executorimpl.c 198.3 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

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

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

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

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

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

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

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

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

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

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

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

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

dengyihao's avatar
dengyihao 已提交
582 583
void doApplyFunctions(SqlFunctionCtx* pCtx, STimeWindow* pWin, SColumnInfoData* pTimeWindowData, int32_t offset,
                      int32_t forwardStep, TSKEY* tsCol, int32_t numOfTotal, int32_t numOfOutput, int32_t order) {
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);
618 619 620 621
      pEntryInfo->numOfRes = 1;
      continue;
    }

622
    if (functionNeedToExecute(&pCtx[k]) && pCtx[k].fpSet.process != NULL) {
623
      pCtx[k].fpSet.process(&pCtx[k]);
624 625 626
    }

    // restore it
627 628 629
    pCtx[k].input.colDataAggIsSet = hasAgg;
    pCtx[k].input.startRowIndex = startOffset;
    pCtx[k].input.numOfRows = numOfRows;
630 631 632
  }
}

L
Liu Jicong 已提交
633
static FORCE_INLINE TSKEY reviseWindowEkey(STaskAttr* pQueryAttr, STimeWindow* pWindow) {
dengyihao's avatar
dengyihao 已提交
634
  TSKEY   ekey = -1;
635 636
  int32_t order = TSDB_ORDER_ASC;
  if (order == TSDB_ORDER_ASC) {
637 638 639 640 641 642 643 644 645 646 647 648 649 650
    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
651 652
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                                   bool createDummyCol);
653

dengyihao's avatar
dengyihao 已提交
654 655
static void doSetInputDataBlockInfo(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock,
                                    int32_t order) {
656
  for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
657
    pCtx[i].order = order;
L
Liu Jicong 已提交
658
    pCtx[i].size = pBlock->info.rows;
659
    setBlockStatisInfo(&pCtx[i], &pOperator->pExpr[i], pBlock);
660 661 662
  }
}

dengyihao's avatar
dengyihao 已提交
663 664
void setInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                       bool createDummyCol) {
665
  if (pBlock->pBlockAgg != NULL) {
H
Haojun Liao 已提交
666
    doSetInputDataBlockInfo(pOperator, pCtx, pBlock, order);
667
  } else {
668
    doSetInputDataBlock(pOperator, pCtx, pBlock, order, createDummyCol);
H
Haojun Liao 已提交
669
  }
670 671
}

L
Liu Jicong 已提交
672 673
static int32_t doCreateConstantValColumnInfo(SInputColumnInfoData* pInput, SFunctParam* pFuncParam, int32_t paramIndex,
                                             int32_t numOfRows) {
674 675 676 677 678 679 680 681
  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)
682 683
    pColInfo->info.type = pFuncParam->param.nType;
    pColInfo->info.bytes = pFuncParam->param.nLen;
684 685

    pInput->pData[paramIndex] = pColInfo;
686 687
  } else {
    pColInfo = pInput->pData[paramIndex];
688 689
  }

690
  colInfoDataEnsureCapacity(pColInfo, 0, numOfRows);
691

692
  int8_t type = pFuncParam->param.nType;
693 694
  if (type == TSDB_DATA_TYPE_BIGINT || type == TSDB_DATA_TYPE_UBIGINT) {
    int64_t v = pFuncParam->param.i;
dengyihao's avatar
dengyihao 已提交
695
    for (int32_t i = 0; i < numOfRows; ++i) {
696 697 698 699
      colDataAppendInt64(pColInfo, i, &v);
    }
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
dengyihao's avatar
dengyihao 已提交
700
    for (int32_t i = 0; i < numOfRows; ++i) {
701 702
      colDataAppendDouble(pColInfo, i, &v);
    }
703
  } else if (type == TSDB_DATA_TYPE_VARCHAR) {
L
Liu Jicong 已提交
704
    char* tmp = taosMemoryMalloc(pFuncParam->param.nLen + VARSTR_HEADER_SIZE);
705
    STR_WITH_SIZE_TO_VARSTR(tmp, pFuncParam->param.pz, pFuncParam->param.nLen);
L
Liu Jicong 已提交
706
    for (int32_t i = 0; i < numOfRows; ++i) {
707 708
      colDataAppend(pColInfo, i, tmp, false);
    }
709 710 711 712 713
  }

  return TSDB_CODE_SUCCESS;
}

dengyihao's avatar
dengyihao 已提交
714 715
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                                   bool createDummyCol) {
716 717
  int32_t code = TSDB_CODE_SUCCESS;

718
  for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
L
Liu Jicong 已提交
719 720 721
    pCtx[i].order = order;
    pCtx[i].size = pBlock->info.rows;
    pCtx[i].pSrcBlock = pBlock;
H
Haojun Liao 已提交
722 723
    pCtx[i].currentStage = MAIN_SCAN;

724
    SInputColumnInfoData* pInput = &pCtx[i].input;
725
    pInput->uid = pBlock->info.uid;
C
Cary Xu 已提交
726
    pInput->colDataAggIsSet = false;
727

728 729
    SExprInfo* pOneExpr = &pOperator->pExpr[i];
    for (int32_t j = 0; j < pOneExpr->base.numOfParams; ++j) {
dengyihao's avatar
dengyihao 已提交
730
      SFunctParam* pFuncParam = &pOneExpr->base.pParam[j];
G
Ganlin Zhao 已提交
731 732
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
dengyihao's avatar
dengyihao 已提交
733
        pInput->pData[j] = taosArrayGet(pBlock->pDataBlock, slotId);
734 735 736
        pInput->totalRows = pBlock->info.rows;
        pInput->numOfRows = pBlock->info.rows;
        pInput->startRowIndex = 0;
737

738 739 740 741
        // the last parameter is the timestamp column
        if (fmIsTimelineFunc(pCtx[i].functionId) && (j == pOneExpr->base.numOfParams - 1)) {
          pInput->pPTS = pInput->pData[j];
        }
742 743
        ASSERT(pInput->pData[j] != NULL);
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
744 745 746
        // 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) {
747 748 749 750
          pInput->totalRows = pBlock->info.rows;
          pInput->numOfRows = pBlock->info.rows;
          pInput->startRowIndex = 0;

751
          code = doCreateConstantValColumnInfo(pInput, pFuncParam, j, pBlock->info.rows);
752 753 754
          if (code != TSDB_CODE_SUCCESS) {
            return code;
          }
755
        }
G
Ganlin Zhao 已提交
756 757
      }
    }
H
Haojun Liao 已提交
758

H
Haojun Liao 已提交
759
    //    setBlockStatisInfo(&pCtx[i], pBlock, pOperator->pExpr[i].base.pColumns);
H
Haojun Liao 已提交
760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797
    //      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);
    //        }
    //      }
  }
798 799

  return code;
H
Haojun Liao 已提交
800 801 802
}

static void doAggregateImpl(SOperatorInfo* pOperator, TSKEY startTs, SqlFunctionCtx* pCtx) {
803
  for (int32_t k = 0; k < pOperator->numOfExprs; ++k) {
H
Haojun Liao 已提交
804
    if (functionNeedToExecute(&pCtx[k])) {
805 806 807 808
      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) {
809
        pCtx[k].fpSet.process(&pCtx[k]);
810
      }
811 812 813 814
    }
  }
}

H
Haojun Liao 已提交
815
static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, SArray* pPseudoList) {
dengyihao's avatar
dengyihao 已提交
816
  size_t num = (pPseudoList != NULL) ? taosArrayGetSize(pPseudoList) : 0;
H
Haojun Liao 已提交
817 818 819 820 821
  for (int32_t i = 0; i < num; ++i) {
    pCtx[i].pOutput = taosArrayGet(pResult->pDataBlock, i);
  }
}

822
int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
X
Xiaoyu Wang 已提交
823
                              int32_t numOfOutput, SArray* pPseudoList) {
H
Haojun Liao 已提交
824
  setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
H
Haojun Liao 已提交
825
  pResult->info.groupId = pSrcBlock->info.groupId;
H
Haojun Liao 已提交
826

dengyihao's avatar
dengyihao 已提交
827 828
  // if the source equals to the destination, it is to create a new column as the result of scalar function or some
  // operators.
829 830
  bool createNewColModel = (pResult == pSrcBlock);

831 832
  int32_t numOfRows = 0;

833
  for (int32_t k = 0; k < numOfOutput; ++k) {
dengyihao's avatar
dengyihao 已提交
834
    int32_t         outputSlotId = pExpr[k].base.resSchema.slotId;
835 836
    SqlFunctionCtx* pfCtx = &pCtx[k];

L
Liu Jicong 已提交
837
    if (pExpr[k].pExpr->nodeType == QUERY_NODE_COLUMN) {  // it is a project query
838
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
839
      if (pResult->info.rows > 0 && !createNewColModel) {
X
Xiaoyu Wang 已提交
840 841
        colDataMergeCol(pColInfoData, pResult->info.rows, &pResult->info.capacity, pfCtx->input.pData[0],
                        pfCtx->input.numOfRows);
842 843 844
      } else {
        colDataAssign(pColInfoData, pfCtx->input.pData[0], pfCtx->input.numOfRows);
      }
845

846
      numOfRows = pfCtx->input.numOfRows;
847
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE) {
848
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
849

dengyihao's avatar
dengyihao 已提交
850
      int32_t offset = createNewColModel ? 0 : pResult->info.rows;
851
      for (int32_t i = 0; i < pSrcBlock->info.rows; ++i) {
dengyihao's avatar
dengyihao 已提交
852 853 854
        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);
855
      }
856 857

      numOfRows = pSrcBlock->info.rows;
H
Haojun Liao 已提交
858
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_OPERATOR) {
859 860 861
      SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
      taosArrayPush(pBlockList, &pSrcBlock);

862
      SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
863
      SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
864

865
      SScalarParam dest = {.columnData = &idata};
866 867
      scalarCalculate(pExpr[k].pExpr->_optrRoot.pRootNode, pBlockList, &dest);

dengyihao's avatar
dengyihao 已提交
868
      int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
869
      colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
870 871

      numOfRows = dest.numOfRows;
872 873
      taosArrayDestroy(pBlockList);
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_FUNCTION) {
874
      ASSERT(!fmIsAggFunc(pfCtx->functionId));
875

876
      if (fmIsPseudoColumnFunc(pfCtx->functionId)) {
H
Haojun Liao 已提交
877
        // do nothing
878
      } else if (fmIsNonstandardSQLFunc(pfCtx->functionId)) {
dengyihao's avatar
dengyihao 已提交
879
        SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[k]);
880
        pfCtx->fpSet.init(&pCtx[k], pResInfo);
881

882
        pfCtx->pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
dengyihao's avatar
dengyihao 已提交
883
        pfCtx->offset = createNewColModel ? 0 : pResult->info.rows;  // set the start offset
H
Haojun Liao 已提交
884

885
        // set the timestamp(_rowts) output buffer
886 887
        if (taosArrayGetSize(pPseudoList) > 0) {
          int32_t* outputColIndex = taosArrayGet(pPseudoList, 0);
888
          pfCtx->pTsOutput = (SColumnInfoData*)pCtx[*outputColIndex].pOutput;
889
        }
H
Haojun Liao 已提交
890

891
        numOfRows = pfCtx->fpSet.process(pfCtx);
H
Haojun Liao 已提交
892 893 894
      } else {
        SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
        taosArrayPush(pBlockList, &pSrcBlock);
G
Ganlin Zhao 已提交
895

896
        SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
897
        SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
H
Haojun Liao 已提交
898

899
        SScalarParam dest = {.columnData = &idata};
X
Xiaoyu Wang 已提交
900
        int32_t      code = scalarCalculate((SNode*)pExpr[k].pExpr->_function.pFunctNode, pBlockList, &dest);
901 902 903 904
        if (code != TSDB_CODE_SUCCESS) {
          taosArrayDestroy(pBlockList);
          return code;
        }
905

dengyihao's avatar
dengyihao 已提交
906
        int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
907
        colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
908 909

        numOfRows = dest.numOfRows;
H
Haojun Liao 已提交
910 911
        taosArrayDestroy(pBlockList);
      }
912
    } else {
913
      ASSERT(0);
914 915
    }
  }
916

917 918 919
  if (!createNewColModel) {
    pResult->info.rows += numOfRows;
  }
920 921

  return TSDB_CODE_SUCCESS;
922 923
}

924 925 926
static void setResultRowKey(SResultRow* pResultRow, char* pData, int16_t type) {
  if (IS_VAR_DATA_TYPE(type)) {
    // todo disable this
927

928 929 930 931 932 933 934 935 936
    //    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);
937

938 939
    pResultRow->win.skey = v;
    pResultRow->win.ekey = v;
940 941 942
  }
}

943 944 945 946 947
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;
948

949 950 951
  SResultRow* pResultRow =
      doSetResultOutBufByKey(pBuf, pResultRowInfo, (char*)pData, bytes, true, groupId, pTaskInfo, false, pAggSup);
  assert(pResultRow != NULL);
952

953
  setResultRowKey(pResultRow, pData, type);
954
  setResultRowInitCtx(pResultRow, pCtx, numOfCols, binfo->rowCellInfoOffset);
955
  return TSDB_CODE_SUCCESS;
956 957
}

958 959
static bool functionNeedToExecute(SqlFunctionCtx* pCtx) {
  struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
960

961 962 963 964 965
  // in case of timestamp column, always generated results.
  int32_t functionId = pCtx->functionId;
  if (functionId == -1) {
    return false;
  }
966

967
  if (isRowEntryCompleted(pResInfo)) {
968 969 970
    return false;
  }

971 972 973 974 975 976 977
  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;
978 979
  }

980 981 982 983 984
  // 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;
  //  }
985 986 987 988

  return true;
}

989 990 991 992 993 994 995
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;
    }
996

997 998 999
    // Set the correct column info (data type and bytes)
    pInput->pData[paramIndex]->info.type = type;
    pInput->pData[paramIndex]->info.bytes = tDataTypes[type].bytes;
1000
  }
H
Haojun Liao 已提交
1001

1002 1003 1004 1005 1006 1007
  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;
1008 1009
    }
  } else {
1010
    da = pInput->pColumnDataAgg[paramIndex];
1011 1012
  }

1013
  ASSERT(!IS_VAR_DATA_TYPE(type));
1014

1015 1016 1017 1018 1019 1020
  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};
1021

1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
    *(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
1034
  } else {
1035
    ASSERT(0);
1036 1037
  }

1038 1039
  return TSDB_CODE_SUCCESS;
}
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050

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;

1051 1052
    for (int32_t j = 0; j < pExprInfo->base.numOfParams; ++j) {
      SFunctParam* pFuncParam = &pExprInfo->base.pParam[j];
1053

1054 1055
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
1056 1057 1058 1059
        pInput->pColumnDataAgg[j] = pBlock->pBlockAgg[slotId];
        if (pInput->pColumnDataAgg[j] == NULL) {
          pInput->colDataAggIsSet = false;
        }
1060 1061 1062 1063

        // 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);
1064 1065
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
        doCreateConstantValColumnAggInfo(pInput, pFuncParam, pFuncParam->param.nType, j, pBlock->info.rows);
1066 1067
      }
    }
1068
  } else {
1069
    pInput->colDataAggIsSet = false;
1070 1071 1072
  }

  // set the statistics data for primary time stamp column
1073 1074 1075 1076 1077
  //  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;
  //  }
1078 1079 1080
}

// set the output buffer for the selectivity + tag query
1081
static int32_t setSelectValueColumnInfo(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1082 1083
  int32_t num = 0;

H
Haojun Liao 已提交
1084
  SqlFunctionCtx*  p = NULL;
1085 1086
  SqlFunctionCtx** pValCtx = taosMemoryCalloc(numOfOutput, POINTER_BYTES);
  if (pValCtx == NULL) {
1087 1088 1089 1090
    return TSDB_CODE_QRY_OUT_OF_MEMORY;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
1091 1092
    if (strcmp(pCtx[i].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
      pValCtx[num++] = &pCtx[i];
1093
    } else if (fmIsAggFunc(pCtx[i].functionId)) {
1094
      p = &pCtx[i];
1095
    }
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
//    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
//    }
1106
  }
1107

1108
  if (p != NULL) {
1109
    p->subsidiaries.pCtx = pValCtx;
1110
    p->subsidiaries.num = num;
1111
  } else {
1112
    taosMemoryFreeClear(pValCtx);
1113 1114 1115 1116 1117
  }

  return TSDB_CODE_SUCCESS;
}

1118
SqlFunctionCtx* createSqlFunctionCtx(SExprInfo* pExprInfo, int32_t numOfOutput, int32_t** rowCellInfoOffset) {
L
Liu Jicong 已提交
1119
  SqlFunctionCtx* pFuncCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfOutput, sizeof(SqlFunctionCtx));
H
Haojun Liao 已提交
1120 1121 1122 1123
  if (pFuncCtx == NULL) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
1124
  *rowCellInfoOffset = taosMemoryCalloc(numOfOutput, sizeof(int32_t));
H
Haojun Liao 已提交
1125
  if (*rowCellInfoOffset == 0) {
wafwerar's avatar
wafwerar 已提交
1126
    taosMemoryFreeClear(pFuncCtx);
H
Haojun Liao 已提交
1127 1128 1129 1130
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
1131
    SExprInfo* pExpr = &pExprInfo[i];
H
Haojun Liao 已提交
1132

L
Liu Jicong 已提交
1133
    SExprBasicInfo* pFunct = &pExpr->base;
H
Haojun Liao 已提交
1134
    SqlFunctionCtx* pCtx = &pFuncCtx[i];
H
Haojun Liao 已提交
1135

1136
    pCtx->functionId = -1;
1137
    pCtx->curBufPage = -1;
L
Liu Jicong 已提交
1138
    pCtx->pExpr = pExpr;
1139

H
Haojun Liao 已提交
1140
    if (pExpr->pExpr->nodeType == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
1141
      SFuncExecEnv env = {0};
H
Haojun Liao 已提交
1142 1143
      pCtx->functionId = pExpr->pExpr->_function.pFunctNode->funcId;

H
Haojun Liao 已提交
1144
      if (fmIsAggFunc(pCtx->functionId) || fmIsNonstandardSQLFunc(pCtx->functionId)) {
1145 1146 1147 1148
        bool isUdaf = fmIsUserDefinedFunc(pCtx->functionId);
        if (!isUdaf) {
          fmGetFuncExecFuncs(pCtx->functionId, &pCtx->fpSet);
        } else {
1149
          char* udfName = pExpr->pExpr->_function.pFunctNode->functionName;
1150 1151 1152
          strncpy(pCtx->udfName, udfName, strlen(udfName));
          fmGetUdafExecFuncs(pCtx->functionId, &pCtx->fpSet);
        }
1153 1154 1155
        pCtx->fpSet.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
      } else {
        fmGetScalarFuncExecFuncs(pCtx->functionId, &pCtx->sfp);
1156 1157 1158
        if (pCtx->sfp.getEnv != NULL) {
          pCtx->sfp.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
        }
1159
      }
H
Haojun Liao 已提交
1160
      pCtx->resDataInfo.interBufSize = env.calcMemSize;
1161 1162
    } 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
1163
      // for simple column, the result buffer needs to hold at least one element.
1164
      pCtx->resDataInfo.interBufSize = pFunct->resSchema.bytes;
H
Haojun Liao 已提交
1165
    }
H
Haojun Liao 已提交
1166

H
Haojun Liao 已提交
1167
    pCtx->input.numOfInputCols = pFunct->numOfParams;
wafwerar's avatar
wafwerar 已提交
1168 1169
    pCtx->input.pData = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
    pCtx->input.pColumnDataAgg = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
H
Haojun Liao 已提交
1170

1171
    pCtx->pTsOutput = NULL;
L
Liu Jicong 已提交
1172
    pCtx->resDataInfo.bytes = pFunct->resSchema.bytes;
1173
    pCtx->resDataInfo.type = pFunct->resSchema.type;
L
Liu Jicong 已提交
1174
    pCtx->order = TSDB_ORDER_ASC;
1175
    pCtx->start.key = INT64_MIN;
L
Liu Jicong 已提交
1176
    pCtx->end.key = INT64_MIN;
1177
    pCtx->numOfParams = pExpr->base.numOfParams;
H
Haojun Liao 已提交
1178

1179
    pCtx->param = pFunct->pParam;
dengyihao's avatar
dengyihao 已提交
1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
    //    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
1214 1215
  }

L
Liu Jicong 已提交
1216 1217 1218
  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
1219
  }
H
Haojun Liao 已提交
1220

1221
  setSelectValueColumnInfo(pFuncCtx, numOfOutput);
H
Haojun Liao 已提交
1222 1223 1224
  return pFuncCtx;
}

1225
static void* destroySqlFunctionCtx(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1226 1227 1228 1229 1230 1231
  if (pCtx == NULL) {
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
    for (int32_t j = 0; j < pCtx[i].numOfParams; ++j) {
1232
      taosVariantDestroy(&pCtx[i].param[j].param);
1233 1234 1235
    }

    taosVariantDestroy(&pCtx[i].tag);
1236
    taosMemoryFreeClear(pCtx[i].subsidiaries.pCtx);
H
Haojun Liao 已提交
1237 1238
    taosMemoryFree(pCtx[i].input.pData);
    taosMemoryFree(pCtx[i].input.pColumnDataAgg);
1239 1240
  }

wafwerar's avatar
wafwerar 已提交
1241
  taosMemoryFreeClear(pCtx);
1242 1243 1244
  return NULL;
}

L
Liu Jicong 已提交
1245
bool isTaskKilled(SExecTaskInfo* pTaskInfo) {
1246 1247
  // query has been executed more than tsShellActivityTimer, and the retrieve has not arrived
  // abort current query execution.
L
Liu Jicong 已提交
1248 1249
  if (pTaskInfo->owner != 0 &&
      ((taosGetTimestampSec() - pTaskInfo->cost.start / 1000) > 10 * getMaximumIdleDurationSec())
1250 1251
      /*(!needBuildResAfterQueryComplete(pTaskInfo))*/) {
    assert(pTaskInfo->cost.start != 0);
L
Liu Jicong 已提交
1252 1253 1254
    //    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;
1255 1256 1257 1258 1259
  }

  return false;
}

L
Liu Jicong 已提交
1260
void setTaskKilled(SExecTaskInfo* pTaskInfo) { pTaskInfo->code = TSDB_CODE_TSC_QUERY_CANCELLED; }
1261

L
Liu Jicong 已提交
1262
static bool isCachedLastQuery(STaskAttr* pQueryAttr) {
1263 1264 1265 1266 1267 1268 1269 1270 1271
  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;
  }

1272 1273
  int32_t order = TSDB_ORDER_ASC;
  if (order != TSDB_ORDER_DESC || !TSWINDOW_IS_EQUAL(pQueryAttr->window, TSWINDOW_DESC_INITIALIZER)) {
1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292
    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
1293
// todo refactor : return window
1294
void getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int64_t key, STimeWindow* win) {
H
Haojun Liao 已提交
1295
  win->skey = taosTimeTruncate(key, pInterval, precision);
1296 1297

  /*
H
Haojun Liao 已提交
1298
   * if the realSkey > INT64_MAX - pInterval->interval, the query duration between
1299 1300
   * realSkey and realEkey must be less than one interval.Therefore, no need to adjust the query ranges.
   */
1301 1302
  win->ekey = taosTimeAdd(win->skey, pInterval->interval, pInterval->intervalUnit, precision) - 1;
  if (win->ekey < win->skey) {
1303 1304 1305 1306
    win->ekey = INT64_MAX;
  }
}

L
Liu Jicong 已提交
1307
static int32_t updateBlockLoadStatus(STaskAttr* pQuery, int32_t status) {
1308 1309 1310
  bool hasFirstLastFunc = false;
  bool hasOtherFunc = false;

1311
  if (status == BLK_DATA_DATA_LOAD || status == BLK_DATA_FILTEROUT) {
1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
    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;
    }
  }

1330
  if (hasFirstLastFunc && status == BLK_DATA_NOT_LOAD) {
L
Liu Jicong 已提交
1331
    if (!hasOtherFunc) {
1332
      return BLK_DATA_FILTEROUT;
1333
    } else {
1334
      return BLK_DATA_DATA_LOAD;
1335 1336 1337 1338 1339 1340
    }
  }

  return status;
}

L
Liu Jicong 已提交
1341 1342
// static void updateDataCheckOrder(SQInfo *pQInfo, SQueryTableReq* pQueryMsg, bool stableQuery) {
//   STaskAttr* pQueryAttr = pQInfo->runtimeEnv.pQueryAttr;
H
Haojun Liao 已提交
1343
//
L
Liu Jicong 已提交
1344 1345 1346 1347
//   // 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
1348
//
L
Liu Jicong 已提交
1349 1350 1351 1352 1353
//   // 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
1354
//
L
Liu Jicong 已提交
1355 1356
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     if (pQueryAttr->window.skey > pQueryAttr->window.ekey) {
wafwerar's avatar
wafwerar 已提交
1357
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1358
//     }
H
Haojun Liao 已提交
1359
//
L
Liu Jicong 已提交
1360 1361 1362
//     pQueryAttr->needReverseScan = false;
//     return;
//   }
H
Haojun Liao 已提交
1363
//
L
Liu Jicong 已提交
1364 1365 1366
//   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
1367
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1368
//     }
H
Haojun Liao 已提交
1369
//
L
Liu Jicong 已提交
1370 1371 1372 1373
//     pQueryAttr->needReverseScan = false;
//     doUpdateLastKey(pQueryAttr);
//     return;
//   }
H
Haojun Liao 已提交
1374
//
L
Liu Jicong 已提交
1375 1376 1377 1378 1379 1380
//   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
1381
//
L
Liu Jicong 已提交
1382 1383 1384
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     return;
//   }
H
Haojun Liao 已提交
1385
//
L
Liu Jicong 已提交
1386 1387 1388 1389
//   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
1390 1391
////               pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey);
//
wafwerar's avatar
wafwerar 已提交
1392
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
//        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
1403
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415
//        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
1416 1417
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1418
//
wafwerar's avatar
wafwerar 已提交
1419
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1420 1421 1422 1423 1424 1425 1426 1427
//          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
1428 1429
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1430
//
wafwerar's avatar
wafwerar 已提交
1431
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1432 1433 1434 1435 1436 1437 1438 1439 1440
//          doUpdateLastKey(pQueryAttr);
//        }
//
//        pQueryAttr->order.order = TSDB_ORDER_DESC;
//        pQueryAttr->needReverseScan = false;
//      }
//    }
//  }
//}
1441

H
Haojun Liao 已提交
1442 1443
static void getIntermediateBufInfo(STaskRuntimeEnv* pRuntimeEnv, int32_t* ps, int32_t* rowsize) {
  STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
L
Liu Jicong 已提交
1444
  int32_t    MIN_ROWS_PER_PAGE = 4;
1445

L
Liu Jicong 已提交
1446 1447
  *rowsize = (int32_t)(pQueryAttr->resultRowSize *
                       getRowNumForMultioutput(pQueryAttr, pQueryAttr->topBotQuery, pQueryAttr->stableQuery));
1448 1449 1450 1451
  int32_t overhead = sizeof(SFilePage);

  // one page contains at least two rows
  *ps = DEFAULT_INTERN_BUF_PAGE_SIZE;
L
Liu Jicong 已提交
1452
  while (((*rowsize) * MIN_ROWS_PER_PAGE) > (*ps) - overhead) {
1453 1454 1455 1456
    *ps = ((*ps) << 1u);
  }
}

L
Liu Jicong 已提交
1457 1458 1459
// static FORCE_INLINE bool doFilterByBlockStatistics(STaskRuntimeEnv* pRuntimeEnv, SDataStatis *pDataStatis,
// SqlFunctionCtx *pCtx, int32_t numOfRows) {
//   STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
1460
//
L
Liu Jicong 已提交
1461 1462 1463
//   if (pDataStatis == NULL || pQueryAttr->pFilters == NULL) {
//     return true;
//   }
1464
//
L
Liu Jicong 已提交
1465 1466
//   return filterRangeExecute(pQueryAttr->pFilters, pDataStatis, pQueryAttr->numOfCols, numOfRows);
// }
1467

H
Haojun Liao 已提交
1468
static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
1469 1470
  STimeWindow w = {0};

dengyihao's avatar
dengyihao 已提交
1471 1472
  TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
  TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
1473

1474
  if (true) {
L
Liu Jicong 已提交
1475
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
1476 1477 1478 1479 1480 1481
    assert(w.ekey >= pBlockInfo->window.skey);

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

L
Liu Jicong 已提交
1482 1483
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1484 1485 1486 1487 1488 1489 1490 1491 1492 1493
      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
1494
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
1495 1496 1497 1498 1499 1500
    assert(w.skey <= pBlockInfo->window.ekey);

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

L
Liu Jicong 已提交
1501 1502
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566
      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
1567
static void         doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes);
1568 1569

static uint32_t doFilterByBlockTimeWindow(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock) {
H
Haojun Liao 已提交
1570
  SqlFunctionCtx* pCtx = pTableScanInfo->pCtx;
1571
  uint32_t        status = BLK_DATA_NOT_LOAD;
1572 1573 1574 1575

  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
1576
    int32_t colId = pTableScanInfo->pExpr[i].base.pParam[0].pCol->colId;
1577 1578 1579

    // group by + first/last should not apply the first/last block filter
    if (functionId < 0) {
1580
      status |= BLK_DATA_DATA_LOAD;
1581 1582
      return status;
    } else {
L
Liu Jicong 已提交
1583
      //      status |= aAggs[functionId].dataReqFunc(&pTableScanInfo->pCtx[i], &pBlock->info.window, colId);
1584
      //      if ((status & BLK_DATA_DATA_LOAD) == BLK_DATA_DATA_LOAD) {
L
Liu Jicong 已提交
1585 1586
      //        return status;
      //      }
1587 1588 1589 1590 1591 1592
    }
  }

  return status;
}

L
Liu Jicong 已提交
1593 1594
int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                              uint32_t* status) {
1595
  *status = BLK_DATA_NOT_LOAD;
1596

H
Haojun Liao 已提交
1597
  pBlock->pDataBlock = NULL;
L
Liu Jicong 已提交
1598
  pBlock->pBlockAgg = NULL;
H
Haojun Liao 已提交
1599

L
Liu Jicong 已提交
1600 1601
  //  int64_t groupId = pRuntimeEnv->current->groupIndex;
  //  bool    ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
1602

H
Haojun Liao 已提交
1603
  STaskCostInfo* pCost = &pTaskInfo->cost;
1604

1605 1606
//  pCost->totalBlocks += 1;
//  pCost->totalRows += pBlock->info.rows;
H
Haojun Liao 已提交
1607
#if 0
1608 1609 1610
  // 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
1611
      (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && overlapWithTimeWindow(pTaskInfo, &pBlock->info))) {
1612
    (*status) = BLK_DATA_DATA_LOAD;
1613 1614 1615
  }

  // check if this data block is required to load
1616
  if ((*status) != BLK_DATA_DATA_LOAD) {
1617 1618 1619 1620 1621 1622 1623
    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
1624
      bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647
      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 {
1648
      (*status) = BLK_DATA_DATA_LOAD;
1649 1650 1651 1652
    }
  }

  SDataBlockInfo* pBlockInfo = &pBlock->info;
H
Haojun Liao 已提交
1653
//  *status = updateBlockLoadStatus(pRuntimeEnv->pQueryAttr, *status);
1654

1655
  if ((*status) == BLK_DATA_NOT_LOAD || (*status) == BLK_DATA_FILTEROUT) {
1656 1657
    //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//           pBlockInfo->window.ekey, pBlockInfo->rows);
1658
    pCost->skipBlocks += 1;
1659
  } else if ((*status) == BLK_DATA_SMA_LOAD) {
1660 1661
    // this function never returns error?
    pCost->loadBlockStatis += 1;
1662
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1663 1664

    if (pBlock->pBlockAgg == NULL) {  // data block statistics does not exist, load data block
1665
//      pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1666 1667 1668
      pCost->totalCheckedRows += pBlock->info.rows;
    }
  } else {
1669
    assert((*status) == BLK_DATA_DATA_LOAD);
1670 1671 1672

    // load the data block statistics to perform further filter
    pCost->loadBlockStatis += 1;
1673
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1674 1675 1676 1677 1678 1679

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

H
Haojun Liao 已提交
1680
          bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697
          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
1698
            pCost->skipBlocks += 1;
1699 1700
            //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId,
//                   pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
1701
            (*status) = BLK_DATA_FILTEROUT;
1702 1703 1704 1705 1706 1707 1708 1709
            return TSDB_CODE_SUCCESS;
          }
        }
      }
    }

    // current block has been discard due to filter applied
//    if (!doFilterByBlockStatistics(pRuntimeEnv, pBlock->pBlockAgg, pTableScanInfo->pCtx, pBlockInfo->rows)) {
1710
//      pCost->skipBlocks += 1;
1711 1712
//      qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//             pBlockInfo->window.ekey, pBlockInfo->rows);
1713
//      (*status) = BLK_DATA_FILTEROUT;
1714 1715 1716 1717 1718
//      return TSDB_CODE_SUCCESS;
//    }

    pCost->totalCheckedRows += pBlockInfo->rows;
    pCost->loadBlocks += 1;
1719
//    pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1720 1721 1722 1723 1724 1725 1726
//    if (pBlock->pDataBlock == NULL) {
//      return terrno;
//    }

//    if (pQueryAttr->pFilters != NULL) {
//      filterSetColFieldData(pQueryAttr->pFilters, pBlock->info.numOfCols, pBlock->pDataBlock);
//    }
1727

1728 1729 1730 1731
//    if (pQueryAttr->pFilters != NULL || pRuntimeEnv->pTsBuf != NULL) {
//      filterColRowsInDataBlock(pRuntimeEnv, pBlock, ascQuery);
//    }
  }
H
Haojun Liao 已提交
1732
#endif
1733 1734 1735 1736
  return TSDB_CODE_SUCCESS;
}

/*
H
Haojun Liao 已提交
1737
 * set tag value in SqlFunctionCtx
1738 1739
 * e.g.,tag information into input buffer
 */
L
Liu Jicong 已提交
1740
static void doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes) {
1741 1742 1743
  taosVariantDestroy(tag);

  char* val = NULL;
L
Liu Jicong 已提交
1744 1745 1746 1747 1748 1749
  //  if (tagColId == TSDB_TBNAME_COLUMN_INDEX) {
  //    val = tsdbGetTableName(pTable);
  //    assert(val != NULL);
  //  } else {
  //    val = tsdbGetTableTagVal(pTable, tagColId, type, bytes);
  //  }
1750 1751 1752 1753 1754 1755 1756 1757

  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
1758
    int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
1759
    taosVariantCreateFromBinary(tag, varDataVal(val), len, type);
L
Liu Jicong 已提交
1760
    // taosVariantCreateFromBinary(tag, varDataVal(val), varDataLen(val), type);
1761 1762 1763 1764 1765 1766 1767 1768
  } 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
1769
  for (int32_t i = 0; i < numOfTags; ++i) {
1770 1771 1772 1773 1774 1775 1776 1777
    if (pTagColList[i].colId == colId) {
      return &pTagColList[i];
    }
  }

  return NULL;
}

L
Liu Jicong 已提交
1778 1779
void setTagValue(SOperatorInfo* pOperatorInfo, void* pTable, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
  SExprInfo* pExpr = pOperatorInfo->pExpr;
1780
  SExprInfo* pExprInfo = &pExpr[0];
L
Liu Jicong 已提交
1781
  int32_t    functionId = getExprFunctionId(pExprInfo);
1782
#if 0
1783 1784 1785
  if (pQueryAttr->numOfOutput == 1 && functionId == FUNCTION_TS_COMP && pQueryAttr->stableQuery) {
    assert(pExprInfo->base.numOfParams == 1);

L
Liu Jicong 已提交
1786
    //    int16_t      tagColId = (int16_t)pExprInfo->base.param[0].i;
H
Haojun Liao 已提交
1787
    int16_t      tagColId = -1;
1788 1789 1790
    SColumnInfo* pColInfo = doGetTagColumnInfoById(pQueryAttr->tagColList, pQueryAttr->numOfTags, tagColId);

    doSetTagValueInParam(pTable, tagColId, &pCtx[0].tag, pColInfo->type, pColInfo->bytes);
H
Haojun Liao 已提交
1791

1792 1793 1794 1795 1796 1797 1798 1799 1800
  } 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
1801
      if (!TSDB_COL_IS_TAG(pLocalExprInfo->base.pParam[0].pCol->flag)) {
1802 1803 1804 1805
        continue;
      }

      // todo use tag column index to optimize performance
L
Liu Jicong 已提交
1806 1807
      doSetTagValueInParam(pTable, pLocalExprInfo->base.pParam[0].pCol->colId, &pCtx[idx].tag,
                           pLocalExprInfo->base.resSchema.type, pLocalExprInfo->base.resSchema.bytes);
1808

L
Liu Jicong 已提交
1809 1810 1811
      if (IS_NUMERIC_TYPE(pLocalExprInfo->base.resSchema.type) ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_BOOL ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_TIMESTAMP) {
1812 1813 1814 1815
        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
1816
        }
1817 1818 1819 1820 1821 1822 1823 1824 1825 1826
      }

      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);
  }
1827
#endif
1828 1829
}

H
Haojun Liao 已提交
1830
void copyToSDataBlock(SSDataBlock* pBlock, int32_t* offset, SGroupResInfo* pGroupResInfo, SDiskbasedBuf* pResBuf) {
1831 1832 1833 1834 1835 1836 1837
  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
1838 1839 1840
      //      if ((code = mergeIntoGroupResult(&pGroupResInfo, pRuntimeEnv, offset)) != TSDB_CODE_SUCCESS) {
      return;
      //      }
1841 1842
    }

L
Liu Jicong 已提交
1843
    //    doCopyToSDataBlock(pResBuf, pGroupResInfo, TSDB_ORDER_ASC, pBlock, );
1844 1845 1846 1847 1848 1849 1850 1851 1852 1853

    // 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
1854 1855 1856
    //    if (pBlock->info.rows >= threshold) {
    //      break;
    //    }
1857 1858 1859
  }
}

L
Liu Jicong 已提交
1860
static void updateTableQueryInfoForReverseScan(STableQueryInfo* pTableQueryInfo) {
1861 1862 1863 1864
  if (pTableQueryInfo == NULL) {
    return;
  }

wafwerar's avatar
wafwerar 已提交
1865
  //  TSWAP(pTableQueryInfo->win.skey, pTableQueryInfo->win.ekey);
L
Liu Jicong 已提交
1866
  //  pTableQueryInfo->lastKey = pTableQueryInfo->win.skey;
1867

L
Liu Jicong 已提交
1868 1869
  //  SWITCH_ORDER(pTableQueryInfo->cur.order);
  //  pTableQueryInfo->cur.vgroupIndex = -1;
1870 1871

  // set the index to be the end slot of result rows array
dengyihao's avatar
dengyihao 已提交
1872 1873 1874 1875 1876 1877
  //  SResultRowInfo* pResultRowInfo = &pTableQueryInfo->resInfo;
  //  if (pResultRowInfo->size > 0) {
  //    pResultRowInfo->curPos = pResultRowInfo->size - 1;
  //  } else {
  //    pResultRowInfo->curPos = -1;
  //  }
1878 1879
}

H
Haojun Liao 已提交
1880
void initResultRow(SResultRow* pResultRow) {
X
Xiaoyu Wang 已提交
1881
  //  pResultRow->pEntryInfo = (struct SResultRowEntryInfo*)((char*)pResultRow + sizeof(SResultRow));
1882 1883 1884 1885 1886
}

/*
 * 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
1887 1888 1889
 * +------------+-----------------result column 1------------+------------------result column 2-----------+
 * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
 * +------------+--------------------------------------------+--------------------------------------------+
1890 1891
 *           offset[0]                                  offset[1]                                   offset[2]
 */
1892
// TODO refactor: some function move away
X
Xiaoyu Wang 已提交
1893 1894
void setFunctionResultOutput(SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage, int32_t numOfExprs,
                             SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
1895 1896 1897
  SqlFunctionCtx* pCtx = pInfo->pCtx;
  SSDataBlock*    pDataBlock = pInfo->pRes;
  int32_t*        rowCellInfoOffset = pInfo->rowCellInfoOffset;
H
Haojun Liao 已提交
1898

H
Haojun Liao 已提交
1899
  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
H
Haojun Liao 已提交
1900
  initResultRowInfo(pResultRowInfo, 16);
H
Haojun Liao 已提交
1901

L
Liu Jicong 已提交
1902 1903
  int64_t     tid = 0;
  int64_t     groupId = 0;
1904 1905
  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
                                            pTaskInfo, false, pSup);
H
Haojun Liao 已提交
1906

H
Haojun Liao 已提交
1907
  ASSERT(pDataBlock->info.numOfCols == numOfExprs);
H
Haojun Liao 已提交
1908 1909 1910 1911
  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
1912
    pCtx[i].resultInfo = pEntry;
H
Haojun Liao 已提交
1913 1914 1915
    pCtx[i].currentStage = stage;

    // set the timestamp output buffer for top/bottom/diff query
L
Liu Jicong 已提交
1916 1917
    //    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
1918
    //      if (i > 0) pCtx[i].pTsOutput = pCtx[i-1].pOutput;
L
Liu Jicong 已提交
1919
    //    }
H
Haojun Liao 已提交
1920 1921 1922 1923 1924
  }

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

L
Liu Jicong 已提交
1925
void updateOutputBuf(SOptrBasicInfo* pBInfo, int32_t* bufCapacity, int32_t numOfInputRows) {
1926 1927
  SSDataBlock* pDataBlock = pBInfo->pRes;

L
Liu Jicong 已提交
1928
  int32_t newSize = pDataBlock->info.rows + numOfInputRows + 5;  // extra output buffer
1929
  if ((*bufCapacity) < newSize) {
L
Liu Jicong 已提交
1930 1931
    for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
      SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1932

wafwerar's avatar
wafwerar 已提交
1933
      char* p = taosMemoryRealloc(pColInfo->pData, newSize * pColInfo->info.bytes);
1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946
      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
1947
    SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1948 1949 1950 1951 1952
    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
1953 1954
    if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF ||
        functionId == FUNCTION_DERIVATIVE) {
dengyihao's avatar
dengyihao 已提交
1955
      //      if (i > 0) pBInfo->pCtx[i].pTsOutput = pBInfo->pCtx[i - 1].pOutput;
1956 1957 1958 1959
    }
  }
}

H
Haojun Liao 已提交
1960
void copyTsColoum(SSDataBlock* pRes, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1961 1962
  bool    needCopyTs = false;
  int32_t tsNum = 0;
L
Liu Jicong 已提交
1963
  char*   src = NULL;
1964 1965 1966 1967
  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
1968 1969
      if (i > 0 && pCtx[i - 1].functionId == FUNCTION_TS_DUMMY) {
        SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i - 1);  // find ts data
1970 1971
        src = pColRes->pData;
      }
L
Liu Jicong 已提交
1972
    } else if (functionId == FUNCTION_TS_DUMMY) {
1973 1974 1975 1976 1977 1978 1979 1980 1981 1982
      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
1983
    if (functionId == FUNCTION_TS_DUMMY) {
1984 1985 1986 1987 1988 1989
      SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i);
      memcpy(pColRes->pData, src, pColRes->info.bytes * pRes->info.rows);
    }
  }
}

H
Haojun Liao 已提交
1990
void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
1991 1992
  for (int32_t j = 0; j < size; ++j) {
    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
dengyihao's avatar
dengyihao 已提交
1993 1994
    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
        fmIsScalarFunc(pCtx[j].functionId)) {
1995 1996 1997
      continue;
    }

H
Haojun Liao 已提交
1998
    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
1999 2000 2001
  }
}

L
Liu Jicong 已提交
2002
void setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status) {
2003
  if (status == TASK_NOT_COMPLETED) {
H
Haojun Liao 已提交
2004
    pTaskInfo->status = status;
2005 2006
  } else {
    // QUERY_NOT_COMPLETED is not compatible with any other status, so clear its position first
2007
    CLEAR_QUERY_STATUS(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
2008
    pTaskInfo->status |= status;
2009 2010 2011
  }
}

2012
// todo merged with the build group result.
2013 2014
void finalizeMultiTupleQueryResult(int32_t numOfOutput, SDiskbasedBuf* pBuf, SResultRowInfo* pResultRowInfo,
                                   int32_t* rowCellInfoOffset) {
2015 2016 2017
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
    SResultRowPosition* pPos = &pResultRowInfo->pPosition[i];

L
Liu Jicong 已提交
2018
    SFilePage*  bufPage = getBufPage(pBuf, pPos->pageId);
2019
    SResultRow* pRow = (SResultRow*)((char*)bufPage + pPos->offset);
H
Haojun Liao 已提交
2020 2021

    // TODO ignore the close status anyway.
dengyihao's avatar
dengyihao 已提交
2022 2023 2024
    //    if (!isResultRowClosed(pRow)) {
    //      continue;
    //    }
2025

2026
    for (int32_t j = 0; j < numOfOutput; ++j) {
2027
      struct SResultRowEntryInfo* pResInfo = getResultCell(pRow, j, rowCellInfoOffset);
H
Haojun Liao 已提交
2028
      if (!isRowEntryInitialized(pResInfo)) {
2029 2030 2031 2032 2033 2034
        continue;
      }

      if (pRow->numOfRows < pResInfo->numOfRes) {
        pRow->numOfRows = pResInfo->numOfRes;
      }
2035
    }
2036 2037 2038

    releaseBufPage(pBuf, bufPage);
  }
2039 2040
}

2041
STableQueryInfo* createTableQueryInfo(void* buf, STimeWindow win) {
L
Liu Jicong 已提交
2042
  STableQueryInfo* pTableQueryInfo = buf;
2043 2044 2045 2046
  pTableQueryInfo->lastKey = win.skey;
  return pTableQueryInfo;
}

L
Liu Jicong 已提交
2047
void destroyTableQueryInfoImpl(STableQueryInfo* pTableQueryInfo) {
2048 2049 2050 2051
  if (pTableQueryInfo == NULL) {
    return;
  }

L
Liu Jicong 已提交
2052
  //  taosVariantDestroy(&pTableQueryInfo->tag);
dengyihao's avatar
dengyihao 已提交
2053
  //  cleanupResultRowInfo(&pTableQueryInfo->resInfo);
2054 2055
}

2056
void setResultRowInitCtx(SResultRow* pResult, SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t* rowCellInfoOffset) {
2057 2058 2059 2060 2061 2062 2063
  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;
    }
2064 2065 2066 2067 2068

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

2069 2070 2071 2072 2073 2074
    if (!pResInfo->initialized) {
      if (pCtx[i].functionId != -1) {
        pCtx[i].fpSet.init(&pCtx[i], pResInfo);
      } else {
        pResInfo->initialized = true;
      }
2075 2076 2077 2078
    }
  }
}

2079 2080 2081 2082 2083 2084
void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock) {
  if (pFilterNode == NULL) {
    return;
  }

  SFilterInfo* filter = NULL;
H
Haojun Liao 已提交
2085

H
Haojun Liao 已提交
2086
  // todo move to the initialization function
H
Haojun Liao 已提交
2087
  int32_t code = filterInitFromNode((SNode*)pFilterNode, &filter, 0);
2088 2089 2090 2091 2092 2093

  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
2094
  filterFreeInfo(filter);
2095

2096
  SSDataBlock* px = createOneDataBlock(pBlock, false);
2097 2098
  blockDataEnsureCapacity(px, pBlock->info.rows);

H
Haojun Liao 已提交
2099
  // todo extract method
2100 2101 2102 2103
  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
2104 2105 2106 2107 2108 2109 2110 2111 2112
    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;
        }
2113

D
dapan1121 已提交
2114 2115 2116 2117 2118 2119
        if (colDataIsNull_s(pSrc, j)) {
          colDataAppendNULL(pDst, numOfRow);
        } else {
          colDataAppend(pDst, numOfRow, colDataGetData(pSrc, j), false);
        }
        numOfRow += 1;
H
Haojun Liao 已提交
2120
      }
D
dapan1121 已提交
2121 2122
    } else {
      numOfRow = 0;
2123 2124 2125 2126 2127 2128
    }

    *pSrc = *pDst;
  }

  pBlock->info.rows = numOfRow;
2129
  blockDataUpdateTsWindow(pBlock);
2130 2131
}

dengyihao's avatar
dengyihao 已提交
2132 2133
void doSetTableGroupOutputBuf(SAggOperatorInfo* pAggInfo, int32_t numOfOutput, uint64_t groupId,
                              SExecTaskInfo* pTaskInfo) {
2134 2135 2136
  // for simple group by query without interval, all the tables belong to one group result.
  int64_t uid = 0;

2137 2138
  SResultRowInfo* pResultRowInfo = &pAggInfo->binfo.resultRowInfo;
  SqlFunctionCtx* pCtx = pAggInfo->binfo.pCtx;
L
Liu Jicong 已提交
2139
  int32_t*        rowCellInfoOffset = pAggInfo->binfo.rowCellInfoOffset;
2140

2141
  SResultRow* pResultRow = doSetResultOutBufByKey(pAggInfo->aggSup.pResultBuf, pResultRowInfo, (char*)&groupId,
L
Liu Jicong 已提交
2142
                                                  sizeof(groupId), true, groupId, pTaskInfo, false, &pAggInfo->aggSup);
L
Liu Jicong 已提交
2143
  assert(pResultRow != NULL);
2144 2145 2146 2147 2148 2149

  /*
   * 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
2150 2151
    int32_t ret =
        addNewWindowResultBuf(pResultRow, pAggInfo->aggSup.pResultBuf, groupId, pAggInfo->binfo.pRes->info.rowSize);
2152 2153 2154 2155 2156
    if (ret != TSDB_CODE_SUCCESS) {
      return;
    }
  }

2157
  setResultRowInitCtx(pResultRow, pCtx, numOfOutput, rowCellInfoOffset);
2158 2159
}

H
Haojun Liao 已提交
2160 2161
void setExecutionContext(int32_t numOfOutput, uint64_t groupId, SExecTaskInfo* pTaskInfo, SAggOperatorInfo* pAggInfo) {
  if (pAggInfo->groupId != INT32_MIN && pAggInfo->groupId == groupId) {
2162 2163 2164
    return;
  }

H
Haojun Liao 已提交
2165
  doSetTableGroupOutputBuf(pAggInfo, numOfOutput, groupId, pTaskInfo);
2166 2167

  // record the current active group id
H
Haojun Liao 已提交
2168
  pAggInfo->groupId = groupId;
2169 2170 2171 2172 2173 2174 2175 2176 2177 2178
}

/**
 * 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
 */
2179
int32_t doCopyToSDataBlock(SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf, SGroupResInfo* pGroupResInfo,
2180
                           int32_t orderType, int32_t* rowCellOffset, SqlFunctionCtx* pCtx) {
2181
  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
L
Liu Jicong 已提交
2182
  int32_t numOfResult = pBlock->info.rows;  // there are already exists result rows
2183 2184

  int32_t start = 0;
2185
  int32_t step = 1;
2186

L
Liu Jicong 已提交
2187
  // qDebug("QInfo:0x%"PRIx64" start to copy data from windowResInfo to output buf", GET_TASKID(pRuntimeEnv));
2188 2189 2190 2191 2192 2193 2194 2195 2196
  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
2197 2198
    SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
    SFilePage*  page = getBufPage(pBuf, pPos->pos.pageId);
2199

2200
    SResultRow* pRow = (SResultRow*)((char*)page + pPos->pos.offset);
2201 2202 2203 2204 2205
    if (pRow->numOfRows == 0) {
      pGroupResInfo->index += 1;
      continue;
    }

2206
    // TODO copy multiple rows?
2207
    int32_t numOfRowsToCopy = pRow->numOfRows;
2208
    if (numOfResult + numOfRowsToCopy >= pBlock->info.capacity) {
2209 2210 2211 2212 2213 2214
      break;
    }

    pGroupResInfo->index += 1;

    for (int32_t j = 0; j < pBlock->info.numOfCols; ++j) {
2215 2216
      int32_t slotId = pExprInfo[j].base.resSchema.slotId;

2217 2218
      pCtx[j].resultInfo = getResultCell(pRow, j, rowCellOffset);
      if (pCtx[j].fpSet.process) {
2219
        pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
2220 2221
      } else if (strcmp(pCtx[j].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
        // do nothing, todo refactor
2222
      } else {
2223 2224 2225 2226 2227 2228 2229
        // 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);
          }
2230
      }
2231 2232
    }

2233 2234
    releaseBufPage(pBuf, page);

2235 2236
    pBlock->info.rows += pRow->numOfRows;
    if (pBlock->info.rows >= pBlock->info.capacity) {  // output buffer is full
2237 2238 2239 2240
      break;
    }
  }

L
Liu Jicong 已提交
2241
  // qDebug("QInfo:0x%"PRIx64" copy data to query buf completed", GET_TASKID(pRuntimeEnv));
2242
  blockDataUpdateTsWindow(pBlock);
2243 2244 2245
  return 0;
}

X
Xiaoyu Wang 已提交
2246 2247
void doBuildResultDatablock(SOptrBasicInfo* pbInfo, SGroupResInfo* pGroupResInfo, SExprInfo* pExprInfo,
                            SDiskbasedBuf* pBuf) {
2248 2249
  assert(pGroupResInfo->currentGroup <= pGroupResInfo->totalGroup);

X
Xiaoyu Wang 已提交
2250 2251 2252
  int32_t*        rowCellOffset = pbInfo->rowCellInfoOffset;
  SSDataBlock*    pBlock = pbInfo->pRes;
  SqlFunctionCtx* pCtx = pbInfo->pCtx;
2253

2254
  blockDataCleanup(pBlock);
2255 2256 2257 2258
  if (!hasRemainDataInCurrentGroup(pGroupResInfo)) {
    return;
  }

2259
  int32_t orderType = TSDB_ORDER_ASC;
2260
  doCopyToSDataBlock(pBlock, pExprInfo, pBuf, pGroupResInfo, orderType, rowCellOffset, pCtx);
2261

H
Haojun Liao 已提交
2262 2263
  // add condition (pBlock->info.rows >= 1) just to runtime happy
  blockDataUpdateTsWindow(pBlock);
2264 2265
}

L
Liu Jicong 已提交
2266 2267
static void updateNumOfRowsInResultRows(SqlFunctionCtx* pCtx, int32_t numOfOutput, SResultRowInfo* pResultRowInfo,
                                        int32_t* rowCellInfoOffset) {
2268
  // update the number of result for each, only update the number of rows for the corresponding window result.
L
Liu Jicong 已提交
2269 2270 2271
  //  if (QUERY_IS_INTERVAL_QUERY(pQueryAttr)) {
  //    return;
  //  }
H
Haojun Liao 已提交
2272
#if 0
2273
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
L
Liu Jicong 已提交
2274
    SResultRow* pResult = pResultRowInfo->pResult[i];
2275 2276 2277 2278 2279 2280 2281

    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;
      }

2282 2283
      SResultRowEntryInfo* pCell = getResultCell(pResult, j, rowCellInfoOffset);
      pResult->numOfRows = (uint16_t)(TMAX(pResult->numOfRows, pCell->numOfRes));
2284 2285
    }
  }
H
Haojun Liao 已提交
2286
#endif
2287 2288
}

L
Liu Jicong 已提交
2289
static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
2290 2291
  int32_t colSize = pColRes->info.bytes * numOfRows;
  return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
L
Liu Jicong 已提交
2292
                                                      colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
2293 2294
}

2295
int32_t doFillTimeIntervalGapsInResults(struct SFillInfo* pFillInfo, SSDataBlock* pBlock, int32_t capacity) {
L
Liu Jicong 已提交
2296 2297 2298 2299
  //  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);
  //  }
2300

2301 2302
  int32_t numOfRows = (int32_t)taosFillResultDataBlock(pFillInfo, pBlock, capacity - pBlock->info.rows);
  pBlock->info.rows += numOfRows;
2303

2304
  return pBlock->info.rows;
2305 2306
}

2307
void publishOperatorProfEvent(SOperatorInfo* pOperator, EQueryProfEventType eventType) {
2308 2309
  SQueryProfEvent event = {0};

L
Liu Jicong 已提交
2310 2311
  event.eventType = eventType;
  event.eventTime = taosGetTimestampUs();
2312
  event.operatorType = pOperator->operatorType;
dengyihao's avatar
dengyihao 已提交
2313 2314 2315
  //    if (pQInfo->summary.queryProfEvents) {
  //      taosArrayPush(pQInfo->summary.queryProfEvents, &event);
  //    }
2316 2317
}

L
Liu Jicong 已提交
2318
void publishQueryAbortEvent(SExecTaskInfo* pTaskInfo, int32_t code) {
2319 2320 2321 2322 2323
  SQueryProfEvent event;
  event.eventType = QUERY_PROF_QUERY_ABORT;
  event.eventTime = taosGetTimestampUs();
  event.abortCode = code;

2324 2325
  if (pTaskInfo->cost.queryProfEvents) {
    taosArrayPush(pTaskInfo->cost.queryProfEvents, &event);
2326 2327 2328
  }
}

L
Liu Jicong 已提交
2329
typedef struct {
2330 2331 2332 2333 2334 2335 2336
  uint8_t operatorType;
  int64_t beginTime;
  int64_t endTime;
  int64_t selfTime;
  int64_t descendantsTime;
} SOperatorStackItem;

L
Liu Jicong 已提交
2337 2338
static void doOperatorExecProfOnce(SOperatorStackItem* item, SQueryProfEvent* event, SArray* opStack,
                                   SHashObj* profResults) {
2339 2340 2341 2342 2343 2344 2345 2346
  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
2347
  uint8_t              operatorType = item->operatorType;
2348 2349 2350 2351 2352 2353 2354 2355 2356
  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
2357
    taosHashPut(profResults, &(operatorType), sizeof(operatorType), &opResult, sizeof(opResult));
2358 2359 2360
  }
}

2361
void calculateOperatorProfResults(void) {
dengyihao's avatar
dengyihao 已提交
2362 2363 2364 2365 2366 2367 2368 2369 2370
  //  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;
  //  }
2371 2372 2373 2374 2375

  SArray* opStack = taosArrayInit(32, sizeof(SOperatorStackItem));
  if (opStack == NULL) {
    return;
  }
2376
#if 0
L
Liu Jicong 已提交
2377
  size_t    size = taosArrayGetSize(pQInfo->summary.queryProfEvents);
2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398
  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);
      }
    }
  }
2399
#endif
2400 2401 2402
  taosArrayDestroy(opStack);
}

L
Liu Jicong 已提交
2403 2404
void queryCostStatis(SExecTaskInfo* pTaskInfo) {
  STaskCostInfo* pSummary = &pTaskInfo->cost;
2405

L
Liu Jicong 已提交
2406 2407 2408
  //  uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
  //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
  //  pSummary->hashSize = hashSize;
2409 2410 2411 2412

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

L
Liu Jicong 已提交
2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423
  //  SResultRowPool* p = pTaskInfo->pool;
  //  if (p != NULL) {
  //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
  //    pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
  //  } else {
  //    pSummary->winInfoSize = 0;
  //    pSummary->numOfTimeWindows = 0;
  //  }
  //
  //  calculateOperatorProfResults(pQInfo);

2424 2425
  SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
  if (pSummary->pRecoder != NULL) {
X
Xiaoyu Wang 已提交
2426 2427
    qDebug("%s :cost summary: elapsed time:%" PRId64 " us, first merge:%" PRId64
           " us, total blocks:%d, "
2428 2429 2430 2431
           "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
2432 2433 2434
  // 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);
2435 2436 2437 2438

  if (pSummary->operatorProfResults) {
    SOperatorProfResult* opRes = taosHashIterate(pSummary->operatorProfResults, NULL);
    while (opRes != NULL) {
L
Liu Jicong 已提交
2439 2440
      // qDebug("QInfo:0x%" PRIx64 " :cost summary: operator : %d, exec times: %" PRId64 ", self time: %" PRId64,
      //             pQInfo->qId, opRes->operatorType, opRes->sumRunTimes, opRes->sumSelfTime);
2441 2442 2443 2444 2445
      opRes = taosHashIterate(pSummary->operatorProfResults, opRes);
    }
  }
}

L
Liu Jicong 已提交
2446 2447 2448
// static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
2449
//
L
Liu Jicong 已提交
2450
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2451
//
L
Liu Jicong 已提交
2452 2453 2454 2455
//   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;
//   }
2456
//
L
Liu Jicong 已提交
2457 2458 2459 2460 2461
//   if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//     pQueryAttr->pos = (int32_t)pQueryAttr->limit.offset;
//   } else {
//     pQueryAttr->pos = pBlockInfo->rows - (int32_t)pQueryAttr->limit.offset - 1;
//   }
2462
//
L
Liu Jicong 已提交
2463
//   assert(pQueryAttr->pos >= 0 && pQueryAttr->pos <= pBlockInfo->rows - 1);
2464
//
L
Liu Jicong 已提交
2465 2466
//   SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//   SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2467
//
L
Liu Jicong 已提交
2468 2469
//   // update the pQueryAttr->limit.offset value, and pQueryAttr->pos value
//   TSKEY *keys = (TSKEY *) pColInfoData->pData;
2470
//
L
Liu Jicong 已提交
2471 2472 2473
//   // update the offset value
//   pTableQueryInfo->lastKey = keys[pQueryAttr->pos];
//   pQueryAttr->limit.offset = 0;
2474
//
L
Liu Jicong 已提交
2475
//   int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
2476
//
L
Liu Jicong 已提交
2477 2478 2479 2480
//   //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);
// }
2481

L
Liu Jicong 已提交
2482 2483
// void skipBlocks(STaskRuntimeEnv *pRuntimeEnv) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
2484
//
L
Liu Jicong 已提交
2485 2486 2487
//   if (pQueryAttr->limit.offset <= 0 || pQueryAttr->numOfFilterCols > 0) {
//     return;
//   }
2488
//
L
Liu Jicong 已提交
2489 2490
//   pQueryAttr->pos = 0;
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2491
//
L
Liu Jicong 已提交
2492 2493
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
//   TsdbQueryHandleT pTsdbReadHandle = pRuntimeEnv->pTsdbReadHandle;
2494
//
L
Liu Jicong 已提交
2495 2496 2497 2498 2499
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pTsdbReadHandle)) {
//     if (isTaskKilled(pRuntimeEnv->qinfo)) {
//       longjmp(pRuntimeEnv->env, TSDB_CODE_TSC_QUERY_CANCELLED);
//     }
2500
//
L
Liu Jicong 已提交
2501
//     tsdbRetrieveDataBlockInfo(pTsdbReadHandle, &blockInfo);
2502
//
L
Liu Jicong 已提交
2503 2504 2505 2506
//     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;
2507
//
L
Liu Jicong 已提交
2508 2509 2510 2511 2512 2513 2514
//       //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;
//     }
//   }
2515
//
L
Liu Jicong 已提交
2516 2517 2518 2519 2520 2521 2522 2523 2524
//   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;
2525
//
L
Liu Jicong 已提交
2526 2527 2528
//   assert(pQueryAttr->limit.offset == 0);
//   STimeWindow tw = *win;
//   getNextTimeWindow(pQueryAttr, &tw);
2529
//
L
Liu Jicong 已提交
2530 2531
//   if ((tw.skey <= pBlockInfo->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) ||
//       (tw.ekey >= pBlockInfo->window.skey && !QUERY_IS_ASC_QUERY(pQueryAttr))) {
2532
//
L
Liu Jicong 已提交
2533 2534 2535 2536
//     // 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);
2537
//
L
Liu Jicong 已提交
2538 2539 2540 2541
//     tw = *win;
//     int32_t startPos =
//         getNextQualifiedWindow(pQueryAttr, &tw, pBlockInfo, pColInfoData->pData, binarySearchForKey, -1);
//     assert(startPos >= 0);
2542
//
L
Liu Jicong 已提交
2543 2544
//     // set the abort info
//     pQueryAttr->pos = startPos;
2545
//
L
Liu Jicong 已提交
2546 2547 2548 2549
//     // reset the query start timestamp
//     pTableQueryInfo->win.skey = ((TSKEY *)pColInfoData->pData)[startPos];
//     pQueryAttr->window.skey = pTableQueryInfo->win.skey;
//     TSKEY key = pTableQueryInfo->win.skey;
2550
//
L
Liu Jicong 已提交
2551 2552
//     pWindowResInfo->prevSKey = tw.skey;
//     int32_t index = pRuntimeEnv->resultRowInfo.curIndex;
2553
//
L
Liu Jicong 已提交
2554 2555
//     int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
//     pRuntimeEnv->resultRowInfo.curIndex = index;  // restore the window index
2556
//
L
Liu Jicong 已提交
2557 2558 2559 2560
//     //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);
2561
//
L
Liu Jicong 已提交
2562 2563 2564 2565 2566
//     return key;
//   } else {  // do nothing
//     pQueryAttr->window.skey      = tw.skey;
//     pWindowResInfo->prevSKey = tw.skey;
//     pTableQueryInfo->lastKey = tw.skey;
2567
//
L
Liu Jicong 已提交
2568 2569
//     return tw.skey;
//   }
2570
//
L
Liu Jicong 已提交
2571 2572 2573 2574 2575 2576 2577 2578 2579 2580
//   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);
//   }
2581
//
L
Liu Jicong 已提交
2582 2583 2584 2585 2586
//   // 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;
//   }
2587
//
L
Liu Jicong 已提交
2588 2589 2590 2591 2592 2593 2594
//   /*
//    * 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);
2595
//
L
Liu Jicong 已提交
2596 2597
//   STimeWindow w = TSWINDOW_INITIALIZER;
//   bool ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
2598
//
L
Liu Jicong 已提交
2599 2600
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
//   STableQueryInfo *pTableQueryInfo = pRuntimeEnv->current;
2601
//
L
Liu Jicong 已提交
2602 2603 2604
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pRuntimeEnv->pTsdbReadHandle)) {
//     tsdbRetrieveDataBlockInfo(pRuntimeEnv->pTsdbReadHandle, &blockInfo);
2605
//
L
Liu Jicong 已提交
2606 2607 2608 2609 2610 2611 2612 2613 2614
//     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;
//     }
2615
//
L
Liu Jicong 已提交
2616 2617
//     // the first time window
//     STimeWindow win = getActiveTimeWindow(pWindowResInfo, pWindowResInfo->prevSKey, pQueryAttr);
2618
//
L
Liu Jicong 已提交
2619 2620
//     while (pQueryAttr->limit.offset > 0) {
//       STimeWindow tw = win;
2621
//
L
Liu Jicong 已提交
2622 2623 2624
//       if ((win.ekey <= blockInfo.window.ekey && ascQuery) || (win.ekey >= blockInfo.window.skey && !ascQuery)) {
//         pQueryAttr->limit.offset -= 1;
//         pWindowResInfo->prevSKey = win.skey;
2625
//
L
Liu Jicong 已提交
2626 2627 2628 2629 2630 2631
//         // 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;
//         }
//       }
2632
//
L
Liu Jicong 已提交
2633 2634 2635 2636
//       if (pQueryAttr->limit.offset == 0) {
//         *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
//         return true;
//       }
2637
//
L
Liu Jicong 已提交
2638 2639
//       // current window does not ended in current data block, try next data block
//       getNextTimeWindow(pQueryAttr, &tw);
2640
//
L
Liu Jicong 已提交
2641 2642 2643 2644 2645 2646 2647 2648 2649
//       /*
//        * 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)) {
2650
//
L
Liu Jicong 已提交
2651 2652
//         SArray *pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//         SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2653
//
L
Liu Jicong 已提交
2654 2655 2656
//         if ((win.ekey > blockInfo.window.ekey && ascQuery) || (win.ekey < blockInfo.window.skey && !ascQuery)) {
//           pQueryAttr->limit.offset -= 1;
//         }
2657
//
L
Liu Jicong 已提交
2658 2659 2660 2661 2662 2663 2664 2665
//         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);
2666
//
L
Liu Jicong 已提交
2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677
//           // 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.
//       }
//     }
//   }
2678
//
L
Liu Jicong 已提交
2679 2680 2681 2682
//   // check for error
//   if (terrno != TSDB_CODE_SUCCESS) {
//     longjmp(pRuntimeEnv->env, terrno);
//   }
2683
//
L
Liu Jicong 已提交
2684 2685
//   return true;
// }
2686

2687
int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t num) {
H
Haojun Liao 已提交
2688
  if (p->pDownstream == NULL) {
H
Haojun Liao 已提交
2689
    assert(p->numOfDownstream == 0);
2690 2691
  }

wafwerar's avatar
wafwerar 已提交
2692
  p->pDownstream = taosMemoryCalloc(1, num * POINTER_BYTES);
2693 2694 2695 2696 2697 2698 2699
  if (p->pDownstream == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

  memcpy(p->pDownstream, pDownstream, num * POINTER_BYTES);
  p->numOfDownstream = num;
  return TSDB_CODE_SUCCESS;
2700 2701 2702 2703
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo);

2704
static void doTableQueryInfoTimeWindowCheck(SExecTaskInfo* pTaskInfo, STableQueryInfo* pTableQueryInfo, int32_t order) {
H
Haojun Liao 已提交
2705 2706
#if 0
    if (order == TSDB_ORDER_ASC) {
2707 2708
    assert(
        (pTableQueryInfo->win.skey <= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2709 2710
        (pTableQueryInfo->lastKey >= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey >= pTaskInfo->window.skey && pTableQueryInfo->win.ekey <= pTaskInfo->window.ekey));
2711 2712 2713
  } else {
    assert(
        (pTableQueryInfo->win.skey >= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2714 2715
        (pTableQueryInfo->lastKey <= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey <= pTaskInfo->window.skey && pTableQueryInfo->win.ekey >= pTaskInfo->window.ekey));
2716
  }
H
Haojun Liao 已提交
2717
#endif
2718 2719
}

L
Liu Jicong 已提交
2720 2721 2722 2723
// 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;
2724
//
L
Liu Jicong 已提交
2725 2726 2727
//   if (pTableQueryInfo->pTable == NULL) {
//     return;
//   }
2728
//
L
Liu Jicong 已提交
2729 2730 2731 2732 2733 2734 2735 2736 2737
//   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));
//   }
// }
2738

2739
int32_t loadRemoteDataCallback(void* param, const SDataBuf* pMsg, int32_t code) {
L
Liu Jicong 已提交
2740
  SSourceDataInfo* pSourceDataInfo = (SSourceDataInfo*)param;
H
Haojun Liao 已提交
2741 2742
  if (code == TSDB_CODE_SUCCESS) {
    pSourceDataInfo->pRsp = pMsg->pData;
2743

H
Haojun Liao 已提交
2744 2745
    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
    pRsp->numOfRows = htonl(pRsp->numOfRows);
dengyihao's avatar
dengyihao 已提交
2746
    pRsp->compLen = htonl(pRsp->compLen);
2747
    pRsp->numOfCols = htonl(pRsp->numOfCols);
dengyihao's avatar
dengyihao 已提交
2748
    pRsp->useconds = htobe64(pRsp->useconds);
H
Haojun Liao 已提交
2749 2750 2751
  } else {
    pSourceDataInfo->code = code;
  }
H
Haojun Liao 已提交
2752

H
Haojun Liao 已提交
2753
  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
2754
  tsem_post(&pSourceDataInfo->pEx->ready);
wmmhello's avatar
wmmhello 已提交
2755
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
2756 2757 2758 2759
}

static void destroySendMsgInfo(SMsgSendInfo* pMsgBody) {
  assert(pMsgBody != NULL);
wafwerar's avatar
wafwerar 已提交
2760 2761
  taosMemoryFreeClear(pMsgBody->msgInfo.pData);
  taosMemoryFreeClear(pMsgBody);
H
Haojun Liao 已提交
2762 2763
}

S
Shengliang Guan 已提交
2764
void qProcessFetchRsp(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
L
Liu Jicong 已提交
2765
  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->ahandle;
H
Haojun Liao 已提交
2766 2767 2768 2769 2770
  assert(pMsg->ahandle != NULL);

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

  if (pMsg->contLen > 0) {
wafwerar's avatar
wafwerar 已提交
2771
    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
H
Haojun Liao 已提交
2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782
    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);
2783 2784
}

L
Liu Jicong 已提交
2785
static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
2786
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
2787

wafwerar's avatar
wafwerar 已提交
2788
  SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
2789 2790 2791 2792
  if (NULL == pMsg) {
    pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return pTaskInfo->code;
  }
2793

L
Liu Jicong 已提交
2794 2795
  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
2796

L
Liu Jicong 已提交
2797 2798
  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);
2799 2800 2801 2802 2803 2804 2805

  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
2806
  SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
2807
  if (NULL == pMsgSendInfo) {
wafwerar's avatar
wafwerar 已提交
2808
    taosMemoryFreeClear(pMsg);
2809 2810 2811
    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
2812 2813
  }

2814 2815 2816 2817 2818
  pMsgSendInfo->param = pDataInfo;
  pMsgSendInfo->msgInfo.pData = pMsg;
  pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
  pMsgSendInfo->msgType = TDMT_VND_FETCH;
  pMsgSendInfo->fp = loadRemoteDataCallback;
2819

2820
  int64_t transporterId = 0;
L
Liu Jicong 已提交
2821
  int32_t code = asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
2822 2823 2824
  return TSDB_CODE_SUCCESS;
}

2825
// NOTE: sources columns are more than the destination SSDatablock columns.
2826
void relocateColumnData(SSDataBlock* pBlock, const SArray* pColMatchInfo, SArray* pCols) {
2827 2828 2829
  size_t numOfSrcCols = taosArrayGetSize(pCols);

  int32_t i = 0, j = 0;
L
Liu Jicong 已提交
2830
  while (i < numOfSrcCols && j < taosArrayGetSize(pColMatchInfo)) {
2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849
    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);
    }
  }
}

2850
int32_t setSDataBlockFromFetchRsp(SSDataBlock* pRes, SLoadRemoteDataInfo* pLoadInfo, int32_t numOfRows, char* pData,
dengyihao's avatar
dengyihao 已提交
2851 2852
                                  int32_t compLen, int32_t numOfOutput, int64_t startTs, uint64_t* total,
                                  SArray* pColList) {
H
Haojun Liao 已提交
2853
  if (pColList == NULL) {  // data from other sources
H
Haojun Liao 已提交
2854 2855
    blockDataEnsureCapacity(pRes, numOfRows);

L
Liu Jicong 已提交
2856
    int32_t dataLen = *(int32_t*)pData;
2857 2858
    pData += sizeof(int32_t);

L
Liu Jicong 已提交
2859
    pRes->info.groupId = *(uint64_t*)pData;
2860 2861
    pData += sizeof(uint64_t);

H
Haojun Liao 已提交
2862 2863
    int32_t* colLen = (int32_t*)pData;

2864
    char* pStart = pData + sizeof(int32_t) * numOfOutput;
2865
    for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
2866
      colLen[i] = htonl(colLen[i]);
2867
      ASSERT(colLen[i] >= 0);
H
Haojun Liao 已提交
2868

2869
      SColumnInfoData* pColInfoData = taosArrayGet(pRes->pDataBlock, i);
H
Haojun Liao 已提交
2870 2871 2872
      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
        pColInfoData->varmeta.length = colLen[i];
        pColInfoData->varmeta.allocLen = colLen[i];
2873

L
Liu Jicong 已提交
2874 2875
        memcpy(pColInfoData->varmeta.offset, pStart, sizeof(int32_t) * numOfRows);
        pStart += sizeof(int32_t) * numOfRows;
H
Haojun Liao 已提交
2876

2877 2878 2879
        if (colLen[i] > 0) {
          pColInfoData->pData = taosMemoryMalloc(colLen[i]);
        }
H
Haojun Liao 已提交
2880 2881 2882
      } else {
        memcpy(pColInfoData->nullbitmap, pStart, BitmapLen(numOfRows));
        pStart += BitmapLen(numOfRows);
2883
      }
H
Haojun Liao 已提交
2884

2885 2886 2887 2888
      if (colLen[i] > 0) {
        memcpy(pColInfoData->pData, pStart, colLen[i]);
      }

L
Liu Jicong 已提交
2889 2890
      // TODO setting this flag to true temporarily so aggregate function on stable will
      // examine NULL value for non-primary key column
2891
      pColInfoData->hasNull = true;
H
Haojun Liao 已提交
2892
      pStart += colLen[i];
2893
    }
H
Haojun Liao 已提交
2894
  } else {  // extract data according to pColList
2895
    ASSERT(numOfOutput == taosArrayGetSize(pColList));
2896 2897 2898 2899 2900
    char* pStart = pData;

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

2901
    // todo refactor:extract method
2902
    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
dengyihao's avatar
dengyihao 已提交
2903
    for (int32_t i = 0; i < numOfCols; ++i) {
2904 2905 2906 2907 2908 2909 2910
      SSysTableSchema* p = (SSysTableSchema*)pStart;

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

H
Haojun Liao 已提交
2911 2912 2913 2914
    SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
    pBlock->pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData));
    pBlock->info.numOfCols = numOfCols;

dengyihao's avatar
dengyihao 已提交
2915
    for (int32_t i = 0; i < numOfCols; ++i) {
2916 2917 2918 2919 2920
      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
2921
      taosArrayPush(pBlock->pDataBlock, &idata);
2922
      if (IS_VAR_DATA_TYPE(idata.info.type)) {
H
Haojun Liao 已提交
2923
        pBlock->info.hasVarCol = true;
2924 2925 2926
      }
    }

H
Haojun Liao 已提交
2927
    blockDataEnsureCapacity(pBlock, numOfRows);
2928

L
Liu Jicong 已提交
2929 2930
    int32_t  dataLen = *(int32_t*)pStart;
    uint64_t groupId = *(uint64_t*)(pStart + sizeof(int32_t));
2931 2932
    pStart += sizeof(int32_t) + sizeof(uint64_t);

L
Liu Jicong 已提交
2933
    int32_t* colLen = (int32_t*)(pStart);
2934 2935 2936 2937
    pStart += sizeof(int32_t) * numOfCols;

    for (int32_t i = 0; i < numOfCols; ++i) {
      colLen[i] = htonl(colLen[i]);
2938
      ASSERT(colLen[i] >= 0);
2939

H
Haojun Liao 已提交
2940
      SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956
      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
2957 2958

    // data from mnode
H
Haojun Liao 已提交
2959 2960 2961
    relocateColumnData(pRes, pColList, pBlock->pDataBlock);
    taosArrayDestroy(pBlock->pDataBlock);
    taosMemoryFree(pBlock);
X
Xiaoyu Wang 已提交
2962
    //    blockDataDestroy(pBlock);
2963
  }
2964

H
Haojun Liao 已提交
2965
  pRes->info.rows = numOfRows;
2966
  blockDataUpdateTsWindow(pRes);
2967

2968
  int64_t el = taosGetTimestampUs() - startTs;
2969

H
Haojun Liao 已提交
2970 2971
  pLoadInfo->totalRows += numOfRows;
  pLoadInfo->totalSize += compLen;
2972

H
Haojun Liao 已提交
2973 2974 2975
  if (total != NULL) {
    *total += numOfRows;
  }
2976

H
Haojun Liao 已提交
2977
  pLoadInfo->totalElapsed += el;
2978 2979
  return TSDB_CODE_SUCCESS;
}
2980

L
Liu Jicong 已提交
2981 2982
static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
2983
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
2984

L
Liu Jicong 已提交
2985
  int64_t              el = taosGetTimestampUs() - startTs;
H
Haojun Liao 已提交
2986 2987
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
  pLoadInfo->totalElapsed += el;
H
Haojun Liao 已提交
2988

2989
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
L
Liu Jicong 已提交
2990 2991 2992
  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);
2993 2994 2995 2996 2997

  doSetOperatorCompleted(pOperator);
  return NULL;
}

L
Liu Jicong 已提交
2998 2999
static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
                                                   SExecTaskInfo* pTaskInfo) {
3000 3001 3002 3003 3004 3005 3006 3007 3008
  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);

3009
      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
3010
        completed += 1;
H
Haojun Liao 已提交
3011 3012
        continue;
      }
3013

3014
      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
3015 3016 3017
        continue;
      }

L
Liu Jicong 已提交
3018
      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3019
      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
3020

L
Liu Jicong 已提交
3021
      SSDataBlock*         pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3022
      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3023
      if (pRsp->numOfRows == 0) {
L
Liu Jicong 已提交
3024 3025
        qDebug("%s vgId:%d, taskID:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
               " try next",
3026
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i + 1, pDataInfo->totalRows,
H
Haojun Liao 已提交
3027
               pExchangeInfo->loadInfo.totalRows);
3028
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3029 3030 3031
        completed += 1;
        continue;
      }
H
Haojun Liao 已提交
3032

H
Haojun Liao 已提交
3033
      SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3034 3035
      code = setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
                                       pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3036
      if (code != 0) {
3037 3038 3039
        goto _error;
      }

3040 3041 3042
      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
3043 3044
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
               pLoadInfo->totalRows, pLoadInfo->totalSize, i + 1, totalSources);
3045
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3046
      } else {
dengyihao's avatar
dengyihao 已提交
3047 3048
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
               ", totalBytes:%" PRIu64,
H
Haojun Liao 已提交
3049 3050
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
               pLoadInfo->totalSize);
3051 3052
      }

3053 3054
      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073
        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
3074 3075 3076
static SSDataBlock* concurrentlyLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3077

H
Haojun Liao 已提交
3078 3079 3080
  if (pOperator->status == OP_RES_TO_RETURN) {
    return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
  }
3081

L
Liu Jicong 已提交
3082
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3083 3084 3085
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3086
  for (int32_t i = 0; i < totalSources; ++i) {
3087 3088
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3089
      return NULL;
3090 3091 3092 3093
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3094 3095
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3096 3097

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3098 3099
  pOperator->status = OP_RES_TO_RETURN;
  return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
3100 3101
}

L
Liu Jicong 已提交
3102 3103 3104
static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3105

L
Liu Jicong 已提交
3106
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3107 3108 3109
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3110
  for (int32_t i = 0; i < totalSources; ++i) {
3111 3112
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3113 3114
      pTaskInfo->code = code;
      return code;
3115 3116 3117 3118
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3119 3120
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3121 3122

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3123
  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
3124

H
Haojun Liao 已提交
3125
  return TSDB_CODE_SUCCESS;
3126 3127
}

L
Liu Jicong 已提交
3128 3129 3130
static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3131

L
Liu Jicong 已提交
3132
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3133
  int64_t startTs = taosGetTimestampUs();
3134

L
Liu Jicong 已提交
3135
  while (1) {
3136 3137
    if (pExchangeInfo->current >= totalSources) {
      return setAllSourcesCompleted(pOperator, startTs);
3138
    }
3139

3140 3141 3142
    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
    tsem_wait(&pExchangeInfo->ready);

dengyihao's avatar
dengyihao 已提交
3143
    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
X
Xiaoyu Wang 已提交
3144
    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
3145

H
Haojun Liao 已提交
3146
    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
dengyihao's avatar
dengyihao 已提交
3147 3148
      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
3149 3150 3151 3152
      pOperator->pTaskInfo->code = pDataInfo->code;
      return NULL;
    }

L
Liu Jicong 已提交
3153
    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
H
Haojun Liao 已提交
3154
    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3155
    if (pRsp->numOfRows == 0) {
dengyihao's avatar
dengyihao 已提交
3156 3157
      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " %d of total completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
             " try next",
3158
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pExchangeInfo->current + 1,
H
Haojun Liao 已提交
3159
             pDataInfo->totalRows, pLoadInfo->totalRows);
H
Haojun Liao 已提交
3160

3161
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3162 3163 3164
      pExchangeInfo->current += 1;
      continue;
    }
H
Haojun Liao 已提交
3165

L
Liu Jicong 已提交
3166
    SSDataBlock*       pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3167
    SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
L
Liu Jicong 已提交
3168 3169
    int32_t            code =
        setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
3170
                                  pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3171 3172

    if (pRsp->completed == 1) {
H
Haojun Liao 已提交
3173
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
L
Liu Jicong 已提交
3174 3175 3176
             ", 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);
3177

3178
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3179 3180
      pExchangeInfo->current += 1;
    } else {
L
Liu Jicong 已提交
3181 3182 3183 3184
      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);
3185 3186 3187 3188
    }

    return pExchangeInfo->pResult;
  }
3189 3190
}

L
Liu Jicong 已提交
3191
static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
3192
  if (OPTR_IS_OPENED(pOperator)) {
H
Haojun Liao 已提交
3193 3194 3195
    return TSDB_CODE_SUCCESS;
  }

L
Liu Jicong 已提交
3196
  SExchangeInfo* pExchangeInfo = pOperator->info;
H
Haojun Liao 已提交
3197 3198 3199 3200 3201 3202 3203 3204 3205
  if (pExchangeInfo->seqLoadData) {
    // do nothing for sequentially load data
  } else {
    int32_t code = prepareConcurrentlyLoad(pOperator);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
    }
  }

3206
  OPTR_SET_OPENED(pOperator);
H
Haojun Liao 已提交
3207 3208 3209
  return TSDB_CODE_SUCCESS;
}

3210
static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3211 3212
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3213

3214
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
3215
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3216 3217
    return NULL;
  }
3218

L
Liu Jicong 已提交
3219
  size_t               totalSources = taosArrayGetSize(pExchangeInfo->pSources);
H
Haojun Liao 已提交
3220
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
H
Haojun Liao 已提交
3221

3222
  if (pOperator->status == OP_EXEC_DONE) {
L
Liu Jicong 已提交
3223 3224 3225
    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);
3226 3227 3228 3229 3230 3231
    return NULL;
  }

  if (pExchangeInfo->seqLoadData) {
    return seqLoadRemoteData(pOperator);
  } else {
H
Haojun Liao 已提交
3232
    return concurrentlyLoadRemoteData(pOperator);
3233
  }
H
Haojun Liao 已提交
3234

3235
#if 0
H
Haojun Liao 已提交
3236
  _error:
wafwerar's avatar
wafwerar 已提交
3237 3238
  taosMemoryFreeClear(pMsg);
  taosMemoryFreeClear(pMsgSendInfo);
H
Haojun Liao 已提交
3239 3240 3241

  terrno = pTaskInfo->code;
  return NULL;
3242
#endif
H
Haojun Liao 已提交
3243
}
3244

H
Haojun Liao 已提交
3245
static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo) {
3246
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
H
Haojun Liao 已提交
3247 3248
  if (pInfo->pSourceDataInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
3249 3250
  }

L
Liu Jicong 已提交
3251
  for (int32_t i = 0; i < numOfSources; ++i) {
3252
    SSourceDataInfo dataInfo = {0};
H
Haojun Liao 已提交
3253
    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
L
Liu Jicong 已提交
3254 3255
    dataInfo.pEx = pInfo;
    dataInfo.index = i;
3256

H
Haojun Liao 已提交
3257 3258 3259 3260 3261 3262 3263 3264 3265 3266
    void* ret = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
    if (ret == NULL) {
      taosArrayDestroy(pInfo->pSourceDataInfo);
      return TSDB_CODE_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
}

3267
SOperatorInfo* createExchangeOperatorInfo(void *pTransporter, const SNodeList* pSources, SSDataBlock* pBlock,
3268
                                          SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
3269
  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
wafwerar's avatar
wafwerar 已提交
3270
  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
3271 3272

  if (pInfo == NULL || pOperator == NULL) {
wafwerar's avatar
wafwerar 已提交
3273 3274
    taosMemoryFreeClear(pInfo);
    taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3275 3276
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return NULL;
H
Haojun Liao 已提交
3277 3278
  }

H
Haojun Liao 已提交
3279
  size_t numOfSources = LIST_LENGTH(pSources);
H
Haojun Liao 已提交
3280
  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
H
Haojun Liao 已提交
3281 3282 3283
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
  if (pInfo->pSourceDataInfo == NULL || pInfo->pSources == NULL) {
    goto _error;
H
Haojun Liao 已提交
3284 3285
  }

L
Liu Jicong 已提交
3286 3287
  for (int32_t i = 0; i < numOfSources; ++i) {
    SNodeListNode* pNode = nodesListGetNode((SNodeList*)pSources, i);
H
Haojun Liao 已提交
3288 3289
    taosArrayPush(pInfo->pSources, pNode);
  }
3290

H
Haojun Liao 已提交
3291 3292 3293
  int32_t code = initDataSource(numOfSources, pInfo);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3294
  }
H
Haojun Liao 已提交
3295

dengyihao's avatar
dengyihao 已提交
3296
  pInfo->pResult = pBlock;
3297 3298 3299
  pInfo->seqLoadData = true;

  tsem_init(&pInfo->ready, 0, 0);
3300

dengyihao's avatar
dengyihao 已提交
3301
  pOperator->name = "ExchangeOperator";
X
Xiaoyu Wang 已提交
3302
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
3303
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
3304 3305
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3306
  pOperator->numOfExprs = pBlock->info.numOfCols;
dengyihao's avatar
dengyihao 已提交
3307
  pOperator->pTaskInfo = pTaskInfo;
3308

L
Liu Jicong 已提交
3309 3310
  pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                         destroyExchangeOperatorInfo, NULL, NULL, NULL);
3311
  pInfo->pTransporter = pTransporter;
3312

3313
  return pOperator;
H
Haojun Liao 已提交
3314

L
Liu Jicong 已提交
3315
_error:
H
Haojun Liao 已提交
3316
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3317
    destroyExchangeOperatorInfo(pInfo, numOfSources);
H
Haojun Liao 已提交
3318 3319
  }

wafwerar's avatar
wafwerar 已提交
3320 3321
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3322
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
H
Haojun Liao 已提交
3323
  return NULL;
3324 3325
}

dengyihao's avatar
dengyihao 已提交
3326 3327
static int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                                const char* pKey);
3328

3329
static void destroySortedMergeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
3330
  SSortedMergeOperatorInfo* pInfo = (SSortedMergeOperatorInfo*)param;
H
Haojun Liao 已提交
3331
  taosArrayDestroy(pInfo->pSortInfo);
3332 3333 3334
  taosArrayDestroy(pInfo->groupInfo);

  if (pInfo->pSortHandle != NULL) {
H
Haojun Liao 已提交
3335
    tsortDestroySortHandle(pInfo->pSortHandle);
3336 3337
  }

H
Haojun Liao 已提交
3338
  blockDataDestroy(pInfo->binfo.pRes);
H
Haojun Liao 已提交
3339
  cleanupAggSup(&pInfo->aggSup);
3340
}
H
Haojun Liao 已提交
3341

L
Liu Jicong 已提交
3342
static bool needToMerge(SSDataBlock* pBlock, SArray* groupInfo, char** buf, int32_t rowIndex) {
3343 3344 3345 3346
  size_t size = taosArrayGetSize(groupInfo);
  if (size == 0) {
    return true;
  }
3347

3348 3349
  for (int32_t i = 0; i < size; ++i) {
    int32_t* index = taosArrayGet(groupInfo, i);
3350

3351
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
L
Liu Jicong 已提交
3352
    bool             isNull = colDataIsNull(pColInfo, rowIndex, pBlock->info.rows, NULL);
3353

3354 3355 3356
    if ((isNull && buf[i] != NULL) || (!isNull && buf[i] == NULL)) {
      return false;
    }
3357

3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370
    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;
      }
3371 3372 3373
    }
  }

3374
  return 0;
3375 3376
}

L
Liu Jicong 已提交
3377 3378 3379
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
3380
    pCtx[j].startRow = rowIndex;
3381 3382
  }

3383 3384
  for (int32_t j = 0; j < numOfExpr; ++j) {
    int32_t functionId = pCtx[j].functionId;
L
Liu Jicong 已提交
3385 3386 3387 3388 3389 3390 3391 3392 3393
    //    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]);
    //    }
3394
  }
3395
}
3396

L
Liu Jicong 已提交
3397 3398
static void doFinalizeResultImpl(SqlFunctionCtx* pCtx, int32_t numOfExpr) {
  for (int32_t j = 0; j < numOfExpr; ++j) {
3399 3400 3401 3402
    int32_t functionId = pCtx[j].functionId;
    //    if (functionId == FUNC_TAG_DUMMY || functionId == FUNC_TS_DUMMY) {
    //      continue;
    //    }
3403

3404 3405 3406 3407
    //    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
3408
    //    pCtx[j].fpSet.finalize(&pCtx[j]);
3409 3410
  }
}
3411

3412
static bool saveCurrentTuple(char** rowColData, SArray* pColumnList, SSDataBlock* pBlock, int32_t rowIndex) {
L
Liu Jicong 已提交
3413
  int32_t size = (int32_t)taosArrayGetSize(pColumnList);
3414

L
Liu Jicong 已提交
3415 3416
  for (int32_t i = 0; i < size; ++i) {
    int32_t*         index = taosArrayGet(pColumnList, i);
3417
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
H
Haojun Liao 已提交
3418

3419 3420 3421
    char* data = colDataGetData(pColInfo, rowIndex);
    memcpy(rowColData[i], data, colDataGetLength(pColInfo, rowIndex));
  }
3422

3423 3424
  return true;
}
3425

3426 3427
static void doMergeImpl(SOperatorInfo* pOperator, int32_t numOfExpr, SSDataBlock* pBlock) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
3428

3429
  SqlFunctionCtx* pCtx = pInfo->binfo.pCtx;
L
Liu Jicong 已提交
3430
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
3431 3432
    pCtx[i].size = 1;
  }
3433

L
Liu Jicong 已提交
3434
  for (int32_t i = 0; i < pBlock->info.rows; ++i) {
3435 3436 3437 3438 3439 3440 3441 3442 3443
    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);
3444 3445
        int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfExprs, NULL);
        //        setTagValueForMultipleRows(pCtx, pOperator->numOfExprs, numOfRows);
3446

3447
        // TODO check for available buffer;
H
Haojun Liao 已提交
3448

3449 3450 3451 3452 3453
        // next group info data
        pInfo->binfo.pRes->info.rows += numOfRows;
        for (int32_t j = 0; j < numOfExpr; ++j) {
          if (pCtx[j].functionId < 0) {
            continue;
3454
          }
3455

H
Haojun Liao 已提交
3456
          pCtx[j].fpSet.process(&pCtx[j]);
3457
        }
3458 3459 3460

        doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
        pInfo->hasGroupVal = saveCurrentTuple(pInfo->groupVal, pInfo->groupInfo, pBlock, i);
H
Haojun Liao 已提交
3461
      }
3462 3463 3464 3465
    }
  }
}

3466 3467
static SSDataBlock* doMerge(SOperatorInfo* pOperator) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
L
Liu Jicong 已提交
3468
  SSortHandle*              pHandle = pInfo->pSortHandle;
3469

3470
  SSDataBlock* pDataBlock = createOneDataBlock(pInfo->binfo.pRes, false);
3471
  blockDataEnsureCapacity(pDataBlock, pOperator->resultInfo.capacity);
3472

L
Liu Jicong 已提交
3473
  while (1) {
3474
    blockDataCleanup(pDataBlock);
3475
    while (1) {
H
Haojun Liao 已提交
3476
      STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
3477 3478
      if (pTupleHandle == NULL) {
        break;
3479
      }
3480

3481 3482
      // build datablock for merge for one group
      appendOneRowToDataBlock(pDataBlock, pTupleHandle);
3483
      if (pDataBlock->info.rows >= pOperator->resultInfo.capacity) {
3484 3485
        break;
      }
3486
    }
3487

3488 3489 3490
    if (pDataBlock->info.rows == 0) {
      break;
    }
3491

3492
    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pDataBlock, TSDB_ORDER_ASC, true);
L
Liu Jicong 已提交
3493 3494
    //  updateOutputBuf(&pInfo->binfo, &pAggInfo->bufCapacity, pBlock->info.rows * pAggInfo->resultRowFactor,
    //  pOperator->pRuntimeEnv, true);
3495
    doMergeImpl(pOperator, pOperator->numOfExprs, pDataBlock);
3496 3497
    // flush to tuple store, and after all data have been handled, return to upstream node or sink node
  }
3498

3499 3500 3501
  doFinalizeResultImpl(pInfo->binfo.pCtx, pOperator->numOfExprs);
  int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfExprs, NULL);
  //        setTagValueForMultipleRows(pCtx, pOperator->numOfExprs, numOfRows);
3502

3503
  // TODO check for available buffer;
3504

3505 3506
  // next group info data
  pInfo->binfo.pRes->info.rows += numOfRows;
L
Liu Jicong 已提交
3507
  return (pInfo->binfo.pRes->info.rows > 0) ? pInfo->binfo.pRes : NULL;
3508
}
3509

3510
static SSDataBlock* doSortedMerge(SOperatorInfo* pOperator) {
3511 3512
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
3513 3514
  }

L
Liu Jicong 已提交
3515
  SExecTaskInfo*            pTaskInfo = pOperator->pTaskInfo;
3516
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
H
Haojun Liao 已提交
3517
  if (pOperator->status == OP_RES_TO_RETURN) {
3518
    return getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity, NULL);
3519 3520
  }

3521
  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
dengyihao's avatar
dengyihao 已提交
3522 3523
  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, NULL, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize,
                                             numOfBufPage, pInfo->binfo.pRes, "GET_TASKID(pTaskInfo)");
H
Haojun Liao 已提交
3524

3525
  tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock, NULL, NULL);
3526

L
Liu Jicong 已提交
3527
  for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
wmmhello's avatar
wmmhello 已提交
3528
    SSortSource* ps = taosMemoryCalloc(1, sizeof(SSortSource));
H
Haojun Liao 已提交
3529
    ps->param = pOperator->pDownstream[i];
H
Haojun Liao 已提交
3530
    tsortAddSource(pInfo->pSortHandle, ps);
3531 3532
  }

H
Haojun Liao 已提交
3533
  int32_t code = tsortOpen(pInfo->pSortHandle);
3534
  if (code != TSDB_CODE_SUCCESS) {
3535
    longjmp(pTaskInfo->env, terrno);
3536 3537
  }

H
Haojun Liao 已提交
3538
  pOperator->status = OP_RES_TO_RETURN;
3539
  return doMerge(pOperator);
3540
}
3541

L
Liu Jicong 已提交
3542 3543
static int32_t initGroupCol(SExprInfo* pExprInfo, int32_t numOfCols, SArray* pGroupInfo,
                            SSortedMergeOperatorInfo* pInfo) {
3544 3545
  if (pGroupInfo == NULL || taosArrayGetSize(pGroupInfo) == 0) {
    return 0;
H
Haojun Liao 已提交
3546 3547
  }

3548 3549 3550 3551 3552 3553 3554 3555
  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
3556 3557
  size_t numOfGroupCol = taosArrayGetSize(pInfo->groupInfo);
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3558
    SColumn* pCol = taosArrayGet(pGroupInfo, i);
L
Liu Jicong 已提交
3559
    for (int32_t j = 0; j < numOfCols; ++j) {
H
Haojun Liao 已提交
3560
      SExprInfo* pe = &pExprInfo[j];
3561
      if (pe->base.resSchema.slotId == pCol->colId) {
3562 3563
        taosArrayPush(plist, pCol);
        taosArrayPush(pInfo->groupInfo, &j);
H
Haojun Liao 已提交
3564
        len += pCol->bytes;
3565 3566
        break;
      }
H
Haojun Liao 已提交
3567 3568 3569
    }
  }

3570
  ASSERT(taosArrayGetSize(pGroupInfo) == taosArrayGetSize(plist));
H
Haojun Liao 已提交
3571

wafwerar's avatar
wafwerar 已提交
3572
  pInfo->groupVal = taosMemoryCalloc(1, (POINTER_BYTES * numOfGroupCol + len));
3573 3574 3575 3576
  if (pInfo->groupVal == NULL) {
    taosArrayDestroy(plist);
    return TSDB_CODE_OUT_OF_MEMORY;
  }
H
Haojun Liao 已提交
3577

3578
  int32_t offset = 0;
L
Liu Jicong 已提交
3579 3580
  char*   start = (char*)(pInfo->groupVal + (POINTER_BYTES * numOfGroupCol));
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3581 3582
    pInfo->groupVal[i] = start + offset;
    SColumn* pCol = taosArrayGet(plist, i);
H
Haojun Liao 已提交
3583
    offset += pCol->bytes;
3584
  }
H
Haojun Liao 已提交
3585

3586
  taosArrayDestroy(plist);
H
Haojun Liao 已提交
3587

3588 3589
  return TSDB_CODE_SUCCESS;
}
H
Haojun Liao 已提交
3590

L
Liu Jicong 已提交
3591 3592 3593
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
3594
  SSortedMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortedMergeOperatorInfo));
L
Liu Jicong 已提交
3595
  SOperatorInfo*            pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
3596
  if (pInfo == NULL || pOperator == NULL) {
3597
    goto _error;
3598
  }
H
Haojun Liao 已提交
3599

3600
  pInfo->binfo.pCtx = createSqlFunctionCtx(pExprInfo, num, &pInfo->binfo.rowCellInfoOffset);
3601
  initResultRowInfo(&pInfo->binfo.resultRowInfo, (int32_t)1);
H
Haojun Liao 已提交
3602

3603 3604 3605
  if (pInfo->binfo.pCtx == NULL || pInfo->binfo.pRes == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
3606

dengyihao's avatar
dengyihao 已提交
3607
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
3608
  int32_t code = doInitAggInfoSup(&pInfo->aggSup, pInfo->binfo.pCtx, num, keyBufSize, pTaskInfo->id.str);
3609 3610 3611
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3612

H
Haojun Liao 已提交
3613
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, num, pTaskInfo);
H
Haojun Liao 已提交
3614
  code = initGroupCol(pExprInfo, num, pGroupInfo, pInfo);
3615 3616 3617
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3618

L
Liu Jicong 已提交
3619 3620 3621 3622 3623
  //  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
3624

3625
  pOperator->resultInfo.capacity = blockDataGetCapacityInRow(pInfo->binfo.pRes, pInfo->bufPageSize);
H
Haojun Liao 已提交
3626

L
Liu Jicong 已提交
3627
  pOperator->name = "SortedMerge";
X
Xiaoyu Wang 已提交
3628
  // pOperator->operatorType = OP_SortedMerge;
3629
  pOperator->blocking = true;
L
Liu Jicong 已提交
3630 3631
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3632
  pOperator->numOfExprs = num;
L
Liu Jicong 已提交
3633
  pOperator->pExpr = pExprInfo;
H
Haojun Liao 已提交
3634

L
Liu Jicong 已提交
3635
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
3636

3637 3638
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSortedMerge, NULL, NULL, destroySortedMergeOperatorInfo,
                                         NULL, NULL, NULL);
3639 3640 3641
  code = appendDownstream(pOperator, downstream, numOfDownstream);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3642
  }
H
Haojun Liao 已提交
3643

3644
  return pOperator;
H
Haojun Liao 已提交
3645

L
Liu Jicong 已提交
3646
_error:
3647
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3648
    destroySortedMergeOperatorInfo(pInfo, num);
H
Haojun Liao 已提交
3649 3650
  }

wafwerar's avatar
wafwerar 已提交
3651 3652
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
3653 3654
  terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
  return NULL;
H
Haojun Liao 已提交
3655 3656
}

3657 3658
int32_t getTableScanOrder(SOperatorInfo* pOperator) {
  if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
H
Haojun Liao 已提交
3659
    if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
3660
      return TSDB_ORDER_ASC;
H
Haojun Liao 已提交
3661 3662
    } else {
      return getTableScanOrder(pOperator->pDownstream[0]);
3663 3664 3665 3666 3667 3668
    }
  }

  STableScanInfo* pTableScanInfo = pOperator->info;
  return pTableScanInfo->cond.order;
}
3669 3670

// this is a blocking operator
L
Liu Jicong 已提交
3671
static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
H
Haojun Liao 已提交
3672 3673
  if (OPTR_IS_OPENED(pOperator)) {
    return TSDB_CODE_SUCCESS;
3674 3675
  }

H
Haojun Liao 已提交
3676
  SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
3677
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3678

dengyihao's avatar
dengyihao 已提交
3679
  SOptrBasicInfo* pInfo = &pAggInfo->binfo;
3680

H
Haojun Liao 已提交
3681
  SOperatorInfo* downstream = pOperator->pDownstream[0];
3682

H
Haojun Liao 已提交
3683
  while (1) {
H
Haojun Liao 已提交
3684
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
3685
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
3686
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
3687 3688 3689 3690

    if (pBlock == NULL) {
      break;
    }
H
Haojun Liao 已提交
3691
    //    if (pAggInfo->current != NULL) {
3692
    //      setTagValue(pOperator, pAggInfo->current->pTable, pInfo->pCtx, pOperator->numOfExprs);
H
Haojun Liao 已提交
3693
    //    }
3694

3695 3696
    int32_t order = getTableScanOrder(pOperator);

3697 3698
    // 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
3699 3700
      int32_t code = projectApplyFunctions(pAggInfo->pScalarExprInfo, pBlock, pBlock, pAggInfo->pScalarCtx,
                                           pAggInfo->numOfScalarExpr, NULL);
3701 3702 3703 3704
      if (code != TSDB_CODE_SUCCESS) {
        pTaskInfo->code = code;
        longjmp(pTaskInfo->env, pTaskInfo->code);
      }
3705 3706
    }

3707
    // the pDataBlock are always the same one, no need to call this again
3708
    setExecutionContext(pOperator->numOfExprs, pBlock->info.groupId, pTaskInfo, pAggInfo);
3709
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, true);
H
Haojun Liao 已提交
3710
    doAggregateImpl(pOperator, 0, pInfo->pCtx);
3711

dengyihao's avatar
dengyihao 已提交
3712
#if 0  // test for encode/decode result info
3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723
    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);
      }
3724
    }
3725
#endif
3726 3727
  }

H
Haojun Liao 已提交
3728
  closeAllResultRows(&pAggInfo->binfo.resultRowInfo);
3729 3730
  finalizeMultiTupleQueryResult(pOperator->numOfExprs, pAggInfo->aggSup.pResultBuf, &pAggInfo->binfo.resultRowInfo,
                                pAggInfo->binfo.rowCellInfoOffset);
H
Haojun Liao 已提交
3731

3732
  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, false);
H
Haojun Liao 已提交
3733 3734 3735 3736
  OPTR_SET_OPENED(pOperator);
  return TSDB_CODE_SUCCESS;
}

3737
static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3738
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3739 3740 3741 3742 3743 3744
  SOptrBasicInfo*   pInfo = &pAggInfo->binfo;

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

L
Liu Jicong 已提交
3745
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3746
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
H
Haojun Liao 已提交
3747 3748 3749 3750
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
    return NULL;
  }

H
Haojun Liao 已提交
3751
  blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
3752
  doBuildResultDatablock(pInfo, &pAggInfo->groupResInfo, pOperator->pExpr, pAggInfo->aggSup.pResultBuf);
H
Haojun Liao 已提交
3753 3754 3755
  if (pInfo->pRes->info.rows == 0 || !hasRemainDataInCurrentGroup(&pAggInfo->groupResInfo)) {
    doSetOperatorCompleted(pOperator);
  }
3756

H
Haojun Liao 已提交
3757
  doSetOperatorCompleted(pOperator);
L
Liu Jicong 已提交
3758
  return (blockDataGetNumOfRows(pInfo->pRes) != 0) ? pInfo->pRes : NULL;
3759 3760
}

dengyihao's avatar
dengyihao 已提交
3761 3762
void aggEncodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char** result,
                        int32_t* length) {
wmmhello's avatar
wmmhello 已提交
3763
  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
3764
  size_t  keyLen = sizeof(uint64_t) * 2;  // estimate the key length
wmmhello's avatar
wmmhello 已提交
3765
  int32_t totalSize = sizeof(int32_t) + size * (sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize);
wafwerar's avatar
wafwerar 已提交
3766
  *result = taosMemoryCalloc(1, totalSize);
L
Liu Jicong 已提交
3767
  if (*result == NULL) {
3768
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
wmmhello's avatar
wmmhello 已提交
3769 3770 3771
  }
  *(int32_t*)(*result) = size;
  int32_t offset = sizeof(int32_t);
3772 3773

  // prepare memory
3774
  SResultRowPosition* pos = &pInfo->resultRowInfo.cur;
dengyihao's avatar
dengyihao 已提交
3775 3776
  void*               pPage = getBufPage(pSup->pResultBuf, pos->pageId);
  SResultRow*         pRow = (SResultRow*)((char*)pPage + pos->offset);
3777 3778 3779
  setBufPageDirty(pPage, true);
  releaseBufPage(pSup->pResultBuf, pPage);

dengyihao's avatar
dengyihao 已提交
3780
  void* pIter = taosHashIterate(pSup->pResultRowHashTable, NULL);
wmmhello's avatar
wmmhello 已提交
3781
  while (pIter) {
dengyihao's avatar
dengyihao 已提交
3782
    void*               key = taosHashGetKey(pIter, &keyLen);
3783
    SResultRowPosition* p1 = (SResultRowPosition*)pIter;
3784

dengyihao's avatar
dengyihao 已提交
3785
    pPage = (SFilePage*)getBufPage(pSup->pResultBuf, p1->pageId);
3786
    pRow = (SResultRow*)((char*)pPage + p1->offset);
3787 3788
    setBufPageDirty(pPage, true);
    releaseBufPage(pSup->pResultBuf, pPage);
wmmhello's avatar
wmmhello 已提交
3789 3790 3791

    // recalculate the result size
    int32_t realTotalSize = offset + sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize;
L
Liu Jicong 已提交
3792 3793 3794
    if (realTotalSize > totalSize) {
      char* tmp = taosMemoryRealloc(*result, realTotalSize);
      if (tmp == NULL) {
wmmhello's avatar
wmmhello 已提交
3795
        terrno = TSDB_CODE_OUT_OF_MEMORY;
wafwerar's avatar
wafwerar 已提交
3796
        taosMemoryFree(*result);
wmmhello's avatar
wmmhello 已提交
3797
        *result = NULL;
3798
        longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
L
Liu Jicong 已提交
3799
      } else {
wmmhello's avatar
wmmhello 已提交
3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811
        *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);
3812
    memcpy(*result + offset, pRow, pSup->resultRowSize);
wmmhello's avatar
wmmhello 已提交
3813 3814 3815 3816 3817
    offset += pSup->resultRowSize;

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

L
Liu Jicong 已提交
3818
  if (length) {
wmmhello's avatar
wmmhello 已提交
3819 3820 3821 3822 3823
    *length = offset;
  }
  return;
}

dengyihao's avatar
dengyihao 已提交
3824 3825
bool aggDecodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char* result,
                        int32_t length) {
L
Liu Jicong 已提交
3826
  if (!result || length <= 0) {
wmmhello's avatar
wmmhello 已提交
3827 3828 3829 3830 3831 3832 3833
    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
3834
  while (count-- > 0 && length > offset) {
wmmhello's avatar
wmmhello 已提交
3835 3836 3837
    int32_t keyLen = *(int32_t*)(result + offset);
    offset += sizeof(int32_t);

L
Liu Jicong 已提交
3838 3839 3840
    uint64_t    tableGroupId = *(uint64_t*)(result + offset);
    SResultRow* resultRow = getNewResultRow_rv(pSup->pResultBuf, tableGroupId, pSup->resultRowSize);
    if (!resultRow) {
3841
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3842
    }
3843

wmmhello's avatar
wmmhello 已提交
3844
    // add a new result set for a new group
3845 3846
    SResultRowPosition pos = {.pageId = resultRow->pageId, .offset = resultRow->offset};
    taosHashPut(pSup->pResultRowHashTable, result + offset, keyLen, &pos, sizeof(SResultRowPosition));
wmmhello's avatar
wmmhello 已提交
3847 3848 3849

    offset += keyLen;
    int32_t valueLen = *(int32_t*)(result + offset);
L
Liu Jicong 已提交
3850
    if (valueLen != pSup->resultRowSize) {
3851
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3852 3853 3854 3855 3856 3857 3858 3859 3860 3861
    }
    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);
3862
    prepareResultListBuffer(&pInfo->resultRowInfo, pOperator->pTaskInfo->env);
dengyihao's avatar
dengyihao 已提交
3863
    //    pInfo->resultRowInfo.cur = pInfo->resultRowInfo.size;
L
Liu Jicong 已提交
3864 3865
    //    pInfo->resultRowInfo.pPosition[pInfo->resultRowInfo.size++] =
    //        (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
dengyihao's avatar
dengyihao 已提交
3866
    pInfo->resultRowInfo.cur = (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
wmmhello's avatar
wmmhello 已提交
3867 3868
  }

L
Liu Jicong 已提交
3869
  if (offset != length) {
3870
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3871 3872 3873 3874
  }
  return true;
}

3875 3876
enum {
  PROJECT_RETRIEVE_CONTINUE = 0x1,
L
Liu Jicong 已提交
3877
  PROJECT_RETRIEVE_DONE = 0x2,
3878 3879 3880 3881 3882
};

static int32_t handleLimitOffset(SOperatorInfo* pOperator, SSDataBlock* pBlock) {
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
L
Liu Jicong 已提交
3883
  SSDataBlock*          pRes = pInfo->pRes;
3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 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

  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;
  }

3932 3933 3934
  // 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);
3935 3936 3937 3938 3939

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

3940
    return PROJECT_RETRIEVE_DONE;
3941
  }
3942

3943
  // todo optimize performance
3944 3945
  // 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
3946 3947
  if (pRes->info.rows >= pOperator->resultInfo.threshold || pProjectInfo->slimit.offset != -1 ||
      pProjectInfo->slimit.limit != -1) {
3948
    return PROJECT_RETRIEVE_DONE;
L
Liu Jicong 已提交
3949
  } else {  // not full enough, continue to accumulate the output data in the buffer.
3950 3951 3952 3953
    return PROJECT_RETRIEVE_CONTINUE;
  }
}

3954
static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
3955
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
L
Liu Jicong 已提交
3956
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
3957 3958

  SSDataBlock* pRes = pInfo->pRes;
3959
  blockDataCleanup(pRes);
3960

3961
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3962 3963 3964
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }
dengyihao's avatar
dengyihao 已提交
3965

H
Haojun Liao 已提交
3966
#if 0
3967 3968 3969 3970 3971 3972
  if (pProjectInfo->existDataBlock) {  // TODO refactor
    SSDataBlock* pBlock = pProjectInfo->existDataBlock;
    pProjectInfo->existDataBlock = NULL;
    *newgroup = true;

    // todo dynamic set tags
L
Liu Jicong 已提交
3973
    //    if (pTableQueryInfo != NULL) {
3974
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
3975
    //    }
3976 3977

    // the pDataBlock are always the same one, no need to call this again
H
Haojun Liao 已提交
3978
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, TSDB_ORDER_ASC);
3979

H
Haojun Liao 已提交
3980
    blockDataEnsureCapacity(pInfo->pRes, pBlock->info.rows);
3981
    projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
3982
    if (pRes->info.rows >= pProjectInfo->binfo.capacity * 0.8) {
3983 3984
      copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfExprs);
      resetResultRowEntryResult(pInfo->pCtx, pOperator->numOfExprs);
3985 3986 3987
      return pRes;
    }
  }
H
Haojun Liao 已提交
3988
#endif
3989

H
Haojun Liao 已提交
3990 3991
  SOperatorInfo* downstream = pOperator->pDownstream[0];

L
Liu Jicong 已提交
3992
  while (1) {
H
Haojun Liao 已提交
3993
    // The downstream exec may change the value of the newgroup, so use a local variable instead.
H
Haojun Liao 已提交
3994
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
3995
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
3996
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
3997 3998

    if (pBlock == NULL) {
3999
      setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
4000 4001 4002 4003
      break;
    }

    // Return result of the previous group in the firstly.
4004
    if (false) {
4005 4006 4007
      if (pRes->info.rows > 0) {
        pProjectInfo->existDataBlock = pBlock;
        break;
L
Liu Jicong 已提交
4008
      } else {  // init output buffer for a new group data
4009
        initCtxOutputBuffer(pInfo->pCtx, pOperator->numOfExprs);
4010 4011 4012
      }
    }

4013 4014
    // todo set tags

H
Haojun Liao 已提交
4015 4016
    //    STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
    //    if (pTableQueryInfo != NULL) {
4017
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfExprs);
H
Haojun Liao 已提交
4018
    //    }
4019 4020

    // the pDataBlock are always the same one, no need to call this again
4021 4022 4023
    int32_t order = getTableScanOrder(pOperator->pDownstream[0]);

    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, false);
4024 4025
    blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);

4026
    pTaskInfo->code = projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs,
dengyihao's avatar
dengyihao 已提交
4027
                          pProjectInfo->pPseudoColInfo);
4028 4029 4030 4031
    if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
      longjmp(pTaskInfo->env, pTaskInfo->code);
    }

4032 4033
    int32_t status = handleLimitOffset(pOperator, pBlock);
    if (status == PROJECT_RETRIEVE_CONTINUE) {
H
Haojun Liao 已提交
4034
      continue;
L
Liu Jicong 已提交
4035
    } else if (status == PROJECT_RETRIEVE_DONE) {
4036 4037 4038
      break;
    }
  }
dengyihao's avatar
dengyihao 已提交
4039

H
Haojun Liao 已提交
4040
  pProjectInfo->curOutput += pInfo->pRes->info.rows;
H
Haojun Liao 已提交
4041

4042
  //  copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
4043
  return (pInfo->pRes->info.rows > 0) ? pInfo->pRes : NULL;
4044 4045
}

L
Liu Jicong 已提交
4046 4047
static void doHandleRemainBlockForNewGroupImpl(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                               SExecTaskInfo* pTaskInfo) {
4048
  pInfo->totalInputRows = pInfo->existNewGroupBlock->info.rows;
H
Haojun Liao 已提交
4049

L
Liu Jicong 已提交
4050 4051
  int64_t ekey = Q_STATUS_EQUAL(pTaskInfo->status, TASK_COMPLETED) ? pTaskInfo->window.ekey
                                                                   : pInfo->existNewGroupBlock->info.window.ekey;
4052 4053
  taosResetFillInfo(pInfo->pFillInfo, getFillInfoStart(pInfo->pFillInfo));

4054
  taosFillSetStartInfo(pInfo->pFillInfo, pInfo->existNewGroupBlock->info.rows, ekey);
4055 4056
  taosFillSetInputDataBlock(pInfo->pFillInfo, pInfo->existNewGroupBlock);

4057
  doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, pResultInfo->capacity);
4058 4059 4060 4061
  pInfo->existNewGroupBlock = NULL;
  *newgroup = true;
}

L
Liu Jicong 已提交
4062 4063
static void doHandleRemainBlockFromNewGroup(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                            SExecTaskInfo* pTaskInfo) {
4064 4065
  if (taosFillHasMoreResults(pInfo->pFillInfo)) {
    *newgroup = false;
4066
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, (int32_t)pResultInfo->capacity);
H
Haojun Liao 已提交
4067
    if (pInfo->pRes->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult)) {
4068 4069 4070 4071 4072 4073
      return;
    }
  }

  // handle the cached new group data block
  if (pInfo->existNewGroupBlock) {
4074
    doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4075 4076 4077
  }
}

4078
static SSDataBlock* doFill(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
4079 4080
  SFillOperatorInfo* pInfo = pOperator->info;
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
4081

H
Haojun Liao 已提交
4082
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4083 4084 4085
  SSDataBlock* pResBlock = pInfo->pRes;

  blockDataCleanup(pResBlock);
4086 4087 4088 4089
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }

4090
  // todo handle different group data interpolation
X
Xiaoyu Wang 已提交
4091 4092
  bool  n = false;
  bool* newgroup = &n;
4093
  doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4094 4095
  if (pResBlock->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult && pResBlock->info.rows > 0)) {
    return pResBlock;
H
Haojun Liao 已提交
4096
  }
4097

H
Haojun Liao 已提交
4098
  SOperatorInfo* pDownstream = pOperator->pDownstream[0];
L
Liu Jicong 已提交
4099
  while (1) {
H
Haojun Liao 已提交
4100
    publishOperatorProfEvent(pDownstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4101
    SSDataBlock* pBlock = pDownstream->fpSet.getNextFn(pDownstream);
H
Haojun Liao 已提交
4102
    publishOperatorProfEvent(pDownstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113

    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
4114
      taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4115 4116 4117 4118 4119 4120 4121
    } else {
      if (pBlock == NULL) {
        if (pInfo->totalInputRows == 0) {
          pOperator->status = OP_EXEC_DONE;
          return NULL;
        }

4122
        taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4123 4124 4125 4126 4127 4128 4129
      } else {
        pInfo->totalInputRows += pBlock->info.rows;
        taosFillSetStartInfo(pInfo->pFillInfo, pBlock->info.rows, pBlock->info.window.ekey);
        taosFillSetInputDataBlock(pInfo->pFillInfo, pBlock);
      }
    }

4130 4131
    blockDataEnsureCapacity(pResBlock, pOperator->resultInfo.capacity);
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pResBlock, pOperator->resultInfo.capacity);
4132 4133

    // current group has no more result to return
4134
    if (pResBlock->info.rows > 0) {
4135 4136
      // 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
4137 4138
      if (pResBlock->info.rows > pResultInfo->threshold || pBlock == NULL || (!pInfo->multigroupResult)) {
        return pResBlock;
4139 4140
      }

4141
      doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4142 4143
      if (pResBlock->info.rows > pOperator->resultInfo.threshold || pBlock == NULL) {
        return pResBlock;
4144 4145 4146
      }
    } else if (pInfo->existNewGroupBlock) {  // try next group
      assert(pBlock != NULL);
4147
      doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4148 4149
      if (pResBlock->info.rows > pResultInfo->threshold) {
        return pResBlock;
4150 4151 4152 4153 4154 4155 4156 4157
      }
    } else {
      return NULL;
    }
  }
}

// todo set the attribute of query scan count
H
Haojun Liao 已提交
4158
static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr) {
L
Liu Jicong 已提交
4159
  for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173
    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;
  }

4174
  if (pOperator->fpSet.closeFn != NULL) {
4175
    pOperator->fpSet.closeFn(pOperator->info, pOperator->numOfExprs);
4176 4177
  }

H
Haojun Liao 已提交
4178
  if (pOperator->pDownstream != NULL) {
L
Liu Jicong 已提交
4179
    for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
H
Haojun Liao 已提交
4180
      destroyOperatorInfo(pOperator->pDownstream[i]);
4181 4182
    }

wafwerar's avatar
wafwerar 已提交
4183
    taosMemoryFreeClear(pOperator->pDownstream);
H
Haojun Liao 已提交
4184
    pOperator->numOfDownstream = 0;
4185 4186
  }

H
Haojun Liao 已提交
4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197
  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
4198
  taosMemoryFreeClear(pOperator->pExpr);
wafwerar's avatar
wafwerar 已提交
4199 4200
  taosMemoryFreeClear(pOperator->info);
  taosMemoryFreeClear(pOperator);
4201 4202
}

dengyihao's avatar
dengyihao 已提交
4203 4204
int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                         const char* pKey) {
4205 4206
  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);

dengyihao's avatar
dengyihao 已提交
4207 4208
  pAggSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
  pAggSup->keyBuf = taosMemoryCalloc(1, keyBufSize + POINTER_BYTES + sizeof(int64_t));
4209 4210
  pAggSup->pResultRowHashTable = taosHashInit(10, hashFn, true, HASH_NO_LOCK);

H
Haojun Liao 已提交
4211
  if (pAggSup->keyBuf == NULL || pAggSup->pResultRowHashTable == NULL) {
4212 4213 4214
    return TSDB_CODE_OUT_OF_MEMORY;
  }

H
Haojun Liao 已提交
4215
  uint32_t defaultPgsz = 4096;
X
Xiaoyu Wang 已提交
4216
  while (defaultPgsz < pAggSup->resultRowSize * 4) {
H
Haojun Liao 已提交
4217 4218 4219 4220 4221 4222 4223 4224 4225 4226
    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
4227 4228 4229 4230
  if (code != TSDB_CODE_SUCCESS) {
    return code;
  }

4231 4232 4233
  return TSDB_CODE_SUCCESS;
}

4234
void cleanupAggSup(SAggSupporter* pAggSup) {
wafwerar's avatar
wafwerar 已提交
4235
  taosMemoryFreeClear(pAggSup->keyBuf);
4236
  taosHashCleanup(pAggSup->pResultRowHashTable);
H
Haojun Liao 已提交
4237
  destroyDiskbasedBuf(pAggSup->pResultBuf);
4238 4239
}

H
Haojun Liao 已提交
4240
int32_t initAggInfo(SOptrBasicInfo* pBasicInfo, SAggSupporter* pAggSup, SExprInfo* pExprInfo, int32_t numOfCols,
4241
                    SSDataBlock* pResultBlock, size_t keyBufSize, const char* pkey) {
4242
  pBasicInfo->pCtx = createSqlFunctionCtx(pExprInfo, numOfCols, &pBasicInfo->rowCellInfoOffset);
H
Haojun Liao 已提交
4243 4244
  pBasicInfo->pRes = pResultBlock;

4245
  doInitAggInfoSup(pAggSup, pBasicInfo->pCtx, numOfCols, keyBufSize, pkey);
4246

L
Liu Jicong 已提交
4247
  for (int32_t i = 0; i < numOfCols; ++i) {
4248 4249 4250
    pBasicInfo->pCtx[i].pBuf = pAggSup->pResultBuf;
  }

4251
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
4252 4253
}

4254 4255 4256 4257 4258 4259 4260 4261 4262
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
4263
static STableQueryInfo* initTableQueryInfo(const STableGroupInfo* pTableGroupInfo) {
L
Liu Jicong 已提交
4264 4265 4266 4267
  if (pTableGroupInfo->numOfTables == 0) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
4268
  STableQueryInfo* pTableQueryInfo = taosMemoryCalloc(pTableGroupInfo->numOfTables, sizeof(STableQueryInfo));
H
Haojun Liao 已提交
4269 4270
  if (pTableQueryInfo == NULL) {
    return NULL;
H
Haojun Liao 已提交
4271
  }
H
Haojun Liao 已提交
4272 4273

  int32_t index = 0;
L
Liu Jicong 已提交
4274
  for (int32_t i = 0; i < taosArrayGetSize(pTableGroupInfo->pGroupList); ++i) {
H
Haojun Liao 已提交
4275
    SArray* pa = taosArrayGetP(pTableGroupInfo->pGroupList, i);
L
Liu Jicong 已提交
4276
    for (int32_t j = 0; j < taosArrayGetSize(pa); ++j) {
H
Haojun Liao 已提交
4277
      STableKeyInfo* pk = taosArrayGet(pa, j);
H
Haojun Liao 已提交
4278
      STableQueryInfo* pTQueryInfo = &pTableQueryInfo[index++];
L
Liu Jicong 已提交
4279
      pTQueryInfo->lastKey = pk->lastKey;
H
Haojun Liao 已提交
4280 4281
    }
  }
H
Haojun Liao 已提交
4282 4283

  STimeWindow win = {0, INT64_MAX};
4284
  createTableQueryInfo(pTableQueryInfo, win);
H
Haojun Liao 已提交
4285
  return pTableQueryInfo;
H
Haojun Liao 已提交
4286 4287
}

L
Liu Jicong 已提交
4288
SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
4289
                                           SSDataBlock* pResultBlock, SExprInfo* pScalarExprInfo,
L
Liu Jicong 已提交
4290 4291
                                           int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo,
                                           const STableGroupInfo* pTableGroupInfo) {
wafwerar's avatar
wafwerar 已提交
4292
  SAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SAggOperatorInfo));
L
Liu Jicong 已提交
4293
  SOperatorInfo*    pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4294 4295 4296
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
4297

4298
  int32_t numOfRows = 10;
dengyihao's avatar
dengyihao 已提交
4299
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4300 4301

  initResultSizeInfo(pOperator, numOfRows);
dengyihao's avatar
dengyihao 已提交
4302 4303
  int32_t code =
      initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResultBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4304
  pInfo->pTableQueryInfo = initTableQueryInfo(pTableGroupInfo);
L
Liu Jicong 已提交
4305
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4306 4307
    goto _error;
  }
H
Haojun Liao 已提交
4308

H
Haojun Liao 已提交
4309 4310 4311 4312
  int32_t numOfGroup = 10;  // todo replaced with true value
  pInfo->groupId = INT32_MIN;
  initResultRowInfo(&pInfo->binfo.resultRowInfo, numOfGroup);

4313 4314
  pInfo->pScalarExprInfo = pScalarExprInfo;
  pInfo->numOfScalarExpr = numOfScalarExpr;
4315 4316 4317
  if (pInfo->pScalarExprInfo != NULL) {
    pInfo->pScalarCtx = createSqlFunctionCtx(pScalarExprInfo, numOfCols, &pInfo->rowCellInfoOffset);
  }
4318

dengyihao's avatar
dengyihao 已提交
4319
  pOperator->name = "TableAggregate";
X
Xiaoyu Wang 已提交
4320
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_AGG;
4321
  pOperator->blocking = true;
dengyihao's avatar
dengyihao 已提交
4322 4323 4324
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
4325
  pOperator->numOfExprs = numOfCols;
dengyihao's avatar
dengyihao 已提交
4326
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
4327

4328 4329
  pOperator->fpSet = createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, NULL, destroyAggOperatorInfo,
                                         aggEncodeResultRow, aggDecodeResultRow, NULL);
H
Haojun Liao 已提交
4330 4331 4332 4333 4334

  code = appendDownstream(pOperator, &downstream, 1);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4335 4336

  return pOperator;
L
Liu Jicong 已提交
4337
_error:
H
Haojun Liao 已提交
4338
  destroyAggOperatorInfo(pInfo, numOfCols);
wafwerar's avatar
wafwerar 已提交
4339 4340
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
4341 4342
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4343 4344
}

H
Haojun Liao 已提交
4345
void doDestroyBasicInfo(SOptrBasicInfo* pInfo, int32_t numOfOutput) {
4346 4347
  assert(pInfo != NULL);

4348
  destroySqlFunctionCtx(pInfo->pCtx, numOfOutput);
wafwerar's avatar
wafwerar 已提交
4349
  taosMemoryFreeClear(pInfo->rowCellInfoOffset);
4350 4351

  cleanupResultRowInfo(&pInfo->resultRowInfo);
H
Haojun Liao 已提交
4352
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
4353 4354
}

H
Haojun Liao 已提交
4355
void destroyBasicOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4356
  SOptrBasicInfo* pInfo = (SOptrBasicInfo*)param;
4357 4358
  doDestroyBasicInfo(pInfo, numOfOutput);
}
H
Haojun Liao 已提交
4359

H
Haojun Liao 已提交
4360
void destroyMergeJoinOperator(void* param, int32_t numOfOutput) {
X
Xiaoyu Wang 已提交
4361
  SJoinOperatorInfo* pJoinOperator = (SJoinOperatorInfo*)param;
H
Haojun Liao 已提交
4362 4363
}

H
Haojun Liao 已提交
4364
void destroyAggOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4365
  SAggOperatorInfo* pInfo = (SAggOperatorInfo*)param;
4366 4367
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
}
4368

H
Haojun Liao 已提交
4369
void destroySFillOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4370
  SFillOperatorInfo* pInfo = (SFillOperatorInfo*)param;
4371
  pInfo->pFillInfo = taosDestroyFillInfo(pInfo->pFillInfo);
H
Haojun Liao 已提交
4372
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
wafwerar's avatar
wafwerar 已提交
4373
  taosMemoryFreeClear(pInfo->p);
4374 4375
}

H
Haojun Liao 已提交
4376
static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
D
fix bug  
dapan 已提交
4377 4378 4379
  if (NULL == param) {
    return;
  }
L
Liu Jicong 已提交
4380
  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
4381
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
H
Haojun Liao 已提交
4382
  cleanupAggSup(&pInfo->aggSup);
H
Haojun Liao 已提交
4383
  taosArrayDestroy(pInfo->pPseudoColInfo);
4384 4385
}

H
Haojun Liao 已提交
4386
void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4387
  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
H
Haojun Liao 已提交
4388 4389 4390 4391 4392 4393 4394 4395 4396
  taosArrayDestroy(pExInfo->pSources);
  taosArrayDestroy(pExInfo->pSourceDataInfo);
  if (pExInfo->pResult != NULL) {
    blockDataDestroy(pExInfo->pResult);
  }

  tsem_destroy(&pExInfo->ready);
}

H
Haojun Liao 已提交
4397 4398
static SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
  SArray* pList = taosArrayInit(4, sizeof(int32_t));
dengyihao's avatar
dengyihao 已提交
4399
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
4400 4401 4402 4403 4404 4405 4406 4407
    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
      taosArrayPush(pList, &i);
    }
  }

  return pList;
}

L
Liu Jicong 已提交
4408
SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t num,
dengyihao's avatar
dengyihao 已提交
4409 4410
                                         SSDataBlock* pResBlock, SLimit* pLimit, SLimit* pSlimit,
                                         SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4411
  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
L
Liu Jicong 已提交
4412
  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4413 4414 4415
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
4416

dengyihao's avatar
dengyihao 已提交
4417 4418 4419
  pInfo->limit = *pLimit;
  pInfo->slimit = *pSlimit;
  pInfo->curOffset = pLimit->offset;
H
Haojun Liao 已提交
4420 4421
  pInfo->curSOffset = pSlimit->offset;

H
Haojun Liao 已提交
4422
  pInfo->binfo.pRes = pResBlock;
H
Haojun Liao 已提交
4423 4424 4425

  int32_t numOfCols = num;
  int32_t numOfRows = 4096;
dengyihao's avatar
dengyihao 已提交
4426
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4427 4428 4429

  initResultSizeInfo(pOperator, numOfRows);
  initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4430
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfCols, pTaskInfo);
H
Haojun Liao 已提交
4431
  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pInfo->binfo.pCtx, numOfCols);
4432

dengyihao's avatar
dengyihao 已提交
4433
  pOperator->name = "ProjectOperator";
H
Haojun Liao 已提交
4434
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
4435
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4436 4437 4438
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
4439
  pOperator->numOfExprs = num;
4440

L
Liu Jicong 已提交
4441 4442
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
                                         destroyProjectOperatorInfo, NULL, NULL, NULL);
L
Liu Jicong 已提交
4443 4444

  pOperator->pTaskInfo = pTaskInfo;
4445
  int32_t code = appendDownstream(pOperator, &downstream, 1);
H
Haojun Liao 已提交
4446
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4447 4448
    goto _error;
  }
4449 4450

  return pOperator;
H
Haojun Liao 已提交
4451

L
Liu Jicong 已提交
4452
_error:
H
Haojun Liao 已提交
4453 4454
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4455 4456
}

4457
static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SNodeListNode* pValNode,
L
Liu Jicong 已提交
4458
                            STimeWindow win, int32_t capacity, const char* id, SInterval* pInterval, int32_t fillType) {
4459
  SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pValNode);
H
Haojun Liao 已提交
4460 4461

  STimeWindow w = TSWINDOW_INITIALIZER;
4462
  getAlignQueryTimeWindow(pInterval, pInterval->precision, win.skey, &w);
H
Haojun Liao 已提交
4463 4464

  int32_t order = TSDB_ORDER_ASC;
4465
  pInfo->pFillInfo = taosCreateFillInfo(order, w.skey, 0, capacity, numOfCols, pInterval, fillType, pColInfo, id);
H
Haojun Liao 已提交
4466

wafwerar's avatar
wafwerar 已提交
4467
  pInfo->p = taosMemoryCalloc(numOfCols, POINTER_BYTES);
H
Haojun Liao 已提交
4468 4469 4470 4471 4472 4473 4474
  if (pInfo->pFillInfo == NULL || pInfo->p == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  } else {
    return TSDB_CODE_SUCCESS;
  }
}

L
Liu Jicong 已提交
4475
SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExpr, int32_t numOfCols,
X
Xiaoyu Wang 已提交
4476 4477 4478
                                      SInterval* pInterval, STimeWindow* pWindow, SSDataBlock* pResBlock,
                                      int32_t fillType, SNodeListNode* pValueNode, bool multigroupResult,
                                      SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4479
  SFillOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SFillOperatorInfo));
L
Liu Jicong 已提交
4480
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4481

L
Liu Jicong 已提交
4482
  pInfo->pRes = pResBlock;
4483 4484
  pInfo->multigroupResult = multigroupResult;

4485 4486
  int32_t type = TSDB_FILL_NONE;
  switch (fillType) {
dengyihao's avatar
dengyihao 已提交
4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504
    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;
4505 4506 4507 4508
    default:
      type = TSDB_FILL_NONE;
  }

H
Haojun Liao 已提交
4509
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4510 4511
  initResultSizeInfo(pOperator, 4096);

X
Xiaoyu Wang 已提交
4512 4513
  int32_t code = initFillInfo(pInfo, pExpr, numOfCols, pValueNode, *pWindow, pResultInfo->capacity, pTaskInfo->id.str,
                              pInterval, type);
4514 4515 4516
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4517

dengyihao's avatar
dengyihao 已提交
4518
  pOperator->name = "FillOperator";
4519
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4520
  pOperator->status = OP_NOT_OPENED;
4521
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_FILL;
dengyihao's avatar
dengyihao 已提交
4522
  pOperator->pExpr = pExpr;
4523
  pOperator->numOfExprs = numOfCols;
dengyihao's avatar
dengyihao 已提交
4524
  pOperator->info = pInfo;
H
Haojun Liao 已提交
4525

L
Liu Jicong 已提交
4526 4527
  pOperator->fpSet =
      createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, NULL, destroySFillOperatorInfo, NULL, NULL, NULL);
4528
  pOperator->pTaskInfo = pTaskInfo;
4529
  code = appendDownstream(pOperator, &downstream, 1);
4530
  return pOperator;
H
Haojun Liao 已提交
4531

L
Liu Jicong 已提交
4532
_error:
wafwerar's avatar
wafwerar 已提交
4533 4534
  taosMemoryFreeClear(pOperator);
  taosMemoryFreeClear(pInfo);
H
Haojun Liao 已提交
4535
  return NULL;
4536 4537
}

L
Liu Jicong 已提交
4538 4539
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
4540
  SResSchema s = {0};
dengyihao's avatar
dengyihao 已提交
4541 4542 4543 4544
  s.scale = scale;
  s.type = type;
  s.bytes = bytes;
  s.slotId = slotId;
H
Haojun Liao 已提交
4545
  s.precision = precision;
H
Haojun Liao 已提交
4546 4547 4548 4549
  strncpy(s.name, name, tListLen(s.name));

  return s;
}
H
Haojun Liao 已提交
4550

H
Haojun Liao 已提交
4551 4552 4553 4554 4555 4556 4557
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
4558 4559 4560 4561 4562
  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
4563 4564 4565 4566 4567
  pCol->dataBlockId = blockId;

  return pCol;
}

H
Haojun Liao 已提交
4568
SExprInfo* createExprInfo(SNodeList* pNodeList, SNodeList* pGroupKeys, int32_t* numOfExprs) {
H
Haojun Liao 已提交
4569
  int32_t numOfFuncs = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
4570 4571 4572 4573
  int32_t numOfGroupKeys = 0;
  if (pGroupKeys != NULL) {
    numOfGroupKeys = LIST_LENGTH(pGroupKeys);
  }
H
Haojun Liao 已提交
4574

H
Haojun Liao 已提交
4575
  *numOfExprs = numOfFuncs + numOfGroupKeys;
wafwerar's avatar
wafwerar 已提交
4576
  SExprInfo* pExprs = taosMemoryCalloc(*numOfExprs, sizeof(SExprInfo));
H
Haojun Liao 已提交
4577

L
Liu Jicong 已提交
4578
  for (int32_t i = 0; i < (*numOfExprs); ++i) {
H
Haojun Liao 已提交
4579 4580 4581 4582 4583 4584
    STargetNode* pTargetNode = NULL;
    if (i < numOfFuncs) {
      pTargetNode = (STargetNode*)nodesListGetNode(pNodeList, i);
    } else {
      pTargetNode = (STargetNode*)nodesListGetNode(pGroupKeys, i - numOfFuncs);
    }
H
Haojun Liao 已提交
4585

4586
    SExprInfo* pExp = &pExprs[i];
H
Haojun Liao 已提交
4587

wafwerar's avatar
wafwerar 已提交
4588
    pExp->pExpr = taosMemoryCalloc(1, sizeof(tExprNode));
H
Haojun Liao 已提交
4589
    pExp->pExpr->_function.num = 1;
H
Haojun Liao 已提交
4590
    pExp->pExpr->_function.functionId = -1;
H
Haojun Liao 已提交
4591

4592
    int32_t type = nodeType(pTargetNode->pExpr);
H
Haojun Liao 已提交
4593
    // it is a project query, or group by column
4594
    if (type == QUERY_NODE_COLUMN) {
H
Haojun Liao 已提交
4595
      pExp->pExpr->nodeType = QUERY_NODE_COLUMN;
L
Liu Jicong 已提交
4596
      SColumnNode* pColNode = (SColumnNode*)pTargetNode->pExpr;
H
Haojun Liao 已提交
4597

G
Ganlin Zhao 已提交
4598 4599 4600
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

H
Haojun Liao 已提交
4601
      SDataType* pType = &pColNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4602 4603
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pColNode->colName);
H
Haojun Liao 已提交
4604 4605
      pExp->base.pParam[0].pCol = createColumn(pColNode->dataBlockId, pColNode->slotId, pType);
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_COLUMN;
4606
    } else if (type == QUERY_NODE_VALUE) {
4607 4608 4609 4610 4611 4612 4613
      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
4614 4615
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pValNode->node.aliasName);
4616 4617
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_VALUE;
      valueNodeToVariant(pValNode, &pExp->base.pParam[0].param);
4618
    } else if (type == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
4619
      pExp->pExpr->nodeType = QUERY_NODE_FUNCTION;
H
Haojun Liao 已提交
4620 4621 4622
      SFunctionNode* pFuncNode = (SFunctionNode*)pTargetNode->pExpr;

      SDataType* pType = &pFuncNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4623 4624
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pFuncNode->node.aliasName);
H
Haojun Liao 已提交
4625

H
Haojun Liao 已提交
4626
      pExp->pExpr->_function.functionId = pFuncNode->funcId;
H
Haojun Liao 已提交
4627
      pExp->pExpr->_function.pFunctNode = pFuncNode;
dengyihao's avatar
dengyihao 已提交
4628 4629
      strncpy(pExp->pExpr->_function.functionName, pFuncNode->functionName,
              tListLen(pExp->pExpr->_function.functionName));
H
Haojun Liao 已提交
4630 4631

      int32_t numOfParam = LIST_LENGTH(pFuncNode->pParameterList);
G
Ganlin Zhao 已提交
4632 4633 4634 4635

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

H
Haojun Liao 已提交
4636
      for (int32_t j = 0; j < numOfParam; ++j) {
4637
        SNode* p1 = nodesListGetNode(pFuncNode->pParameterList, j);
G
Ganlin Zhao 已提交
4638
        if (p1->type == QUERY_NODE_COLUMN) {
dengyihao's avatar
dengyihao 已提交
4639
          SColumnNode* pcn = (SColumnNode*)p1;
G
Ganlin Zhao 已提交
4640 4641

          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_COLUMN;
H
Haojun Liao 已提交
4642
          pExp->base.pParam[j].pCol = createColumn(pcn->dataBlockId, pcn->slotId, &pcn->node.resType);
G
Ganlin Zhao 已提交
4643 4644 4645
        } else if (p1->type == QUERY_NODE_VALUE) {
          SValueNode* pvn = (SValueNode*)p1;
          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_VALUE;
4646
          valueNodeToVariant(pvn, &pExp->base.pParam[j].param);
G
Ganlin Zhao 已提交
4647
        }
H
Haojun Liao 已提交
4648
      }
4649
    } else if (type == QUERY_NODE_OPERATOR) {
H
Haojun Liao 已提交
4650
      pExp->pExpr->nodeType = QUERY_NODE_OPERATOR;
L
Liu Jicong 已提交
4651
      SOperatorNode* pNode = (SOperatorNode*)pTargetNode->pExpr;
4652

G
Ganlin Zhao 已提交
4653 4654 4655
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

4656
      SDataType* pType = &pNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4657 4658
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pNode->node.aliasName);
4659 4660 4661
      pExp->pExpr->_optrRoot.pRootNode = pTargetNode->pExpr;
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4662 4663 4664
    }
  }

H
Haojun Liao 已提交
4665
  return pExprs;
H
Haojun Liao 已提交
4666 4667
}

4668
static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model) {
wafwerar's avatar
wafwerar 已提交
4669
  SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
4670
  setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
4671

4672
  pTaskInfo->cost.created = taosGetTimestampMs();
H
Haojun Liao 已提交
4673
  pTaskInfo->id.queryId = queryId;
dengyihao's avatar
dengyihao 已提交
4674
  pTaskInfo->execModel = model;
H
Haojun Liao 已提交
4675

wafwerar's avatar
wafwerar 已提交
4676
  char* p = taosMemoryCalloc(1, 128);
L
Liu Jicong 已提交
4677
  snprintf(p, 128, "TID:0x%" PRIx64 " QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
4678
  pTaskInfo->id.str = p;
H
Haojun Liao 已提交
4679

4680 4681
  return pTaskInfo;
}
H
Haojun Liao 已提交
4682

L
Liu Jicong 已提交
4683 4684
static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                                      STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId);
H
Haojun Liao 已提交
4685

L
Liu Jicong 已提交
4686 4687
static int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                                  uint64_t queryId, uint64_t taskId);
4688
static SArray* extractTableIdList(const STableGroupInfo* pTableGroupInfo);
H
Haojun Liao 已提交
4689
static SArray* extractColumnInfo(SNodeList* pNodeList);
X
Xiaoyu Wang 已提交
4690 4691
static SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols,
                                   int32_t type);
4692

4693
static SArray* createSortInfo(SNodeList* pNodeList);
4694
static SArray* extractPartitionColInfo(SNodeList* pNodeList);
4695
static int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode);
4696
static void    setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode);
4697 4698 4699

static SInterval extractIntervalInfo(const STableScanPhysiNode* pTableScanNode) {
  SInterval interval = {
L
Liu Jicong 已提交
4700 4701
      .interval = pTableScanNode->interval,
      .sliding = pTableScanNode->sliding,
4702
      .intervalUnit = pTableScanNode->intervalUnit,
L
Liu Jicong 已提交
4703 4704
      .slidingUnit = pTableScanNode->slidingUnit,
      .offset = pTableScanNode->offset,
4705 4706 4707 4708
  };

  return interval;
}
H
Haojun Liao 已提交
4709

H
Haojun Liao 已提交
4710
SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
4711
                                  uint64_t queryId, uint64_t taskId, STableGroupInfo* pTableGroupInfo) {
4712 4713
  int32_t type = nodeType(pPhyNode);

X
Xiaoyu Wang 已提交
4714
  if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
H
Haojun Liao 已提交
4715
    if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == type) {
dengyihao's avatar
dengyihao 已提交
4716 4717
      SScanPhysiNode*      pScanPhyNode = (SScanPhysiNode*)pPhyNode;
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4718

L
Liu Jicong 已提交
4719
      int32_t     numOfCols = 0;
H
Haojun Liao 已提交
4720
      tsdbReaderT pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableGroupInfo, (uint64_t)queryId, taskId);
4721
      if (pDataReader == NULL && terrno != 0) {
4722 4723
        return NULL;
      }
4724

4725 4726 4727 4728
      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;

      SArray* pColList = extractColMatchInfo(pScanPhyNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID);
      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4729

4730
      SQueryTableDataCond cond = {0};
X
Xiaoyu Wang 已提交
4731
      int32_t             code = initQueryTableDataCond(&cond, pTableScanNode);
4732 4733 4734
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
      }
4735

X
Xiaoyu Wang 已提交
4736 4737 4738 4739
      SInterval      interval = extractIntervalInfo(pTableScanNode);
      SOperatorInfo* pOperator = createTableScanOperatorInfo(
          pDataReader, &cond, numOfCols, pTableScanNode->dataRequired, pTableScanNode->scanSeq, pColList, pResBlock,
          pScanPhyNode->node.pConditions, &interval, pTableScanNode->ratio, pTaskInfo);
4740 4741 4742
      STableScanInfo* pScanInfo = pOperator->info;
      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
      return pOperator;
H
Haojun Liao 已提交
4743
    } else if (QUERY_NODE_PHYSICAL_PLAN_EXCHANGE == type) {
H
Haojun Liao 已提交
4744
      SExchangePhysiNode* pExchange = (SExchangePhysiNode*)pPhyNode;
4745
      SSDataBlock*        pResBlock = createResDataBlock(pExchange->node.pOutputDataBlockDesc);
4746
      return createExchangeOperatorInfo(pHandle->pMsgCb->clientRpc, pExchange->pSrcEndPoints, pResBlock, pTaskInfo);
H
Haojun Liao 已提交
4747
    } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN == type) {
H
Haojun Liao 已提交
4748
      SScanPhysiNode* pScanPhyNode = (SScanPhysiNode*)pPhyNode;  // simple child table.
4749

X
Xiaoyu Wang 已提交
4750 4751
      int32_t code = doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo,
                                        queryId, taskId);
H
Haojun Liao 已提交
4752
      SArray* tableIdList = extractTableIdList(pTableGroupInfo);
H
Haojun Liao 已提交
4753

4754
      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;
X
Xiaoyu Wang 已提交
4755
      SSDataBlock*        pResBlock = createResDataBlock(pDescNode);
4756

X
Xiaoyu Wang 已提交
4757 4758
      int32_t        numOfCols = 0;
      SArray*        pCols = extractColMatchInfo(pScanPhyNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID);
X
Xiaoyu Wang 已提交
4759 4760
      SOperatorInfo* pOperator = createStreamScanOperatorInfo(pHandle->reader, pResBlock, pCols, tableIdList, pTaskInfo,
                                                              pScanPhyNode->node.pConditions);
4761
      taosArrayDestroy(tableIdList);
H
Haojun Liao 已提交
4762
      return pOperator;
H
Haojun Liao 已提交
4763
    } else if (QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN == type) {
L
Liu Jicong 已提交
4764
      SSystemTableScanPhysiNode* pSysScanPhyNode = (SSystemTableScanPhysiNode*)pPhyNode;
L
Liu Jicong 已提交
4765
      SScanPhysiNode*            pScanNode = &pSysScanPhyNode->scan;
H
Haojun Liao 已提交
4766

4767 4768 4769
      SDataBlockDescNode* pDescNode = pScanNode->node.pOutputDataBlockDesc;

      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4770

4771
      int32_t numOfOutputCols = 0;
4772
      SArray* colList = extractColMatchInfo(pScanNode->pScanCols, pDescNode, &numOfOutputCols, COL_MATCH_FROM_COL_ID);
L
Liu Jicong 已提交
4773
      SOperatorInfo* pOperator = createSysTableScanOperatorInfo(
L
Liu Jicong 已提交
4774 4775
          pHandle, pResBlock, &pScanNode->tableName, pScanNode->node.pConditions, pSysScanPhyNode->mgmtEpSet, colList,
          pTaskInfo, pSysScanPhyNode->showRewrite, pSysScanPhyNode->accountId);
H
Haojun Liao 已提交
4776
      return pOperator;
4777
    } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
X
Xiaoyu Wang 已提交
4778
      STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
4779 4780 4781 4782

      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;

      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4783

X
Xiaoyu Wang 已提交
4784 4785
      int32_t code = doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo,
                                        queryId, taskId);
4786 4787 4788 4789
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
      }

X
Xiaoyu Wang 已提交
4790
      int32_t    num = 0;
4791 4792 4793
      SExprInfo* pExprInfo = createExprInfo(pScanPhyNode->pScanPseudoCols, NULL, &num);

      int32_t numOfOutputCols = 0;
X
Xiaoyu Wang 已提交
4794 4795
      SArray* colList =
          extractColMatchInfo(pScanPhyNode->pScanPseudoCols, pDescNode, &numOfOutputCols, COL_MATCH_FROM_COL_ID);
4796

X
Xiaoyu Wang 已提交
4797 4798
      SOperatorInfo* pOperator =
          createTagScanOperatorInfo(pHandle, pExprInfo, num, pResBlock, colList, pTableGroupInfo, pTaskInfo);
4799
      return pOperator;
H
Haojun Liao 已提交
4800 4801
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4802 4803 4804
    }
  }

4805 4806
  int32_t num = 0;
  size_t  size = LIST_LENGTH(pPhyNode->pChildren);
H
Haojun Liao 已提交
4807

4808
  SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
dengyihao's avatar
dengyihao 已提交
4809
  for (int32_t i = 0; i < size; ++i) {
4810 4811
    SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableGroupInfo);
4812 4813 4814
    if (ops[i] == NULL) {
      return NULL;
    }
4815
  }
H
Haojun Liao 已提交
4816

4817
  SOperatorInfo* pOptr = NULL;
H
Haojun Liao 已提交
4818
  if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
dengyihao's avatar
dengyihao 已提交
4819 4820
    SProjectPhysiNode* pProjPhyNode = (SProjectPhysiNode*)pPhyNode;
    SExprInfo*         pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &num);
H
Haojun Liao 已提交
4821

4822
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
dengyihao's avatar
dengyihao 已提交
4823 4824
    SLimit       limit = {.limit = pProjPhyNode->limit, .offset = pProjPhyNode->offset};
    SLimit       slimit = {.limit = pProjPhyNode->slimit, .offset = pProjPhyNode->soffset};
4825
    pOptr = createProjectOperatorInfo(ops[0], pExprInfo, num, pResBlock, &limit, &slimit, pTaskInfo);
H
Haojun Liao 已提交
4826 4827 4828
  } else if (QUERY_NODE_PHYSICAL_PLAN_AGG == type) {
    SAggPhysiNode* pAggNode = (SAggPhysiNode*)pPhyNode;
    SExprInfo*     pExprInfo = createExprInfo(pAggNode->pAggFuncs, pAggNode->pGroupKeys, &num);
4829
    SSDataBlock*   pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4830

dengyihao's avatar
dengyihao 已提交
4831
    int32_t    numOfScalarExpr = 0;
4832 4833 4834 4835 4836
    SExprInfo* pScalarExprInfo = NULL;
    if (pAggNode->pExprs != NULL) {
      pScalarExprInfo = createExprInfo(pAggNode->pExprs, NULL, &numOfScalarExpr);
    }

H
Haojun Liao 已提交
4837 4838
    if (pAggNode->pGroupKeys != NULL) {
      SArray* pColList = extractColumnInfo(pAggNode->pGroupKeys);
dengyihao's avatar
dengyihao 已提交
4839 4840
      pOptr = createGroupOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pAggNode->node.pConditions,
                                      pScalarExprInfo, numOfScalarExpr, pTaskInfo, NULL);
H
Haojun Liao 已提交
4841
    } else {
dengyihao's avatar
dengyihao 已提交
4842 4843
      pOptr = createAggregateOperatorInfo(ops[0], pExprInfo, num, pResBlock, pScalarExprInfo, numOfScalarExpr,
                                          pTaskInfo, pTableGroupInfo);
H
Haojun Liao 已提交
4844
    }
X
Xiaoyu Wang 已提交
4845
  } else if (QUERY_NODE_PHYSICAL_PLAN_INTERVAL == type || QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL == type) {
H
Haojun Liao 已提交
4846
    SIntervalPhysiNode* pIntervalPhyNode = (SIntervalPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4847

H
Haojun Liao 已提交
4848
    SExprInfo*   pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
4849
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4850

dengyihao's avatar
dengyihao 已提交
4851 4852 4853 4854 4855 4856
    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
4857

dengyihao's avatar
dengyihao 已提交
4858 4859
    STimeWindowAggSupp as = {.waterMark = pIntervalPhyNode->window.watermark,
                             .calTrigger = pIntervalPhyNode->window.triggerType};
4860

4861 4862 4863 4864
    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
4865
  } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {
H
Haojun Liao 已提交
4866
    SSortPhysiNode* pSortPhyNode = (SSortPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4867

4868 4869 4870
    SDataBlockDescNode* pDescNode = pPhyNode->pOutputDataBlockDesc;

    SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4871
    SArray*      info = createSortInfo(pSortPhyNode->pSortKeys);
4872 4873

    int32_t    numOfCols = 0;
4874 4875 4876
    SExprInfo* pExprInfo = createExprInfo(pSortPhyNode->pExprs, NULL, &numOfCols);

    int32_t numOfOutputCols = 0;
4877
    SArray* pColList = extractColMatchInfo(pSortPhyNode->pTargets, pDescNode, &numOfOutputCols, COL_MATCH_FROM_SLOT_ID);
4878

4879
    pOptr = createSortOperatorInfo(ops[0], pResBlock, info, pExprInfo, numOfCols, pColList, pTaskInfo);
H
Haojun Liao 已提交
4880
  } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
H
Haojun Liao 已提交
4881 4882
    SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;

X
Xiaoyu Wang 已提交
4883 4884
    STimeWindowAggSupp as = {.waterMark = pSessionNode->window.watermark,
                             .calTrigger = pSessionNode->window.triggerType};
4885

H
Haojun Liao 已提交
4886
    SExprInfo*   pExprInfo = createExprInfo(pSessionNode->window.pFuncs, NULL, &num);
4887
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4888 4889
    int32_t      tsSlotId = ((SColumnNode*)pSessionNode->window.pTspk)->slotId;

X
Xiaoyu Wang 已提交
4890 4891
    pOptr =
        createSessionAggOperatorInfo(ops[0], pExprInfo, num, pResBlock, pSessionNode->gap, tsSlotId, &as, pTaskInfo);
H
Haojun Liao 已提交
4892
  } else if (QUERY_NODE_PHYSICAL_PLAN_PARTITION == type) {
dengyihao's avatar
dengyihao 已提交
4893 4894 4895
    SPartitionPhysiNode* pPartNode = (SPartitionPhysiNode*)pPhyNode;
    SArray*              pColList = extractPartitionColInfo(pPartNode->pPartitionKeys);
    SSDataBlock*         pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4896 4897

    SExprInfo* pExprInfo = createExprInfo(pPartNode->pTargets, NULL, &num);
4898
    pOptr = createPartitionOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pTaskInfo, NULL);
4899
  } else if (QUERY_NODE_PHYSICAL_PLAN_STATE_WINDOW == type) {
dengyihao's avatar
dengyihao 已提交
4900
    SStateWinodwPhysiNode* pStateNode = (SStateWinodwPhysiNode*)pPhyNode;
4901

4902 4903
    STimeWindowAggSupp as = {.waterMark = pStateNode->window.watermark, .calTrigger = pStateNode->window.triggerType};

dengyihao's avatar
dengyihao 已提交
4904
    SExprInfo*   pExprInfo = createExprInfo(pStateNode->window.pFuncs, NULL, &num);
4905
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4906 4907 4908
    int32_t      tsSlotId = ((SColumnNode*)pStateNode->window.pTspk)->slotId;

    pOptr = createStatewindowOperatorInfo(ops[0], pExprInfo, num, pResBlock, &as, tsSlotId, pTaskInfo);
4909
  } else if (QUERY_NODE_PHYSICAL_PLAN_JOIN == type) {
dengyihao's avatar
dengyihao 已提交
4910 4911
    SJoinPhysiNode* pJoinNode = (SJoinPhysiNode*)pPhyNode;
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4912 4913

    SExprInfo* pExprInfo = createExprInfo(pJoinNode->pTargets, NULL, &num);
H
Haojun Liao 已提交
4914
    pOptr = createMergeJoinOperatorInfo(ops, size, pExprInfo, num, pResBlock, pJoinNode->pOnConditions, pTaskInfo);
4915 4916
  } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
    SFillPhysiNode* pFillNode = (SFillPhysiNode*)pPhyNode;
X
Xiaoyu Wang 已提交
4917 4918
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
    SExprInfo*      pExprInfo = createExprInfo(pFillNode->pTargets, NULL, &num);
4919

4920
    SInterval* pInterval = &((SIntervalAggOperatorInfo*)ops[0]->info)->interval;
X
Xiaoyu Wang 已提交
4921 4922
    pOptr = createFillOperatorInfo(ops[0], pExprInfo, num, pInterval, &pFillNode->timeRange, pResBlock, pFillNode->mode,
                                   (SNodeListNode*)pFillNode->pValues, false, pTaskInfo);
H
Haojun Liao 已提交
4923 4924
  } else {
    ASSERT(0);
H
Haojun Liao 已提交
4925
  }
4926 4927 4928

  taosMemoryFree(ops);
  return pOptr;
4929
}
H
Haojun Liao 已提交
4930

L
Liu Jicong 已提交
4931
static int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode) {
4932 4933 4934 4935 4936 4937
  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
4938
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
4939 4940 4941 4942 4943 4944
    return terrno;
  }

  pCond->twindow = pTableScanNode->scanRange;

#if 1
L
Liu Jicong 已提交
4945
  // todo work around a problem, remove it later
4946 4947
  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
4948
    TSWAP(pCond->twindow.skey, pCond->twindow.ekey);
H
Haojun Liao 已提交
4949
  }
4950
#endif
H
Haojun Liao 已提交
4951

4952 4953
  pCond->type = BLOCK_LOAD_OFFSET_SEQ_ORDER;
  //  pCond->type = pTableScanNode->scanFlag;
H
Haojun Liao 已提交
4954

H
Haojun Liao 已提交
4955
  int32_t j = 0;
4956
  for (int32_t i = 0; i < pCond->numOfCols; ++i) {
H
Haojun Liao 已提交
4957 4958
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pTableScanNode->scan.pScanCols, i);
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;
H
Haojun Liao 已提交
4959 4960 4961
    if (pColNode->colType == COLUMN_TYPE_TAG) {
      continue;
    }
H
Haojun Liao 已提交
4962

4963 4964 4965
    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
4966
    j += 1;
H
Haojun Liao 已提交
4967 4968
  }

4969 4970
  pCond->numOfCols = j;
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
4971
}
H
Haojun Liao 已提交
4972

H
Haojun Liao 已提交
4973
SArray* extractColumnInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
4974
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
4975 4976 4977 4978 4979 4980
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
4981 4982
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
H
Haojun Liao 已提交
4983

4984 4985 4986 4987 4988
    if (nodeType(pNode->pExpr) == QUERY_NODE_COLUMN) {
      SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

      // todo extract method
      SColumn c = {0};
L
Liu Jicong 已提交
4989 4990 4991 4992 4993
      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;
4994 4995 4996 4997
      c.precision = pColNode->node.resType.precision;

      taosArrayPush(pList, &c);
    } else if (nodeType(pNode->pExpr) == QUERY_NODE_VALUE) {
L
Liu Jicong 已提交
4998 4999
      SValueNode* pValNode = (SValueNode*)pNode->pExpr;
      SColumn     c = {0};
5000
      c.slotId = pNode->slotId;
L
Liu Jicong 已提交
5001 5002 5003 5004
      c.colId = pNode->slotId;
      c.type = pValNode->node.type;
      c.bytes = pValNode->node.resType.bytes;
      c.scale = pValNode->node.resType.scale;
5005 5006 5007 5008
      c.precision = pValNode->node.resType.precision;

      taosArrayPush(pList, &c);
    }
H
Haojun Liao 已提交
5009 5010 5011 5012 5013
  }

  return pList;
}

5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027
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
5028 5029 5030
    c.colId = pColNode->colId;
    c.type = pColNode->node.resType.type;
    c.bytes = pColNode->node.resType.bytes;
5031 5032 5033 5034 5035 5036 5037 5038 5039
    c.precision = pColNode->node.resType.precision;
    c.scale = pColNode->node.resType.scale;

    taosArrayPush(pList, &c);
  }

  return pList;
}

5040
SArray* createSortInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
5041
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5042 5043 5044 5045 5046 5047
  SArray* pList = taosArrayInit(numOfCols, sizeof(SBlockOrderInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return pList;
  }

L
Liu Jicong 已提交
5048
  for (int32_t i = 0; i < numOfCols; ++i) {
5049
    SOrderByExprNode* pSortKey = (SOrderByExprNode*)nodesListGetNode(pNodeList, i);
L
Liu Jicong 已提交
5050 5051
    SBlockOrderInfo   bi = {0};
    bi.order = (pSortKey->order == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
H
Haojun Liao 已提交
5052 5053 5054
    bi.nullFirst = (pSortKey->nullOrder == NULL_ORDER_FIRST);

    SColumnNode* pColNode = (SColumnNode*)pSortKey->pExpr;
5055
    bi.slotId = pColNode->slotId;
H
Haojun Liao 已提交
5056 5057 5058 5059 5060 5061
    taosArrayPush(pList, &bi);
  }

  return pList;
}

X
Xiaoyu Wang 已提交
5062 5063
SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols,
                            int32_t type) {
L
Liu Jicong 已提交
5064
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5065 5066 5067 5068 5069 5070
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColMatchInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
5071 5072 5073
  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
5074 5075

    SColMatchInfo c = {0};
X
Xiaoyu Wang 已提交
5076 5077 5078 5079
    c.output = true;
    c.colId = pColNode->colId;
    c.srcSlotId = pColNode->slotId;
    c.matchType = type;
H
Haojun Liao 已提交
5080 5081 5082 5083
    c.targetSlotId = pNode->slotId;
    taosArrayPush(pList, &c);
  }

H
Haojun Liao 已提交
5084 5085
  *numOfOutputCols = 0;
  int32_t num = LIST_LENGTH(pOutputNodeList->pSlots);
L
Liu Jicong 已提交
5086 5087
  for (int32_t i = 0; i < num; ++i) {
    SSlotDescNode* pNode = (SSlotDescNode*)nodesListGetNode(pOutputNodeList->pSlots, i);
5088

5089
    // todo: add reserve flag check
5090 5091
    // it is a column reserved for the arithmetic expression calculation
    if (pNode->slotId >= numOfCols) {
5092 5093 5094 5095
      (*numOfOutputCols) += 1;
      continue;
    }

H
Haojun Liao 已提交
5096
    SColMatchInfo* info = taosArrayGet(pList, pNode->slotId);
H
Haojun Liao 已提交
5097 5098 5099 5100 5101
    if (pNode->output) {
      (*numOfOutputCols) += 1;
    } else {
      info->output = false;
    }
H
Haojun Liao 已提交
5102 5103
  }

H
Haojun Liao 已提交
5104 5105 5106
  return pList;
}

dengyihao's avatar
dengyihao 已提交
5107 5108
int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                           uint64_t queryId, uint64_t taskId) {
5109
  int32_t code = 0;
H
Haojun Liao 已提交
5110
  if (tableType == TSDB_SUPER_TABLE) {
H
Haojun Liao 已提交
5111
    code = tsdbQuerySTableByTagCond(metaHandle, tableUid, 0, NULL, 0, 0, NULL, pGroupInfo, NULL, 0, queryId, taskId);
H
Haojun Liao 已提交
5112
  } else {  // Create one table group.
H
Haojun Liao 已提交
5113
    code = tsdbGetOneTableGroup(metaHandle, tableUid, 0, pGroupInfo);
5114 5115 5116 5117
  }

  return code;
}
H
Haojun Liao 已提交
5118

5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136
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
5137 5138
tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                               STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId) {
5139
  uint64_t uid = pTableScanNode->scan.uid;
L
Liu Jicong 已提交
5140 5141
  int32_t  code =
      doCreateTableGroup(pHandle->meta, pTableScanNode->scan.tableType, uid, pTableGroupInfo, queryId, taskId);
5142 5143
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
H
Haojun Liao 已提交
5144
  }
H
Haojun Liao 已提交
5145

H
Haojun Liao 已提交
5146
  if (pTableGroupInfo->numOfTables == 0) {
H
Haojun Liao 已提交
5147
    code = 0;
L
Liu Jicong 已提交
5148
    qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
5149 5150
    goto _error;
  }
H
Haojun Liao 已提交
5151

5152 5153 5154 5155
  SQueryTableDataCond cond = {0};
  code = initQueryTableDataCond(&cond, pTableScanNode);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
X
Xiaoyu Wang 已提交
5156
  }
5157

5158
  return tsdbQueryTables(pHandle->vnode, &cond, pTableGroupInfo, queryId, taskId);
H
Haojun Liao 已提交
5159

L
Liu Jicong 已提交
5160
_error:
H
Haojun Liao 已提交
5161 5162 5163 5164
  terrno = code;
  return NULL;
}

dengyihao's avatar
dengyihao 已提交
5165 5166
int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
                               EOPTR_EXEC_MODEL model) {
H
Haojun Liao 已提交
5167 5168
  uint64_t queryId = pPlan->id.queryId;

H
Haojun Liao 已提交
5169
  int32_t code = TSDB_CODE_SUCCESS;
5170
  *pTaskInfo = createExecTaskInfo(queryId, taskId, model);
H
Haojun Liao 已提交
5171 5172 5173 5174
  if (*pTaskInfo == NULL) {
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    goto _complete;
  }
H
Haojun Liao 已提交
5175

L
Liu Jicong 已提交
5176 5177
  (*pTaskInfo)->pRoot =
      createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId, &(*pTaskInfo)->tableqinfoGroupInfo);
D
dapan1121 已提交
5178
  if (NULL == (*pTaskInfo)->pRoot) {
L
Liu Jicong 已提交
5179
    code = terrno;
D
dapan1121 已提交
5180 5181
    goto _complete;
  }
H
Haojun Liao 已提交
5182

5183
  if ((*pTaskInfo)->pRoot == NULL) {
H
Haojun Liao 已提交
5184
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
H
Haojun Liao 已提交
5185
    goto _complete;
5186 5187
  }

H
Haojun Liao 已提交
5188 5189
  return code;

H
Haojun Liao 已提交
5190
_complete:
wafwerar's avatar
wafwerar 已提交
5191
  taosMemoryFreeClear(*pTaskInfo);
H
Haojun Liao 已提交
5192 5193 5194

  terrno = code;
  return code;
H
Haojun Liao 已提交
5195 5196
}

H
Haojun Liao 已提交
5197
void setResultBufSize(STaskAttr* pQueryAttr, SResultInfo* pResultInfo) {
5198 5199 5200 5201 5202 5203 5204 5205
  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
5206 5207 5208 5209 5210 5211 5212 5213 5214 5215
  //  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;
  //  }
5216 5217

  pResultInfo->threshold = (int32_t)(pResultInfo->capacity * THRESHOLD_RATIO);
H
Haojun Liao 已提交
5218
  pResultInfo->totalRows = 0;
5219 5220
}

L
Liu Jicong 已提交
5221
// TODO refactor
5222
void freeColumnFilterInfo(SColumnFilterInfo* pFilter, int32_t numOfFilters) {
L
Liu Jicong 已提交
5223 5224 5225
  if (pFilter == NULL || numOfFilters == 0) {
    return;
  }
5226

L
Liu Jicong 已提交
5227 5228 5229
  for (int32_t i = 0; i < numOfFilters; i++) {
    if (pFilter[i].filterstr && pFilter[i].pz) {
      taosMemoryFree((void*)(pFilter[i].pz));
5230
    }
L
Liu Jicong 已提交
5231
  }
5232

L
Liu Jicong 已提交
5233
  taosMemoryFree(pFilter);
5234 5235 5236 5237
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo) {
  if (pTableqinfoGroupInfo->pGroupList != NULL) {
L
Liu Jicong 已提交
5238
    int32_t numOfGroups = (int32_t)taosArrayGetSize(pTableqinfoGroupInfo->pGroupList);
5239
    for (int32_t i = 0; i < numOfGroups; ++i) {
L
Liu Jicong 已提交
5240
      SArray* p = taosArrayGetP(pTableqinfoGroupInfo->pGroupList, i);
5241 5242

      size_t num = taosArrayGetSize(p);
L
Liu Jicong 已提交
5243
      for (int32_t j = 0; j < num; ++j) {
5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259
        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
5260
void doDestroyTask(SExecTaskInfo* pTaskInfo) {
H
Haojun Liao 已提交
5261 5262
  qDebug("%s execTask is freed", GET_TASKID(pTaskInfo));

H
Haojun Liao 已提交
5263
  doDestroyTableQueryInfo(&pTaskInfo->tableqinfoGroupInfo);
H
Haojun Liao 已提交
5264
  destroyOperatorInfo(pTaskInfo->pRoot);
L
Liu Jicong 已提交
5265 5266
  //  taosArrayDestroy(pTaskInfo->summary.queryProfEvents);
  //  taosHashCleanup(pTaskInfo->summary.operatorProfResults);
5267

wafwerar's avatar
wafwerar 已提交
5268 5269 5270
  taosMemoryFreeClear(pTaskInfo->sql);
  taosMemoryFreeClear(pTaskInfo->id.str);
  taosMemoryFreeClear(pTaskInfo);
5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282
}

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
5283
      int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295
      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
5296 5297
  //  size_t s3 = sizeof(STableCheckInfo);  buffer consumption in tsdb
  return (int64_t)(s1 * 1.5 * numOfTables);
5298 5299 5300 5301 5302 5303 5304
}

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
5305
    while (1) {
5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331
      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
5332

dengyihao's avatar
dengyihao 已提交
5333 5334
int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SExplainExecInfo** pRes, int32_t* capacity,
                                   int32_t* resNum) {
D
dapan1121 已提交
5335 5336
  if (*resNum >= *capacity) {
    *capacity += 10;
dengyihao's avatar
dengyihao 已提交
5337

D
dapan1121 已提交
5338 5339
    *pRes = taosMemoryRealloc(*pRes, (*capacity) * sizeof(SExplainExecInfo));
    if (NULL == *pRes) {
D
dapan1121 已提交
5340
      qError("malloc %d failed", (*capacity) * (int32_t)sizeof(SExplainExecInfo));
D
dapan1121 已提交
5341 5342 5343 5344
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

D
dapan1121 已提交
5345 5346 5347 5348
  (*pRes)[*resNum].numOfRows = operatorInfo->resultInfo.totalRows;
  (*pRes)[*resNum].startupCost = operatorInfo->cost.openCost;
  (*pRes)[*resNum].totalCost = operatorInfo->cost.totalCost;

5349 5350
  if (operatorInfo->fpSet.getExplainFn) {
    int32_t code = (*operatorInfo->fpSet.getExplainFn)(operatorInfo, &(*pRes)->verboseInfo);
D
dapan1121 已提交
5351 5352 5353 5354 5355
    if (code) {
      qError("operator getExplainFn failed, error:%s", tstrerror(code));
      return code;
    }
  }
dengyihao's avatar
dengyihao 已提交
5356

D
dapan1121 已提交
5357
  ++(*resNum);
dengyihao's avatar
dengyihao 已提交
5358

D
dapan1121 已提交
5359
  int32_t code = 0;
D
dapan1121 已提交
5360 5361
  for (int32_t i = 0; i < operatorInfo->numOfDownstream; ++i) {
    code = getOperatorExplainExecInfo(operatorInfo->pDownstream[i], pRes, capacity, resNum);
D
dapan1121 已提交
5362 5363 5364 5365 5366 5367 5368
    if (code) {
      taosMemoryFreeClear(*pRes);
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
D
dapan1121 已提交
5369 5370
}

5371
static SSDataBlock* doMergeJoin(struct SOperatorInfo* pOperator) {
5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383
  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);
5384
      pJoinInfo->pLeft = ds1->fpSet.getNextFn(ds1);
5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396
      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);
5397
      pJoinInfo->pRight = ds2->fpSet.getNextFn(ds2);
5398 5399 5400 5401 5402 5403 5404 5405 5406 5407
      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
5408
    char*            pLeftVal = colDataGetData(pLeftCol, pJoinInfo->leftPos);
5409 5410

    SColumnInfoData* pRightCol = taosArrayGet(pJoinInfo->pRight->pDataBlock, pJoinInfo->rightCol.slotId);
dengyihao's avatar
dengyihao 已提交
5411
    char*            pRightVal = colDataGetData(pRightCol, pJoinInfo->rightPos);
5412 5413

    // only the timestamp match support for ordinary table
dengyihao's avatar
dengyihao 已提交
5414 5415
    ASSERT(pLeftCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
    if (*(int64_t*)pLeftVal == *(int64_t*)pRightVal) {
5416
      for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
dengyihao's avatar
dengyihao 已提交
5417
        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, i);
5418

dengyihao's avatar
dengyihao 已提交
5419
        SExprInfo* pExprInfo = &pOperator->pExpr[i];
5420

dengyihao's avatar
dengyihao 已提交
5421 5422
        int32_t blockId = pExprInfo->base.pParam[0].pCol->dataBlockId;
        int32_t slotId = pExprInfo->base.pParam[0].pCol->slotId;
5423

dengyihao's avatar
dengyihao 已提交
5424 5425 5426 5427 5428 5429
        SColumnInfoData* pSrc = NULL;
        if (pJoinInfo->pLeft->info.blockId == blockId) {
          pSrc = taosArrayGet(pJoinInfo->pLeft->pDataBlock, slotId);
        } else {
          pSrc = taosArrayGet(pJoinInfo->pRight->pDataBlock, slotId);
        }
5430

dengyihao's avatar
dengyihao 已提交
5431 5432 5433 5434 5435
        if (colDataIsNull_s(pSrc, pJoinInfo->leftPos)) {
          colDataAppendNULL(pDst, nrows);
        } else {
          char* p = colDataGetData(pSrc, pJoinInfo->leftPos);
          colDataAppend(pDst, nrows, p, false);
5436
        }
dengyihao's avatar
dengyihao 已提交
5437
      }
5438

dengyihao's avatar
dengyihao 已提交
5439 5440
      pJoinInfo->leftPos += 1;
      pJoinInfo->rightPos += 1;
5441

dengyihao's avatar
dengyihao 已提交
5442 5443 5444
      nrows += 1;
    } else if (*(int64_t*)pLeftVal < *(int64_t*)pRightVal) {
      pJoinInfo->leftPos += 1;
D
dapan1121 已提交
5445

dengyihao's avatar
dengyihao 已提交
5446 5447
      if (pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
        continue;
5448
      }
dengyihao's avatar
dengyihao 已提交
5449 5450 5451 5452
    } else if (*(int64_t*)pLeftVal > *(int64_t*)pRightVal) {
      pJoinInfo->rightPos += 1;
      if (pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
        continue;
5453
      }
dengyihao's avatar
dengyihao 已提交
5454
    }
5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465

    // 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
5466
SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
X
Xiaoyu Wang 已提交
5467 5468
                                           int32_t numOfCols, SSDataBlock* pResBlock, SNode* pOnCondition,
                                           SExecTaskInfo* pTaskInfo) {
5469
  SJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SJoinOperatorInfo));
dengyihao's avatar
dengyihao 已提交
5470
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
5471 5472 5473 5474
  if (pOperator == NULL || pInfo == NULL) {
    goto _error;
  }

5475
  initResultSizeInfo(pOperator, 4096);
5476

X
Xiaoyu Wang 已提交
5477 5478
  pInfo->pRes = pResBlock;
  pOperator->name = "MergeJoinOperator";
5479
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_JOIN;
X
Xiaoyu Wang 已提交
5480 5481 5482
  pOperator->blocking = false;
  pOperator->status = OP_NOT_OPENED;
  pOperator->pExpr = pExprInfo;
5483
  pOperator->numOfExprs = numOfCols;
X
Xiaoyu Wang 已提交
5484
  pOperator->info = pInfo;
dengyihao's avatar
dengyihao 已提交
5485
  pOperator->pTaskInfo = pTaskInfo;
5486

5487 5488 5489 5490
  SOperatorNode* pNode = (SOperatorNode*)pOnCondition;
  setJoinColumnInfo(&pInfo->leftCol, (SColumnNode*)pNode->pLeft);
  setJoinColumnInfo(&pInfo->rightCol, (SColumnNode*)pNode->pRight);

L
Liu Jicong 已提交
5491
  pOperator->fpSet =
H
Haojun Liao 已提交
5492
      createOperatorFpSet(operatorDummyOpenFn, doMergeJoin, NULL, NULL, destroyMergeJoinOperator, NULL, NULL, NULL);
5493
  int32_t code = appendDownstream(pOperator, pDownstream, numOfDownstream);
5494 5495 5496 5497
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }

5498 5499
  return pOperator;

dengyihao's avatar
dengyihao 已提交
5500
_error:
5501 5502 5503 5504
  taosMemoryFree(pInfo);
  taosMemoryFree(pOperator);
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
dengyihao's avatar
dengyihao 已提交
5505
}
5506 5507

void setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode) {
X
Xiaoyu Wang 已提交
5508 5509 5510
  pColumn->slotId = pColumnNode->slotId;
  pColumn->type = pColumnNode->node.resType.type;
  pColumn->bytes = pColumnNode->node.resType.bytes;
5511
  pColumn->precision = pColumnNode->node.resType.precision;
X
Xiaoyu Wang 已提交
5512
  pColumn->scale = pColumnNode->node.resType.scale;
5513
}