
    taosHashPut(pSup->pResultRowHashTable, result + offset, keyLen, &pos, sizeof(SResultRowPosition));
wmmhello's avatar
wmmhello 已提交
3864 3865 3866

    offset += keyLen;
    int32_t valueLen = *(int32_t*)(result + offset);
L
Liu Jicong 已提交
3867
    if (valueLen != pSup->resultRowSize) {
3868
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3869 3870 3871 3872 3873 3874 3875 3876 3877 3878
    }
    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);
3879
    prepareResultListBuffer(&pInfo->resultRowInfo, pOperator->pTaskInfo->env);
dengyihao's avatar
dengyihao 已提交
3880
    //    pInfo->resultRowInfo.cur = pInfo->resultRowInfo.size;
L
Liu Jicong 已提交
3881 3882
    //    pInfo->resultRowInfo.pPosition[pInfo->resultRowInfo.size++] =
    //        (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
dengyihao's avatar
dengyihao 已提交
3883
    pInfo->resultRowInfo.cur = (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
wmmhello's avatar
wmmhello 已提交
3884 3885
  }

L
Liu Jicong 已提交
3886
  if (offset != length) {
3887
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3888 3889 3890 3891
  }
  return true;
}

3892 3893
enum {
  PROJECT_RETRIEVE_CONTINUE = 0x1,
L
Liu Jicong 已提交
3894
  PROJECT_RETRIEVE_DONE = 0x2,
3895 3896 3897 3898 3899
};

static int32_t handleLimitOffset(SOperatorInfo* pOperator, SSDataBlock* pBlock) {
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
L
Liu Jicong 已提交
3900
  SSDataBlock*          pRes = pInfo->pRes;
3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948

  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;
  }

3949 3950 3951
  // 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);
3952 3953 3954 3955 3956

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

3957
    return PROJECT_RETRIEVE_DONE;
3958
  }
3959

3960
  // todo optimize performance
3961 3962
  // 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
3963 3964
  if (pRes->info.rows >= pOperator->resultInfo.threshold || pProjectInfo->slimit.offset != -1 ||
      pProjectInfo->slimit.limit != -1) {
3965
    return PROJECT_RETRIEVE_DONE;
L
Liu Jicong 已提交
3966
  } else {  // not full enough, continue to accumulate the output data in the buffer.
3967 3968 3969 3970
    return PROJECT_RETRIEVE_CONTINUE;
  }
}

3971
static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
3972
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
L
Liu Jicong 已提交
3973
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
3974 3975

  SSDataBlock* pRes = pInfo->pRes;
3976
  blockDataCleanup(pRes);
3977

3978
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3979 3980 3981
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }
dengyihao's avatar
dengyihao 已提交
3982

H
Haojun Liao 已提交
3983
#if 0
3984 3985 3986 3987 3988 3989
  if (pProjectInfo->existDataBlock) {  // TODO refactor
    SSDataBlock* pBlock = pProjectInfo->existDataBlock;
    pProjectInfo->existDataBlock = NULL;
    *newgroup = true;

    // todo dynamic set tags
L
Liu Jicong 已提交
3990
    //    if (pTableQueryInfo != NULL) {
3991
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
3992
    //    }
3993 3994

    // the pDataBlock are always the same one, no need to call this again
H
Haojun Liao 已提交
3995
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, TSDB_ORDER_ASC);
3996

H
Haojun Liao 已提交
3997
    blockDataEnsureCapacity(pInfo->pRes, pBlock->info.rows);
3998
    projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
3999
    if (pRes->info.rows >= pProjectInfo->binfo.capacity * 0.8) {
4000 4001
      copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfExprs);
      resetResultRowEntryResult(pInfo->pCtx, pOperator->numOfExprs);
4002 4003 4004
      return pRes;
    }
  }
H
Haojun Liao 已提交
4005
#endif
4006

H
Haojun Liao 已提交
4007 4008
  SOperatorInfo* downstream = pOperator->pDownstream[0];

L
Liu Jicong 已提交
4009
  while (1) {
H
Haojun Liao 已提交
4010
    // The downstream exec may change the value of the newgroup, so use a local variable instead.
H
Haojun Liao 已提交
4011
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4012
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
4013
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4014 4015

    if (pBlock == NULL) {
4016
      setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
4017 4018 4019 4020
      break;
    }

    // Return result of the previous group in the firstly.
4021
    if (false) {
4022 4023 4024
      if (pRes->info.rows > 0) {
        pProjectInfo->existDataBlock = pBlock;
        break;
L
Liu Jicong 已提交
4025
      } else {  // init output buffer for a new group data
4026
        initCtxOutputBuffer(pInfo->pCtx, pOperator->numOfExprs);
4027 4028 4029
      }
    }

4030 4031
    // todo set tags

H
Haojun Liao 已提交
4032 4033
    //    STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
    //    if (pTableQueryInfo != NULL) {
4034
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfExprs);
H
Haojun Liao 已提交
4035
    //    }
4036 4037

    // the pDataBlock are always the same one, no need to call this again
4038 4039 4040
    int32_t order = getTableScanOrder(pOperator->pDownstream[0]);

    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, false);
4041 4042
    blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);

4043
    pTaskInfo->code = projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs,
dengyihao's avatar
dengyihao 已提交
4044
                          pProjectInfo->pPseudoColInfo);
4045 4046 4047 4048
    if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
      longjmp(pTaskInfo->env, pTaskInfo->code);
    }

4049 4050
    int32_t status = handleLimitOffset(pOperator, pBlock);
    if (status == PROJECT_RETRIEVE_CONTINUE) {
H
Haojun Liao 已提交
4051
      continue;
L
Liu Jicong 已提交
4052
    } else if (status == PROJECT_RETRIEVE_DONE) {
4053 4054 4055
      break;
    }
  }
dengyihao's avatar
dengyihao 已提交
4056

H
Haojun Liao 已提交
4057
  pProjectInfo->curOutput += pInfo->pRes->info.rows;
H
Haojun Liao 已提交
4058

4059
  //  copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
4060
  return (pInfo->pRes->info.rows > 0) ? pInfo->pRes : NULL;
4061 4062
}

L
Liu Jicong 已提交
4063 4064
static void doHandleRemainBlockForNewGroupImpl(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                               SExecTaskInfo* pTaskInfo) {
4065
  pInfo->totalInputRows = pInfo->existNewGroupBlock->info.rows;
H
Haojun Liao 已提交
4066

L
Liu Jicong 已提交
4067 4068
  int64_t ekey = Q_STATUS_EQUAL(pTaskInfo->status, TASK_COMPLETED) ? pTaskInfo->window.ekey
                                                                   : pInfo->existNewGroupBlock->info.window.ekey;
4069 4070
  taosResetFillInfo(pInfo->pFillInfo, getFillInfoStart(pInfo->pFillInfo));

4071
  taosFillSetStartInfo(pInfo->pFillInfo, pInfo->existNewGroupBlock->info.rows, ekey);
4072 4073
  taosFillSetInputDataBlock(pInfo->pFillInfo, pInfo->existNewGroupBlock);

4074
  doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, pResultInfo->capacity);
4075 4076 4077 4078
  pInfo->existNewGroupBlock = NULL;
  *newgroup = true;
}

L
Liu Jicong 已提交
4079 4080
static void doHandleRemainBlockFromNewGroup(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                            SExecTaskInfo* pTaskInfo) {
4081 4082
  if (taosFillHasMoreResults(pInfo->pFillInfo)) {
    *newgroup = false;
4083
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, (int32_t)pResultInfo->capacity);
H
Haojun Liao 已提交
4084
    if (pInfo->pRes->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult)) {
4085 4086 4087 4088 4089 4090
      return;
    }
  }

  // handle the cached new group data block
  if (pInfo->existNewGroupBlock) {
4091
    doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4092 4093 4094
  }
}

4095
static SSDataBlock* doFill(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
4096 4097
  SFillOperatorInfo* pInfo = pOperator->info;
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
4098

H
Haojun Liao 已提交
4099
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4100 4101 4102
  SSDataBlock* pResBlock = pInfo->pRes;

  blockDataCleanup(pResBlock);
4103 4104 4105 4106
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }

4107
  // todo handle different group data interpolation
X
Xiaoyu Wang 已提交
4108 4109
  bool  n = false;
  bool* newgroup = &n;
4110
  doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4111 4112
  if (pResBlock->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult && pResBlock->info.rows > 0)) {
    return pResBlock;
H
Haojun Liao 已提交
4113
  }
4114

H
Haojun Liao 已提交
4115
  SOperatorInfo* pDownstream = pOperator->pDownstream[0];
L
Liu Jicong 已提交
4116
  while (1) {
H
Haojun Liao 已提交
4117
    publishOperatorProfEvent(pDownstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4118
    SSDataBlock* pBlock = pDownstream->fpSet.getNextFn(pDownstream);
H
Haojun Liao 已提交
4119
    publishOperatorProfEvent(pDownstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130

    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
4131
      taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4132 4133 4134 4135 4136 4137 4138
    } else {
      if (pBlock == NULL) {
        if (pInfo->totalInputRows == 0) {
          pOperator->status = OP_EXEC_DONE;
          return NULL;
        }

4139
        taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4140 4141 4142 4143 4144 4145 4146
      } else {
        pInfo->totalInputRows += pBlock->info.rows;
        taosFillSetStartInfo(pInfo->pFillInfo, pBlock->info.rows, pBlock->info.window.ekey);
        taosFillSetInputDataBlock(pInfo->pFillInfo, pBlock);
      }
    }

4147 4148
    blockDataEnsureCapacity(pResBlock, pOperator->resultInfo.capacity);
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pResBlock, pOperator->resultInfo.capacity);
4149 4150

    // current group has no more result to return
4151
    if (pResBlock->info.rows > 0) {
4152 4153
      // 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
4154 4155
      if (pResBlock->info.rows > pResultInfo->threshold || pBlock == NULL || (!pInfo->multigroupResult)) {
        return pResBlock;
4156 4157
      }

4158
      doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4159 4160
      if (pResBlock->info.rows > pOperator->resultInfo.threshold || pBlock == NULL) {
        return pResBlock;
4161 4162 4163
      }
    } else if (pInfo->existNewGroupBlock) {  // try next group
      assert(pBlock != NULL);
4164
      doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4165 4166
      if (pResBlock->info.rows > pResultInfo->threshold) {
        return pResBlock;
4167 4168 4169 4170 4171 4172 4173 4174
      }
    } else {
      return NULL;
    }
  }
}

// todo set the attribute of query scan count
H
Haojun Liao 已提交
4175
static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr) {
L
Liu Jicong 已提交
4176
  for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190
    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;
  }

4191
  if (pOperator->fpSet.closeFn != NULL) {
4192
    pOperator->fpSet.closeFn(pOperator->info, pOperator->numOfExprs);
4193 4194
  }

H
Haojun Liao 已提交
4195
  if (pOperator->pDownstream != NULL) {
L
Liu Jicong 已提交
4196
    for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
H
Haojun Liao 已提交
4197
      destroyOperatorInfo(pOperator->pDownstream[i]);
4198 4199
    }

wafwerar's avatar
wafwerar 已提交
4200
    taosMemoryFreeClear(pOperator->pDownstream);
H
Haojun Liao 已提交
4201
    pOperator->numOfDownstream = 0;
4202 4203
  }

H
Haojun Liao 已提交
4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214
  if (pOperator->pExpr != NULL) {
    for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
      SExprInfo* pExprInfo = &pOperator->pExpr[i];
      if (pExprInfo->pExpr->nodeType == QUERY_NODE_COLUMN) {
        taosMemoryFree(pExprInfo->base.pParam[0].pCol);
      }
      taosMemoryFree(pExprInfo->base.pParam);
      taosMemoryFree(pExprInfo->pExpr);
    }
  }

D
fix bug  
dapan 已提交
4215
  taosMemoryFreeClear(pOperator->pExpr);
wafwerar's avatar
wafwerar 已提交
4216 4217
  taosMemoryFreeClear(pOperator->info);
  taosMemoryFreeClear(pOperator);
4218 4219
}

dengyihao's avatar
dengyihao 已提交
4220 4221
int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                         const char* pKey) {
4222 4223
  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);

dengyihao's avatar
dengyihao 已提交
4224 4225
  pAggSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
  pAggSup->keyBuf = taosMemoryCalloc(1, keyBufSize + POINTER_BYTES + sizeof(int64_t));
4226 4227
  pAggSup->pResultRowHashTable = taosHashInit(10, hashFn, true, HASH_NO_LOCK);

H
Haojun Liao 已提交
4228
  if (pAggSup->keyBuf == NULL || pAggSup->pResultRowHashTable == NULL) {
4229 4230 4231
    return TSDB_CODE_OUT_OF_MEMORY;
  }

H
Haojun Liao 已提交
4232
  uint32_t defaultPgsz = 4096;
X
Xiaoyu Wang 已提交
4233
  while (defaultPgsz < pAggSup->resultRowSize * 4) {
H
Haojun Liao 已提交
4234 4235 4236 4237 4238 4239 4240 4241 4242 4243
    defaultPgsz <<= 1u;
  }

  // at least four pages need to be in buffer
  int32_t defaultBufsz = 4096 * 256;
  if (defaultBufsz <= defaultPgsz) {
    defaultBufsz = defaultPgsz * 4;
  }

  int32_t code = createDiskbasedBuf(&pAggSup->pResultBuf, defaultPgsz, defaultBufsz, pKey, "/tmp/");
H
Haojun Liao 已提交
4244 4245 4246 4247
  if (code != TSDB_CODE_SUCCESS) {
    return code;
  }

4248 4249 4250
  return TSDB_CODE_SUCCESS;
}

4251
void cleanupAggSup(SAggSupporter* pAggSup) {
wafwerar's avatar
wafwerar 已提交
4252
  taosMemoryFreeClear(pAggSup->keyBuf);
4253
  taosHashCleanup(pAggSup->pResultRowHashTable);
H
Haojun Liao 已提交
4254
  destroyDiskbasedBuf(pAggSup->pResultBuf);
4255 4256
}

H
Haojun Liao 已提交
4257
int32_t initAggInfo(SOptrBasicInfo* pBasicInfo, SAggSupporter* pAggSup, SExprInfo* pExprInfo, int32_t numOfCols,
4258
                    SSDataBlock* pResultBlock, size_t keyBufSize, const char* pkey) {
4259
  pBasicInfo->pCtx = createSqlFunctionCtx(pExprInfo, numOfCols, &pBasicInfo->rowCellInfoOffset);
H
Haojun Liao 已提交
4260 4261
  pBasicInfo->pRes = pResultBlock;

4262
  doInitAggInfoSup(pAggSup, pBasicInfo->pCtx, numOfCols, keyBufSize, pkey);
4263

L
Liu Jicong 已提交
4264
  for (int32_t i = 0; i < numOfCols; ++i) {
4265 4266 4267
    pBasicInfo->pCtx[i].pBuf = pAggSup->pResultBuf;
  }

4268
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
4269 4270
}

4271 4272 4273 4274 4275 4276 4277 4278 4279
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
4280
static STableQueryInfo* initTableQueryInfo(const STableGroupInfo* pTableGroupInfo) {
L
Liu Jicong 已提交
4281 4282 4283 4284
  if (pTableGroupInfo->numOfTables == 0) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
4285
  STableQueryInfo* pTableQueryInfo = taosMemoryCalloc(pTableGroupInfo->numOfTables, sizeof(STableQueryInfo));
H
Haojun Liao 已提交
4286 4287
  if (pTableQueryInfo == NULL) {
    return NULL;
H
Haojun Liao 已提交
4288
  }
H
Haojun Liao 已提交
4289 4290

  int32_t index = 0;
L
Liu Jicong 已提交
4291
  for (int32_t i = 0; i < taosArrayGetSize(pTableGroupInfo->pGroupList); ++i) {
H
Haojun Liao 已提交
4292
    SArray* pa = taosArrayGetP(pTableGroupInfo->pGroupList, i);
L
Liu Jicong 已提交
4293
    for (int32_t j = 0; j < taosArrayGetSize(pa); ++j) {
H
Haojun Liao 已提交
4294
      STableKeyInfo* pk = taosArrayGet(pa, j);
H
Haojun Liao 已提交
4295
      STableQueryInfo* pTQueryInfo = &pTableQueryInfo[index++];
L
Liu Jicong 已提交
4296
      pTQueryInfo->lastKey = pk->lastKey;
H
Haojun Liao 已提交
4297 4298
    }
  }
H
Haojun Liao 已提交
4299 4300

  STimeWindow win = {0, INT64_MAX};
4301
  createTableQueryInfo(pTableQueryInfo, win);
H
Haojun Liao 已提交
4302
  return pTableQueryInfo;
H
Haojun Liao 已提交
4303 4304
}

L
Liu Jicong 已提交
4305
SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
4306
                                           SSDataBlock* pResultBlock, SExprInfo* pScalarExprInfo,
L
Liu Jicong 已提交
4307 4308
                                           int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo,
                                           const STableGroupInfo* pTableGroupInfo) {
wafwerar's avatar
wafwerar 已提交
4309
  SAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SAggOperatorInfo));
L
Liu Jicong 已提交
4310
  SOperatorInfo*    pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4311 4312 4313
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
4314

4315
  int32_t numOfRows = 10;
dengyihao's avatar
dengyihao 已提交
4316
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4317 4318

  initResultSizeInfo(pOperator, numOfRows);
dengyihao's avatar
dengyihao 已提交
4319 4320
  int32_t code =
      initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResultBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4321
  pInfo->pTableQueryInfo = initTableQueryInfo(pTableGroupInfo);
L
Liu Jicong 已提交
4322
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4323 4324
    goto _error;
  }
H
Haojun Liao 已提交
4325

H
Haojun Liao 已提交
4326 4327 4328 4329
  int32_t numOfGroup = 10;  // todo replaced with true value
  pInfo->groupId = INT32_MIN;
  initResultRowInfo(&pInfo->binfo.resultRowInfo, numOfGroup);

4330 4331
  pInfo->pScalarExprInfo = pScalarExprInfo;
  pInfo->numOfScalarExpr = numOfScalarExpr;
4332 4333 4334
  if (pInfo->pScalarExprInfo != NULL) {
    pInfo->pScalarCtx = createSqlFunctionCtx(pScalarExprInfo, numOfCols, &pInfo->rowCellInfoOffset);
  }
4335

dengyihao's avatar
dengyihao 已提交
4336
  pOperator->name = "TableAggregate";
X
Xiaoyu Wang 已提交
4337
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_AGG;
4338
  pOperator->blocking = true;
dengyihao's avatar
dengyihao 已提交
4339 4340 4341
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
4342
  pOperator->numOfExprs = numOfCols;
dengyihao's avatar
dengyihao 已提交
4343
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
4344

4345 4346
  pOperator->fpSet = createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, NULL, destroyAggOperatorInfo,
                                         aggEncodeResultRow, aggDecodeResultRow, NULL);
H
Haojun Liao 已提交
4347 4348 4349 4350 4351

  code = appendDownstream(pOperator, &downstream, 1);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4352 4353

  return pOperator;
L
Liu Jicong 已提交
4354
_error:
H
Haojun Liao 已提交
4355
  destroyAggOperatorInfo(pInfo, numOfCols);
wafwerar's avatar
wafwerar 已提交
4356 4357
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
4358 4359
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4360 4361
}

H
Haojun Liao 已提交
4362
void doDestroyBasicInfo(SOptrBasicInfo* pInfo, int32_t numOfOutput) {
4363 4364
  assert(pInfo != NULL);

4365
  destroySqlFunctionCtx(pInfo->pCtx, numOfOutput);
wafwerar's avatar
wafwerar 已提交
4366
  taosMemoryFreeClear(pInfo->rowCellInfoOffset);
4367 4368

  cleanupResultRowInfo(&pInfo->resultRowInfo);
H
Haojun Liao 已提交
4369
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
4370 4371
}

H
Haojun Liao 已提交
4372
void destroyBasicOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4373
  SOptrBasicInfo* pInfo = (SOptrBasicInfo*)param;
4374 4375
  doDestroyBasicInfo(pInfo, numOfOutput);
}
H
Haojun Liao 已提交
4376

H
Haojun Liao 已提交
4377
void destroyMergeJoinOperator(void* param, int32_t numOfOutput) {
X
Xiaoyu Wang 已提交
4378
  SJoinOperatorInfo* pJoinOperator = (SJoinOperatorInfo*)param;
H
Haojun Liao 已提交
4379 4380
}

H
Haojun Liao 已提交
4381
void destroyAggOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4382
  SAggOperatorInfo* pInfo = (SAggOperatorInfo*)param;
4383 4384
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
}
4385

H
Haojun Liao 已提交
4386
void destroySFillOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4387
  SFillOperatorInfo* pInfo = (SFillOperatorInfo*)param;
4388
  pInfo->pFillInfo = taosDestroyFillInfo(pInfo->pFillInfo);
H
Haojun Liao 已提交
4389
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
wafwerar's avatar
wafwerar 已提交
4390
  taosMemoryFreeClear(pInfo->p);
4391 4392
}

H
Haojun Liao 已提交
4393
static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
D
fix bug  
dapan 已提交
4394 4395 4396
  if (NULL == param) {
    return;
  }
L
Liu Jicong 已提交
4397
  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
4398
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
H
Haojun Liao 已提交
4399
  cleanupAggSup(&pInfo->aggSup);
H
Haojun Liao 已提交
4400
  taosArrayDestroy(pInfo->pPseudoColInfo);
4401 4402
}

H
Haojun Liao 已提交
4403
void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4404
  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
H
Haojun Liao 已提交
4405 4406 4407 4408 4409 4410 4411 4412 4413
  taosArrayDestroy(pExInfo->pSources);
  taosArrayDestroy(pExInfo->pSourceDataInfo);
  if (pExInfo->pResult != NULL) {
    blockDataDestroy(pExInfo->pResult);
  }

  tsem_destroy(&pExInfo->ready);
}

H
Haojun Liao 已提交
4414 4415
static SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
  SArray* pList = taosArrayInit(4, sizeof(int32_t));
dengyihao's avatar
dengyihao 已提交
4416
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
4417 4418 4419 4420 4421 4422 4423 4424
    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
      taosArrayPush(pList, &i);
    }
  }

  return pList;
}

L
Liu Jicong 已提交
4425
SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t num,
dengyihao's avatar
dengyihao 已提交
4426 4427
                                         SSDataBlock* pResBlock, SLimit* pLimit, SLimit* pSlimit,
                                         SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4428
  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
L
Liu Jicong 已提交
4429
  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4430 4431 4432
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
4433

dengyihao's avatar
dengyihao 已提交
4434 4435 4436
  pInfo->limit = *pLimit;
  pInfo->slimit = *pSlimit;
  pInfo->curOffset = pLimit->offset;
H
Haojun Liao 已提交
4437 4438
  pInfo->curSOffset = pSlimit->offset;

H
Haojun Liao 已提交
4439
  pInfo->binfo.pRes = pResBlock;
H
Haojun Liao 已提交
4440 4441 4442

  int32_t numOfCols = num;
  int32_t numOfRows = 4096;
dengyihao's avatar
dengyihao 已提交
4443
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4444 4445 4446

  initResultSizeInfo(pOperator, numOfRows);
  initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4447
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfCols, pTaskInfo);
H
Haojun Liao 已提交
4448
  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pInfo->binfo.pCtx, numOfCols);
4449

dengyihao's avatar
dengyihao 已提交
4450
  pOperator->name = "ProjectOperator";
H
Haojun Liao 已提交
4451
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
4452
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4453 4454 4455
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
4456
  pOperator->numOfExprs = num;
4457

L
Liu Jicong 已提交
4458 4459
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
                                         destroyProjectOperatorInfo, NULL, NULL, NULL);
L
Liu Jicong 已提交
4460 4461

  pOperator->pTaskInfo = pTaskInfo;
4462
  int32_t code = appendDownstream(pOperator, &downstream, 1);
H
Haojun Liao 已提交
4463
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4464 4465
    goto _error;
  }
4466 4467

  return pOperator;
H
Haojun Liao 已提交
4468

L
Liu Jicong 已提交
4469
_error:
H
Haojun Liao 已提交
4470 4471
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4472 4473
}

4474
static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SNodeListNode* pValNode,
L
Liu Jicong 已提交
4475
                            STimeWindow win, int32_t capacity, const char* id, SInterval* pInterval, int32_t fillType) {
4476
  SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pValNode);
H
Haojun Liao 已提交
4477 4478

  STimeWindow w = TSWINDOW_INITIALIZER;
4479
  getAlignQueryTimeWindow(pInterval, pInterval->precision, win.skey, &w);
H
Haojun Liao 已提交
4480 4481

  int32_t order = TSDB_ORDER_ASC;
4482
  pInfo->pFillInfo = taosCreateFillInfo(order, w.skey, 0, capacity, numOfCols, pInterval, fillType, pColInfo, id);
H
Haojun Liao 已提交
4483

wafwerar's avatar
wafwerar 已提交
4484
  pInfo->p = taosMemoryCalloc(numOfCols, POINTER_BYTES);
H
Haojun Liao 已提交
4485 4486 4487 4488 4489 4490 4491
  if (pInfo->pFillInfo == NULL || pInfo->p == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  } else {
    return TSDB_CODE_SUCCESS;
  }
}

L
Liu Jicong 已提交
4492
SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExpr, int32_t numOfCols,
X
Xiaoyu Wang 已提交
4493 4494 4495
                                      SInterval* pInterval, STimeWindow* pWindow, SSDataBlock* pResBlock,
                                      int32_t fillType, SNodeListNode* pValueNode, bool multigroupResult,
                                      SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4496
  SFillOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SFillOperatorInfo));
L
Liu Jicong 已提交
4497
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4498

L
Liu Jicong 已提交
4499
  pInfo->pRes = pResBlock;
4500 4501
  pInfo->multigroupResult = multigroupResult;

4502 4503
  int32_t type = TSDB_FILL_NONE;
  switch (fillType) {
dengyihao's avatar
dengyihao 已提交
4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521
    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;
4522 4523 4524 4525
    default:
      type = TSDB_FILL_NONE;
  }

H
Haojun Liao 已提交
4526
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4527 4528
  initResultSizeInfo(pOperator, 4096);

X
Xiaoyu Wang 已提交
4529 4530
  int32_t code = initFillInfo(pInfo, pExpr, numOfCols, pValueNode, *pWindow, pResultInfo->capacity, pTaskInfo->id.str,
                              pInterval, type);
4531 4532 4533
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4534

dengyihao's avatar
dengyihao 已提交
4535
  pOperator->name = "FillOperator";
4536
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4537
  pOperator->status = OP_NOT_OPENED;
4538
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_FILL;
dengyihao's avatar
dengyihao 已提交
4539
  pOperator->pExpr = pExpr;
4540
  pOperator->numOfExprs = numOfCols;
dengyihao's avatar
dengyihao 已提交
4541
  pOperator->info = pInfo;
H
Haojun Liao 已提交
4542

L
Liu Jicong 已提交
4543 4544
  pOperator->fpSet =
      createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, NULL, destroySFillOperatorInfo, NULL, NULL, NULL);
4545
  pOperator->pTaskInfo = pTaskInfo;
4546
  code = appendDownstream(pOperator, &downstream, 1);
4547
  return pOperator;
H
Haojun Liao 已提交
4548

L
Liu Jicong 已提交
4549
_error:
wafwerar's avatar
wafwerar 已提交
4550 4551
  taosMemoryFreeClear(pOperator);
  taosMemoryFreeClear(pInfo);
H
Haojun Liao 已提交
4552
  return NULL;
4553 4554
}

L
Liu Jicong 已提交
4555 4556
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
4557
  SResSchema s = {0};
dengyihao's avatar
dengyihao 已提交
4558 4559 4560 4561
  s.scale = scale;
  s.type = type;
  s.bytes = bytes;
  s.slotId = slotId;
H
Haojun Liao 已提交
4562
  s.precision = precision;
H
Haojun Liao 已提交
4563 4564 4565 4566
  strncpy(s.name, name, tListLen(s.name));

  return s;
}
H
Haojun Liao 已提交
4567

H
Haojun Liao 已提交
4568 4569 4570 4571 4572 4573 4574
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
4575 4576 4577 4578 4579
  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
4580 4581 4582 4583 4584
  pCol->dataBlockId = blockId;

  return pCol;
}

H
Haojun Liao 已提交
4585
SExprInfo* createExprInfo(SNodeList* pNodeList, SNodeList* pGroupKeys, int32_t* numOfExprs) {
H
Haojun Liao 已提交
4586
  int32_t numOfFuncs = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
4587 4588 4589 4590
  int32_t numOfGroupKeys = 0;
  if (pGroupKeys != NULL) {
    numOfGroupKeys = LIST_LENGTH(pGroupKeys);
  }
H
Haojun Liao 已提交
4591

H
Haojun Liao 已提交
4592
  *numOfExprs = numOfFuncs + numOfGroupKeys;
wafwerar's avatar
wafwerar 已提交
4593
  SExprInfo* pExprs = taosMemoryCalloc(*numOfExprs, sizeof(SExprInfo));
H
Haojun Liao 已提交
4594

L
Liu Jicong 已提交
4595
  for (int32_t i = 0; i < (*numOfExprs); ++i) {
H
Haojun Liao 已提交
4596 4597 4598 4599 4600 4601
    STargetNode* pTargetNode = NULL;
    if (i < numOfFuncs) {
      pTargetNode = (STargetNode*)nodesListGetNode(pNodeList, i);
    } else {
      pTargetNode = (STargetNode*)nodesListGetNode(pGroupKeys, i - numOfFuncs);
    }
H
Haojun Liao 已提交
4602

4603
    SExprInfo* pExp = &pExprs[i];
H
Haojun Liao 已提交
4604

wafwerar's avatar
wafwerar 已提交
4605
    pExp->pExpr = taosMemoryCalloc(1, sizeof(tExprNode));
H
Haojun Liao 已提交
4606
    pExp->pExpr->_function.num = 1;
H
Haojun Liao 已提交
4607
    pExp->pExpr->_function.functionId = -1;
H
Haojun Liao 已提交
4608

4609
    int32_t type = nodeType(pTargetNode->pExpr);
H
Haojun Liao 已提交
4610
    // it is a project query, or group by column
4611
    if (type == QUERY_NODE_COLUMN) {
H
Haojun Liao 已提交
4612
      pExp->pExpr->nodeType = QUERY_NODE_COLUMN;
L
Liu Jicong 已提交
4613
      SColumnNode* pColNode = (SColumnNode*)pTargetNode->pExpr;
H
Haojun Liao 已提交
4614

G
Ganlin Zhao 已提交
4615 4616 4617
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

H
Haojun Liao 已提交
4618
      SDataType* pType = &pColNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4619 4620
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pColNode->colName);
H
Haojun Liao 已提交
4621 4622
      pExp->base.pParam[0].pCol = createColumn(pColNode->dataBlockId, pColNode->slotId, pType);
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_COLUMN;
4623
    } else if (type == QUERY_NODE_VALUE) {
4624 4625 4626 4627 4628 4629 4630
      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
4631 4632
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pValNode->node.aliasName);
4633 4634
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_VALUE;
      valueNodeToVariant(pValNode, &pExp->base.pParam[0].param);
4635
    } else if (type == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
4636
      pExp->pExpr->nodeType = QUERY_NODE_FUNCTION;
H
Haojun Liao 已提交
4637 4638 4639
      SFunctionNode* pFuncNode = (SFunctionNode*)pTargetNode->pExpr;

      SDataType* pType = &pFuncNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4640 4641
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pFuncNode->node.aliasName);
H
Haojun Liao 已提交
4642

H
Haojun Liao 已提交
4643
      pExp->pExpr->_function.functionId = pFuncNode->funcId;
H
Haojun Liao 已提交
4644
      pExp->pExpr->_function.pFunctNode = pFuncNode;
dengyihao's avatar
dengyihao 已提交
4645 4646
      strncpy(pExp->pExpr->_function.functionName, pFuncNode->functionName,
              tListLen(pExp->pExpr->_function.functionName));
H
Haojun Liao 已提交
4647 4648

      int32_t numOfParam = LIST_LENGTH(pFuncNode->pParameterList);
G
Ganlin Zhao 已提交
4649 4650 4651 4652

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

H
Haojun Liao 已提交
4653
      for (int32_t j = 0; j < numOfParam; ++j) {
4654
        SNode* p1 = nodesListGetNode(pFuncNode->pParameterList, j);
G
Ganlin Zhao 已提交
4655
        if (p1->type == QUERY_NODE_COLUMN) {
dengyihao's avatar
dengyihao 已提交
4656
          SColumnNode* pcn = (SColumnNode*)p1;
G
Ganlin Zhao 已提交
4657 4658

          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_COLUMN;
H
Haojun Liao 已提交
4659
          pExp->base.pParam[j].pCol = createColumn(pcn->dataBlockId, pcn->slotId, &pcn->node.resType);
G
Ganlin Zhao 已提交
4660 4661 4662
        } else if (p1->type == QUERY_NODE_VALUE) {
          SValueNode* pvn = (SValueNode*)p1;
          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_VALUE;
4663
          valueNodeToVariant(pvn, &pExp->base.pParam[j].param);
G
Ganlin Zhao 已提交
4664
        }
H
Haojun Liao 已提交
4665
      }
4666
    } else if (type == QUERY_NODE_OPERATOR) {
H
Haojun Liao 已提交
4667
      pExp->pExpr->nodeType = QUERY_NODE_OPERATOR;
L
Liu Jicong 已提交
4668
      SOperatorNode* pNode = (SOperatorNode*)pTargetNode->pExpr;
4669

G
Ganlin Zhao 已提交
4670 4671 4672
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

4673
      SDataType* pType = &pNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4674 4675
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pNode->node.aliasName);
4676 4677 4678
      pExp->pExpr->_optrRoot.pRootNode = pTargetNode->pExpr;
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4679 4680 4681
    }
  }

H
Haojun Liao 已提交
4682
  return pExprs;
H
Haojun Liao 已提交
4683 4684
}

4685
static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model) {
wafwerar's avatar
wafwerar 已提交
4686
  SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
4687
  setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
4688

4689
  pTaskInfo->cost.created = taosGetTimestampMs();
H
Haojun Liao 已提交
4690
  pTaskInfo->id.queryId = queryId;
dengyihao's avatar
dengyihao 已提交
4691
  pTaskInfo->execModel = model;
H
Haojun Liao 已提交
4692

wafwerar's avatar
wafwerar 已提交
4693
  char* p = taosMemoryCalloc(1, 128);
L
Liu Jicong 已提交
4694
  snprintf(p, 128, "TID:0x%" PRIx64 " QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
4695
  pTaskInfo->id.str = p;
H
Haojun Liao 已提交
4696

4697 4698
  return pTaskInfo;
}
H
Haojun Liao 已提交
4699

L
Liu Jicong 已提交
4700 4701
static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                                      STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId);
H
Haojun Liao 已提交
4702

L
Liu Jicong 已提交
4703 4704
static int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                                  uint64_t queryId, uint64_t taskId);
4705
static SArray* extractTableIdList(const STableGroupInfo* pTableGroupInfo);
H
Haojun Liao 已提交
4706
static SArray* extractColumnInfo(SNodeList* pNodeList);
X
Xiaoyu Wang 已提交
4707 4708
static SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols,
                                   int32_t type);
4709

4710
static SArray* createSortInfo(SNodeList* pNodeList);
4711
static SArray* extractPartitionColInfo(SNodeList* pNodeList);
4712
static int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode);
4713
static void    setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode);
4714 4715 4716

static SInterval extractIntervalInfo(const STableScanPhysiNode* pTableScanNode) {
  SInterval interval = {
L
Liu Jicong 已提交
4717 4718
      .interval = pTableScanNode->interval,
      .sliding = pTableScanNode->sliding,
4719
      .intervalUnit = pTableScanNode->intervalUnit,
L
Liu Jicong 已提交
4720 4721
      .slidingUnit = pTableScanNode->slidingUnit,
      .offset = pTableScanNode->offset,
4722 4723 4724 4725
  };

  return interval;
}
H
Haojun Liao 已提交
4726

H
Haojun Liao 已提交
4727
SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
4728
                                  uint64_t queryId, uint64_t taskId, STableGroupInfo* pTableGroupInfo) {
4729 4730
  int32_t type = nodeType(pPhyNode);

X
Xiaoyu Wang 已提交
4731
  if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
H
Haojun Liao 已提交
4732
    if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == type) {
dengyihao's avatar
dengyihao 已提交
4733 4734
      SScanPhysiNode*      pScanPhyNode = (SScanPhysiNode*)pPhyNode;
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4735

L
Liu Jicong 已提交
4736
      int32_t     numOfCols = 0;
H
Haojun Liao 已提交
4737
      tsdbReaderT pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableGroupInfo, (uint64_t)queryId, taskId);
4738
      if (pDataReader == NULL && terrno != 0) {
4739 4740
        return NULL;
      }
4741

4742 4743 4744 4745
      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;

      SArray* pColList = extractColMatchInfo(pScanPhyNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID);
      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4746

4747
      SQueryTableDataCond cond = {0};
X
Xiaoyu Wang 已提交
4748
      int32_t             code = initQueryTableDataCond(&cond, pTableScanNode);
4749 4750 4751
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
      }
4752

X
Xiaoyu Wang 已提交
4753 4754 4755 4756
      SInterval      interval = extractIntervalInfo(pTableScanNode);
      SOperatorInfo* pOperator = createTableScanOperatorInfo(
          pDataReader, &cond, numOfCols, pTableScanNode->dataRequired, pTableScanNode->scanSeq, pColList, pResBlock,
          pScanPhyNode->node.pConditions, &interval, pTableScanNode->ratio, pTaskInfo);
4757 4758 4759
      STableScanInfo* pScanInfo = pOperator->info;
      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
      return pOperator;
H
Haojun Liao 已提交
4760
    } else if (QUERY_NODE_PHYSICAL_PLAN_EXCHANGE == type) {
H
Haojun Liao 已提交
4761
      SExchangePhysiNode* pExchange = (SExchangePhysiNode*)pPhyNode;
4762
      SSDataBlock*        pResBlock = createResDataBlock(pExchange->node.pOutputDataBlockDesc);
4763
      return createExchangeOperatorInfo(pHandle->pMsgCb->clientRpc, pExchange->pSrcEndPoints, pResBlock, pTaskInfo);
H
Haojun Liao 已提交
4764
    } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN == type) {
H
Haojun Liao 已提交
4765
      SScanPhysiNode* pScanPhyNode = (SScanPhysiNode*)pPhyNode;  // simple child table.
4766

X
Xiaoyu Wang 已提交
4767 4768
      int32_t code = doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo,
                                        queryId, taskId);
H
Haojun Liao 已提交
4769
      SArray* tableIdList = extractTableIdList(pTableGroupInfo);
H
Haojun Liao 已提交
4770

4771
      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;
X
Xiaoyu Wang 已提交
4772
      SSDataBlock*        pResBlock = createResDataBlock(pDescNode);
4773

X
Xiaoyu Wang 已提交
4774 4775
      int32_t        numOfCols = 0;
      SArray*        pCols = extractColMatchInfo(pScanPhyNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID);
X
Xiaoyu Wang 已提交
4776 4777
      SOperatorInfo* pOperator = createStreamScanOperatorInfo(pHandle->reader, pResBlock, pCols, tableIdList, pTaskInfo,
                                                              pScanPhyNode->node.pConditions);
4778
      taosArrayDestroy(tableIdList);
H
Haojun Liao 已提交
4779
      return pOperator;
H
Haojun Liao 已提交
4780
    } else if (QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN == type) {
L
Liu Jicong 已提交
4781
      SSystemTableScanPhysiNode* pSysScanPhyNode = (SSystemTableScanPhysiNode*)pPhyNode;
L
Liu Jicong 已提交
4782
      SScanPhysiNode*            pScanNode = &pSysScanPhyNode->scan;
H
Haojun Liao 已提交
4783

4784 4785 4786
      SDataBlockDescNode* pDescNode = pScanNode->node.pOutputDataBlockDesc;

      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4787

4788
      int32_t numOfOutputCols = 0;
4789
      SArray* colList = extractColMatchInfo(pScanNode->pScanCols, pDescNode, &numOfOutputCols, COL_MATCH_FROM_COL_ID);
L
Liu Jicong 已提交
4790
      SOperatorInfo* pOperator = createSysTableScanOperatorInfo(
L
Liu Jicong 已提交
4791 4792
          pHandle, pResBlock, &pScanNode->tableName, pScanNode->node.pConditions, pSysScanPhyNode->mgmtEpSet, colList,
          pTaskInfo, pSysScanPhyNode->showRewrite, pSysScanPhyNode->accountId);
H
Haojun Liao 已提交
4793
      return pOperator;
4794
    } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
X
Xiaoyu Wang 已提交
4795
      STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
4796 4797 4798 4799

      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;

      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4800

X
Xiaoyu Wang 已提交
4801 4802
      int32_t code = doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo,
                                        queryId, taskId);
4803 4804 4805 4806
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
      }

X
Xiaoyu Wang 已提交
4807
      int32_t    num = 0;
4808 4809 4810
      SExprInfo* pExprInfo = createExprInfo(pScanPhyNode->pScanPseudoCols, NULL, &num);

      int32_t numOfOutputCols = 0;
X
Xiaoyu Wang 已提交
4811 4812
      SArray* colList =
          extractColMatchInfo(pScanPhyNode->pScanPseudoCols, pDescNode, &numOfOutputCols, COL_MATCH_FROM_COL_ID);
4813

X
Xiaoyu Wang 已提交
4814 4815
      SOperatorInfo* pOperator =
          createTagScanOperatorInfo(pHandle, pExprInfo, num, pResBlock, colList, pTableGroupInfo, pTaskInfo);
4816
      return pOperator;
H
Haojun Liao 已提交
4817 4818
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4819 4820 4821
    }
  }

4822 4823
  int32_t num = 0;
  size_t  size = LIST_LENGTH(pPhyNode->pChildren);
H
Haojun Liao 已提交
4824

4825
  SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
dengyihao's avatar
dengyihao 已提交
4826
  for (int32_t i = 0; i < size; ++i) {
4827 4828
    SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableGroupInfo);
4829 4830 4831
    if (ops[i] == NULL) {
      return NULL;
    }
4832
  }
H
Haojun Liao 已提交
4833

4834
  SOperatorInfo* pOptr = NULL;
H
Haojun Liao 已提交
4835
  if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
dengyihao's avatar
dengyihao 已提交
4836 4837
    SProjectPhysiNode* pProjPhyNode = (SProjectPhysiNode*)pPhyNode;
    SExprInfo*         pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &num);
H
Haojun Liao 已提交
4838

4839
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
dengyihao's avatar
dengyihao 已提交
4840 4841
    SLimit       limit = {.limit = pProjPhyNode->limit, .offset = pProjPhyNode->offset};
    SLimit       slimit = {.limit = pProjPhyNode->slimit, .offset = pProjPhyNode->soffset};
4842
    pOptr = createProjectOperatorInfo(ops[0], pExprInfo, num, pResBlock, &limit, &slimit, pTaskInfo);
H
Haojun Liao 已提交
4843 4844 4845
  } else if (QUERY_NODE_PHYSICAL_PLAN_AGG == type) {
    SAggPhysiNode* pAggNode = (SAggPhysiNode*)pPhyNode;
    SExprInfo*     pExprInfo = createExprInfo(pAggNode->pAggFuncs, pAggNode->pGroupKeys, &num);
4846
    SSDataBlock*   pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4847

dengyihao's avatar
dengyihao 已提交
4848
    int32_t    numOfScalarExpr = 0;
4849 4850 4851 4852 4853
    SExprInfo* pScalarExprInfo = NULL;
    if (pAggNode->pExprs != NULL) {
      pScalarExprInfo = createExprInfo(pAggNode->pExprs, NULL, &numOfScalarExpr);
    }

H
Haojun Liao 已提交
4854 4855
    if (pAggNode->pGroupKeys != NULL) {
      SArray* pColList = extractColumnInfo(pAggNode->pGroupKeys);
dengyihao's avatar
dengyihao 已提交
4856 4857
      pOptr = createGroupOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pAggNode->node.pConditions,
                                      pScalarExprInfo, numOfScalarExpr, pTaskInfo, NULL);
H
Haojun Liao 已提交
4858
    } else {
dengyihao's avatar
dengyihao 已提交
4859 4860
      pOptr = createAggregateOperatorInfo(ops[0], pExprInfo, num, pResBlock, pScalarExprInfo, numOfScalarExpr,
                                          pTaskInfo, pTableGroupInfo);
H
Haojun Liao 已提交
4861
    }
X
Xiaoyu Wang 已提交
4862
  } else if (QUERY_NODE_PHYSICAL_PLAN_INTERVAL == type || QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL == type) {
H
Haojun Liao 已提交
4863
    SIntervalPhysiNode* pIntervalPhyNode = (SIntervalPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4864

H
Haojun Liao 已提交
4865
    SExprInfo*   pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
4866
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4867

dengyihao's avatar
dengyihao 已提交
4868 4869 4870 4871 4872 4873
    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
4874

dengyihao's avatar
dengyihao 已提交
4875 4876
    STimeWindowAggSupp as = {.waterMark = pIntervalPhyNode->window.watermark,
                             .calTrigger = pIntervalPhyNode->window.triggerType};
4877

4878 4879 4880 4881
    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
4882
  } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {
H
Haojun Liao 已提交
4883
    SSortPhysiNode* pSortPhyNode = (SSortPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4884

4885 4886 4887
    SDataBlockDescNode* pDescNode = pPhyNode->pOutputDataBlockDesc;

    SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4888
    SArray*      info = createSortInfo(pSortPhyNode->pSortKeys);
4889 4890

    int32_t    numOfCols = 0;
4891 4892 4893
    SExprInfo* pExprInfo = createExprInfo(pSortPhyNode->pExprs, NULL, &numOfCols);

    int32_t numOfOutputCols = 0;
4894
    SArray* pColList = extractColMatchInfo(pSortPhyNode->pTargets, pDescNode, &numOfOutputCols, COL_MATCH_FROM_SLOT_ID);
4895

4896
    pOptr = createSortOperatorInfo(ops[0], pResBlock, info, pExprInfo, numOfCols, pColList, pTaskInfo);
H
Haojun Liao 已提交
4897
  } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
H
Haojun Liao 已提交
4898 4899
    SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;

X
Xiaoyu Wang 已提交
4900 4901
    STimeWindowAggSupp as = {.waterMark = pSessionNode->window.watermark,
                             .calTrigger = pSessionNode->window.triggerType};
4902

H
Haojun Liao 已提交
4903
    SExprInfo*   pExprInfo = createExprInfo(pSessionNode->window.pFuncs, NULL, &num);
4904
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4905 4906
    int32_t      tsSlotId = ((SColumnNode*)pSessionNode->window.pTspk)->slotId;

X
Xiaoyu Wang 已提交
4907 4908
    pOptr =
        createSessionAggOperatorInfo(ops[0], pExprInfo, num, pResBlock, pSessionNode->gap, tsSlotId, &as, pTaskInfo);
H
Haojun Liao 已提交
4909
  } else if (QUERY_NODE_PHYSICAL_PLAN_PARTITION == type) {
dengyihao's avatar
dengyihao 已提交
4910 4911 4912
    SPartitionPhysiNode* pPartNode = (SPartitionPhysiNode*)pPhyNode;
    SArray*              pColList = extractPartitionColInfo(pPartNode->pPartitionKeys);
    SSDataBlock*         pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4913 4914

    SExprInfo* pExprInfo = createExprInfo(pPartNode->pTargets, NULL, &num);
4915
    pOptr = createPartitionOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pTaskInfo, NULL);
4916
  } else if (QUERY_NODE_PHYSICAL_PLAN_STATE_WINDOW == type) {
dengyihao's avatar
dengyihao 已提交
4917
    SStateWinodwPhysiNode* pStateNode = (SStateWinodwPhysiNode*)pPhyNode;
4918

4919 4920
    STimeWindowAggSupp as = {.waterMark = pStateNode->window.watermark, .calTrigger = pStateNode->window.triggerType};

dengyihao's avatar
dengyihao 已提交
4921
    SExprInfo*   pExprInfo = createExprInfo(pStateNode->window.pFuncs, NULL, &num);
4922
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4923 4924 4925
    int32_t      tsSlotId = ((SColumnNode*)pStateNode->window.pTspk)->slotId;

    pOptr = createStatewindowOperatorInfo(ops[0], pExprInfo, num, pResBlock, &as, tsSlotId, pTaskInfo);
4926
  } else if (QUERY_NODE_PHYSICAL_PLAN_JOIN == type) {
dengyihao's avatar
dengyihao 已提交
4927 4928
    SJoinPhysiNode* pJoinNode = (SJoinPhysiNode*)pPhyNode;
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4929 4930

    SExprInfo* pExprInfo = createExprInfo(pJoinNode->pTargets, NULL, &num);
H
Haojun Liao 已提交
4931
    pOptr = createMergeJoinOperatorInfo(ops, size, pExprInfo, num, pResBlock, pJoinNode->pOnConditions, pTaskInfo);
4932 4933
  } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
    SFillPhysiNode* pFillNode = (SFillPhysiNode*)pPhyNode;
X
Xiaoyu Wang 已提交
4934 4935
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
    SExprInfo*      pExprInfo = createExprInfo(pFillNode->pTargets, NULL, &num);
4936

4937
    SInterval* pInterval = &((SIntervalAggOperatorInfo*)ops[0]->info)->interval;
X
Xiaoyu Wang 已提交
4938 4939
    pOptr = createFillOperatorInfo(ops[0], pExprInfo, num, pInterval, &pFillNode->timeRange, pResBlock, pFillNode->mode,
                                   (SNodeListNode*)pFillNode->pValues, false, pTaskInfo);
H
Haojun Liao 已提交
4940 4941
  } else {
    ASSERT(0);
H
Haojun Liao 已提交
4942
  }
4943 4944 4945

  taosMemoryFree(ops);
  return pOptr;
4946
}
H
Haojun Liao 已提交
4947

L
Liu Jicong 已提交
4948
static int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode) {
4949 4950 4951 4952 4953 4954
  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
4955
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
4956 4957 4958 4959 4960 4961
    return terrno;
  }

  pCond->twindow = pTableScanNode->scanRange;

#if 1
L
Liu Jicong 已提交
4962
  // todo work around a problem, remove it later
4963 4964
  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
4965
    TSWAP(pCond->twindow.skey, pCond->twindow.ekey);
H
Haojun Liao 已提交
4966
  }
4967
#endif
H
Haojun Liao 已提交
4968

4969 4970
  pCond->type = BLOCK_LOAD_OFFSET_SEQ_ORDER;
  //  pCond->type = pTableScanNode->scanFlag;
H
Haojun Liao 已提交
4971

H
Haojun Liao 已提交
4972
  int32_t j = 0;
4973
  for (int32_t i = 0; i < pCond->numOfCols; ++i) {
H
Haojun Liao 已提交
4974 4975
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pTableScanNode->scan.pScanCols, i);
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;
H
Haojun Liao 已提交
4976 4977 4978
    if (pColNode->colType == COLUMN_TYPE_TAG) {
      continue;
    }
H
Haojun Liao 已提交
4979

4980 4981 4982
    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
4983
    j += 1;
H
Haojun Liao 已提交
4984 4985
  }

4986 4987
  pCond->numOfCols = j;
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
4988
}
H
Haojun Liao 已提交
4989

H
Haojun Liao 已提交
4990
SArray* extractColumnInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
4991
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
4992 4993 4994 4995 4996 4997
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
4998 4999
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
H
Haojun Liao 已提交
5000

5001 5002 5003 5004 5005
    if (nodeType(pNode->pExpr) == QUERY_NODE_COLUMN) {
      SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

      // todo extract method
      SColumn c = {0};
L
Liu Jicong 已提交
5006 5007 5008 5009 5010
      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;
5011 5012 5013 5014
      c.precision = pColNode->node.resType.precision;

      taosArrayPush(pList, &c);
    } else if (nodeType(pNode->pExpr) == QUERY_NODE_VALUE) {
L
Liu Jicong 已提交
5015 5016
      SValueNode* pValNode = (SValueNode*)pNode->pExpr;
      SColumn     c = {0};
5017
      c.slotId = pNode->slotId;
L
Liu Jicong 已提交
5018 5019 5020 5021
      c.colId = pNode->slotId;
      c.type = pValNode->node.type;
      c.bytes = pValNode->node.resType.bytes;
      c.scale = pValNode->node.resType.scale;
5022 5023 5024 5025
      c.precision = pValNode->node.resType.precision;

      taosArrayPush(pList, &c);
    }
H
Haojun Liao 已提交
5026 5027 5028 5029 5030
  }

  return pList;
}

5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044
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
5045 5046 5047
    c.colId = pColNode->colId;
    c.type = pColNode->node.resType.type;
    c.bytes = pColNode->node.resType.bytes;
5048 5049 5050 5051 5052 5053 5054 5055 5056
    c.precision = pColNode->node.resType.precision;
    c.scale = pColNode->node.resType.scale;

    taosArrayPush(pList, &c);
  }

  return pList;
}

5057
SArray* createSortInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
5058
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5059 5060 5061 5062 5063 5064
  SArray* pList = taosArrayInit(numOfCols, sizeof(SBlockOrderInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return pList;
  }

L
Liu Jicong 已提交
5065
  for (int32_t i = 0; i < numOfCols; ++i) {
5066
    SOrderByExprNode* pSortKey = (SOrderByExprNode*)nodesListGetNode(pNodeList, i);
L
Liu Jicong 已提交
5067 5068
    SBlockOrderInfo   bi = {0};
    bi.order = (pSortKey->order == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
H
Haojun Liao 已提交
5069 5070 5071
    bi.nullFirst = (pSortKey->nullOrder == NULL_ORDER_FIRST);

    SColumnNode* pColNode = (SColumnNode*)pSortKey->pExpr;
5072
    bi.slotId = pColNode->slotId;
H
Haojun Liao 已提交
5073 5074 5075 5076 5077 5078
    taosArrayPush(pList, &bi);
  }

  return pList;
}

X
Xiaoyu Wang 已提交
5079 5080
SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols,
                            int32_t type) {
L
Liu Jicong 已提交
5081
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5082 5083 5084 5085 5086 5087
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColMatchInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
5088 5089 5090
  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
5091 5092

    SColMatchInfo c = {0};
X
Xiaoyu Wang 已提交
5093 5094 5095 5096
    c.output = true;
    c.colId = pColNode->colId;
    c.srcSlotId = pColNode->slotId;
    c.matchType = type;
H
Haojun Liao 已提交
5097 5098 5099 5100
    c.targetSlotId = pNode->slotId;
    taosArrayPush(pList, &c);
  }

H
Haojun Liao 已提交
5101 5102
  *numOfOutputCols = 0;
  int32_t num = LIST_LENGTH(pOutputNodeList->pSlots);
L
Liu Jicong 已提交
5103 5104
  for (int32_t i = 0; i < num; ++i) {
    SSlotDescNode* pNode = (SSlotDescNode*)nodesListGetNode(pOutputNodeList->pSlots, i);
5105

5106
    // todo: add reserve flag check
5107 5108
    // it is a column reserved for the arithmetic expression calculation
    if (pNode->slotId >= numOfCols) {
5109 5110 5111 5112
      (*numOfOutputCols) += 1;
      continue;
    }

H
Haojun Liao 已提交
5113
    SColMatchInfo* info = taosArrayGet(pList, pNode->slotId);
H
Haojun Liao 已提交
5114 5115 5116 5117 5118
    if (pNode->output) {
      (*numOfOutputCols) += 1;
    } else {
      info->output = false;
    }
H
Haojun Liao 已提交
5119 5120
  }

H
Haojun Liao 已提交
5121 5122 5123
  return pList;
}

dengyihao's avatar
dengyihao 已提交
5124 5125
int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                           uint64_t queryId, uint64_t taskId) {
5126
  int32_t code = 0;
H
Haojun Liao 已提交
5127
  if (tableType == TSDB_SUPER_TABLE) {
H
Haojun Liao 已提交
5128
    code = tsdbQuerySTableByTagCond(metaHandle, tableUid, 0, NULL, 0, 0, NULL, pGroupInfo, NULL, 0, queryId, taskId);
H
Haojun Liao 已提交
5129
  } else {  // Create one table group.
H
Haojun Liao 已提交
5130
    code = tsdbGetOneTableGroup(metaHandle, tableUid, 0, pGroupInfo);
5131 5132 5133 5134
  }

  return code;
}
H
Haojun Liao 已提交
5135

5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153
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
5154 5155
tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                               STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId) {
5156
  uint64_t uid = pTableScanNode->scan.uid;
L
Liu Jicong 已提交
5157 5158
  int32_t  code =
      doCreateTableGroup(pHandle->meta, pTableScanNode->scan.tableType, uid, pTableGroupInfo, queryId, taskId);
5159 5160
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
H
Haojun Liao 已提交
5161
  }
H
Haojun Liao 已提交
5162

H
Haojun Liao 已提交
5163
  if (pTableGroupInfo->numOfTables == 0) {
H
Haojun Liao 已提交
5164
    code = 0;
L
Liu Jicong 已提交
5165
    qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
5166 5167
    goto _error;
  }
H
Haojun Liao 已提交
5168

5169 5170 5171 5172
  SQueryTableDataCond cond = {0};
  code = initQueryTableDataCond(&cond, pTableScanNode);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
X
Xiaoyu Wang 已提交
5173
  }
5174

5175
  return tsdbQueryTables(pHandle->vnode, &cond, pTableGroupInfo, queryId, taskId);
H
Haojun Liao 已提交
5176

L
Liu Jicong 已提交
5177
_error:
H
Haojun Liao 已提交
5178 5179 5180 5181
  terrno = code;
  return NULL;
}

dengyihao's avatar
dengyihao 已提交
5182 5183
int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
                               EOPTR_EXEC_MODEL model) {
H
Haojun Liao 已提交
5184 5185
  uint64_t queryId = pPlan->id.queryId;

H
Haojun Liao 已提交
5186
  int32_t code = TSDB_CODE_SUCCESS;
5187
  *pTaskInfo = createExecTaskInfo(queryId, taskId, model);
H
Haojun Liao 已提交
5188 5189 5190 5191
  if (*pTaskInfo == NULL) {
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    goto _complete;
  }
H
Haojun Liao 已提交
5192

L
Liu Jicong 已提交
5193 5194
  (*pTaskInfo)->pRoot =
      createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId, &(*pTaskInfo)->tableqinfoGroupInfo);
D
dapan1121 已提交
5195
  if (NULL == (*pTaskInfo)->pRoot) {
L
Liu Jicong 已提交
5196
    code = terrno;
D
dapan1121 已提交
5197 5198
    goto _complete;
  }
H
Haojun Liao 已提交
5199

5200
  if ((*pTaskInfo)->pRoot == NULL) {
H
Haojun Liao 已提交
5201
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
H
Haojun Liao 已提交
5202
    goto _complete;
5203 5204
  }

H
Haojun Liao 已提交
5205 5206
  return code;

H
Haojun Liao 已提交
5207
_complete:
wafwerar's avatar
wafwerar 已提交
5208
  taosMemoryFreeClear(*pTaskInfo);
H
Haojun Liao 已提交
5209 5210 5211

  terrno = code;
  return code;
H
Haojun Liao 已提交
5212 5213
}

H
Haojun Liao 已提交
5214
void setResultBufSize(STaskAttr* pQueryAttr, SResultInfo* pResultInfo) {
5215 5216 5217 5218 5219 5220 5221 5222
  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
5223 5224 5225 5226 5227 5228 5229 5230 5231 5232
  //  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;
  //  }
5233 5234

  pResultInfo->threshold = (int32_t)(pResultInfo->capacity * THRESHOLD_RATIO);
H
Haojun Liao 已提交
5235
  pResultInfo->totalRows = 0;
5236 5237
}

L
Liu Jicong 已提交
5238
// TODO refactor
5239
void freeColumnFilterInfo(SColumnFilterInfo* pFilter, int32_t numOfFilters) {
L
Liu Jicong 已提交
5240 5241 5242
  if (pFilter == NULL || numOfFilters == 0) {
    return;
  }
5243

L
Liu Jicong 已提交
5244 5245 5246
  for (int32_t i = 0; i < numOfFilters; i++) {
    if (pFilter[i].filterstr && pFilter[i].pz) {
      taosMemoryFree((void*)(pFilter[i].pz));
5247
    }
L
Liu Jicong 已提交
5248
  }
5249

L
Liu Jicong 已提交
5250
  taosMemoryFree(pFilter);
5251 5252 5253 5254
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo) {
  if (pTableqinfoGroupInfo->pGroupList != NULL) {
L
Liu Jicong 已提交
5255
    int32_t numOfGroups = (int32_t)taosArrayGetSize(pTableqinfoGroupInfo->pGroupList);
5256
    for (int32_t i = 0; i < numOfGroups; ++i) {
L
Liu Jicong 已提交
5257
      SArray* p = taosArrayGetP(pTableqinfoGroupInfo->pGroupList, i);
5258 5259

      size_t num = taosArrayGetSize(p);
L
Liu Jicong 已提交
5260
      for (int32_t j = 0; j < num; ++j) {
5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276
        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
5277
void doDestroyTask(SExecTaskInfo* pTaskInfo) {
H
Haojun Liao 已提交
5278 5279
  qDebug("%s execTask is freed", GET_TASKID(pTaskInfo));

H
Haojun Liao 已提交
5280
  doDestroyTableQueryInfo(&pTaskInfo->tableqinfoGroupInfo);
H
Haojun Liao 已提交
5281
  destroyOperatorInfo(pTaskInfo->pRoot);
L
Liu Jicong 已提交
5282 5283
  //  taosArrayDestroy(pTaskInfo->summary.queryProfEvents);
  //  taosHashCleanup(pTaskInfo->summary.operatorProfResults);
5284

wafwerar's avatar
wafwerar 已提交
5285 5286 5287
  taosMemoryFreeClear(pTaskInfo->sql);
  taosMemoryFreeClear(pTaskInfo->id.str);
  taosMemoryFreeClear(pTaskInfo);
5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299
}

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
5300
      int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312
      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
5313 5314
  //  size_t s3 = sizeof(STableCheckInfo);  buffer consumption in tsdb
  return (int64_t)(s1 * 1.5 * numOfTables);
5315 5316 5317 5318 5319 5320 5321
}

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
5322
    while (1) {
5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348
      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
5349

dengyihao's avatar
dengyihao 已提交
5350 5351
int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SExplainExecInfo** pRes, int32_t* capacity,
                                   int32_t* resNum) {
D
dapan1121 已提交
5352 5353
  if (*resNum >= *capacity) {
    *capacity += 10;
dengyihao's avatar
dengyihao 已提交
5354

D
dapan1121 已提交
5355 5356
    *pRes = taosMemoryRealloc(*pRes, (*capacity) * sizeof(SExplainExecInfo));
    if (NULL == *pRes) {
D
dapan1121 已提交
5357
      qError("malloc %d failed", (*capacity) * (int32_t)sizeof(SExplainExecInfo));
D
dapan1121 已提交
5358 5359 5360 5361
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

D
dapan1121 已提交
5362 5363 5364 5365
  (*pRes)[*resNum].numOfRows = operatorInfo->resultInfo.totalRows;
  (*pRes)[*resNum].startupCost = operatorInfo->cost.openCost;
  (*pRes)[*resNum].totalCost = operatorInfo->cost.totalCost;

5366 5367
  if (operatorInfo->fpSet.getExplainFn) {
    int32_t code = (*operatorInfo->fpSet.getExplainFn)(operatorInfo, &(*pRes)->verboseInfo);
D
dapan1121 已提交
5368 5369 5370 5371 5372
    if (code) {
      qError("operator getExplainFn failed, error:%s", tstrerror(code));
      return code;
    }
  }
dengyihao's avatar
dengyihao 已提交
5373

D
dapan1121 已提交
5374
  ++(*resNum);
dengyihao's avatar
dengyihao 已提交
5375

D
dapan1121 已提交
5376
  int32_t code = 0;
D
dapan1121 已提交
5377 5378
  for (int32_t i = 0; i < operatorInfo->numOfDownstream; ++i) {
    code = getOperatorExplainExecInfo(operatorInfo->pDownstream[i], pRes, capacity, resNum);
D
dapan1121 已提交
5379 5380 5381 5382 5383 5384 5385
    if (code) {
      taosMemoryFreeClear(*pRes);
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
D
dapan1121 已提交
5386 5387
}

5388
static SSDataBlock* doMergeJoin(struct SOperatorInfo* pOperator) {
5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400
  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);
5401
      pJoinInfo->pLeft = ds1->fpSet.getNextFn(ds1);
5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413
      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);
5414
      pJoinInfo->pRight = ds2->fpSet.getNextFn(ds2);
5415 5416 5417 5418 5419 5420 5421 5422 5423 5424
      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
5425
    char*            pLeftVal = colDataGetData(pLeftCol, pJoinInfo->leftPos);
5426 5427

    SColumnInfoData* pRightCol = taosArrayGet(pJoinInfo->pRight->pDataBlock, pJoinInfo->rightCol.slotId);
dengyihao's avatar
dengyihao 已提交
5428
    char*            pRightVal = colDataGetData(pRightCol, pJoinInfo->rightPos);
5429 5430

    // only the timestamp match support for ordinary table
dengyihao's avatar
dengyihao 已提交
5431 5432
    ASSERT(pLeftCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
    if (*(int64_t*)pLeftVal == *(int64_t*)pRightVal) {
5433
      for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
dengyihao's avatar
dengyihao 已提交
5434
        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, i);
5435

dengyihao's avatar
dengyihao 已提交
5436
        SExprInfo* pExprInfo = &pOperator->pExpr[i];
5437

dengyihao's avatar
dengyihao 已提交
5438 5439
        int32_t blockId = pExprInfo->base.pParam[0].pCol->dataBlockId;
        int32_t slotId = pExprInfo->base.pParam[0].pCol->slotId;
5440

dengyihao's avatar
dengyihao 已提交
5441 5442 5443 5444 5445 5446
        SColumnInfoData* pSrc = NULL;
        if (pJoinInfo->pLeft->info.blockId == blockId) {
          pSrc = taosArrayGet(pJoinInfo->pLeft->pDataBlock, slotId);
        } else {
          pSrc = taosArrayGet(pJoinInfo->pRight->pDataBlock, slotId);
        }
5447

dengyihao's avatar
dengyihao 已提交
5448 5449 5450 5451 5452
        if (colDataIsNull_s(pSrc, pJoinInfo->leftPos)) {
          colDataAppendNULL(pDst, nrows);
        } else {
          char* p = colDataGetData(pSrc, pJoinInfo->leftPos);
          colDataAppend(pDst, nrows, p, false);
5453
        }
dengyihao's avatar
dengyihao 已提交
5454
      }
5455

dengyihao's avatar
dengyihao 已提交
5456 5457
      pJoinInfo->leftPos += 1;
      pJoinInfo->rightPos += 1;
5458

dengyihao's avatar
dengyihao 已提交
5459 5460 5461
      nrows += 1;
    } else if (*(int64_t*)pLeftVal < *(int64_t*)pRightVal) {
      pJoinInfo->leftPos += 1;
D
dapan1121 已提交
5462

dengyihao's avatar
dengyihao 已提交
5463 5464
      if (pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
        continue;
5465
      }
dengyihao's avatar
dengyihao 已提交
5466 5467 5468 5469
    } else if (*(int64_t*)pLeftVal > *(int64_t*)pRightVal) {
      pJoinInfo->rightPos += 1;
      if (pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
        continue;
5470
      }
dengyihao's avatar
dengyihao 已提交
5471
    }
5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482

    // 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;
}

H
Haojun Liao 已提交
5483
SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
X
Xiaoyu Wang 已提交
5484 5485
                                           int32_t numOfCols, SSDataBlock* pResBlock, SNode* pOnCondition,
                                           SExecTaskInfo* pTaskInfo) {
5486
  SJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SJoinOperatorInfo));
dengyihao's avatar
dengyihao 已提交
5487
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
5488 5489 5490 5491
  if (pOperator == NULL || pInfo == NULL) {
    goto _error;
  }

5492
  initResultSizeInfo(pOperator, 4096);
5493

X
Xiaoyu Wang 已提交
5494 5495
  pInfo->pRes = pResBlock;
  pOperator->name = "MergeJoinOperator";
5496
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_JOIN;
X
Xiaoyu Wang 已提交
5497 5498 5499
  pOperator->blocking = false;
  pOperator->status = OP_NOT_OPENED;
  pOperator->pExpr = pExprInfo;
5500
  pOperator->numOfExprs = numOfCols;
X
Xiaoyu Wang 已提交
5501
  pOperator->info = pInfo;
dengyihao's avatar
dengyihao 已提交
5502
  pOperator->pTaskInfo = pTaskInfo;
5503

5504 5505 5506 5507
  SOperatorNode* pNode = (SOperatorNode*)pOnCondition;
  setJoinColumnInfo(&pInfo->leftCol, (SColumnNode*)pNode->pLeft);
  setJoinColumnInfo(&pInfo->rightCol, (SColumnNode*)pNode->pRight);

L
Liu Jicong 已提交
5508
  pOperator->fpSet =
H
Haojun Liao 已提交
5509
      createOperatorFpSet(operatorDummyOpenFn, doMergeJoin, NULL, NULL, destroyMergeJoinOperator, NULL, NULL, NULL);
5510
  int32_t code = appendDownstream(pOperator, pDownstream, numOfDownstream);
5511 5512 5513 5514
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }

5515 5516
  return pOperator;

dengyihao's avatar
dengyihao 已提交
5517
_error:
5518 5519 5520 5521
  taosMemoryFree(pInfo);
  taosMemoryFree(pOperator);
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
dengyihao's avatar
dengyihao 已提交
5522
}
5523 5524

void setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode) {
X
Xiaoyu Wang 已提交
5525 5526 5527
  pColumn->slotId = pColumnNode->slotId;
  pColumn->type = pColumnNode->node.resType.type;
  pColumn->bytes = pColumnNode->node.resType.bytes;
5528
  pColumn->precision = pColumnNode->node.resType.precision;
X
Xiaoyu Wang 已提交
5529
  pColumn->scale = pColumnNode->node.resType.scale;
5530
}