• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

taosdata / TDengine / #4997

20 Mar 2026 06:10AM UTC coverage: 71.739% (-0.3%) from 72.069%
#4997

push

travis-ci

web-flow
feat: add query phase tracking for SHOW QUERIES (#34706)

148 of 183 new or added lines in 10 files covered. (80.87%)

9273 existing lines in 172 files now uncovered.

244572 of 340921 relevant lines covered (71.74%)

133392941.95 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

66.58
/source/libs/executor/src/hashjoinoperator.c
1
/*
2
 * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
3
 *
4
 * This program is free software: you can use, redistribute, and/or modify
5
 * it under the terms of the GNU Affero General Public License, version 3
6
 * or later ("AGPL"), as published by the Free Software Foundation.
7
 *
8
 * This program is distributed in the hope that it will be useful, but WITHOUT
9
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10
 * FITNESS FOR A PARTICULAR PURPOSE.
11
 *
12
 * You should have received a copy of the GNU Affero General Public License
13
 * along with this program. If not, see <http://www.gnu.org/licenses/>.
14
 */
15

16
#include "executorInt.h"
17
#include "filter.h"
18
#include "function.h"
19
#include "operator.h"
20
#include "os.h"
21
#include "querynodes.h"
22
#include "querytask.h"
23
#include "tcompare.h"
24
#include "tdatablock.h"
25
#include "thash.h"
26
#include "tmsg.h"
27
#include "ttypes.h"
28
#include "hashjoin.h"
29
#include "functionMgt.h"
30

31

32
bool hJoinBlkReachThreshold(SHJoinOperatorInfo* pInfo, int64_t blkRows) {
2,333,298✔
33
  if (INT64_MAX == pInfo->ctx.limit || pInfo->pFinFilter != NULL) {
2,333,298✔
34
    return blkRows >= pInfo->blkThreshold;
2,333,298✔
35
  }
36
  
37
  return (pInfo->execInfo.resRows + blkRows) >= pInfo->ctx.limit;
×
38
}
39

40
int32_t hJoinHandleMidRemains(SHJoinOperatorInfo* pJoin, SHJoinCtx* pCtx) {
×
41
  TSWAP(pJoin->midBlk, pJoin->finBlk);
×
42

43
  pCtx->midRemains = false;
×
44

45
  return TSDB_CODE_SUCCESS;
×
46
}
47

48
int32_t hJoinCopyMergeMidBlk(SHJoinCtx* pCtx, SSDataBlock** ppMid, SSDataBlock** ppFin) {
×
49
  SSDataBlock* pLess = *ppMid;
×
50
  SSDataBlock* pMore = *ppFin;
×
51

52
/*
53
  if ((*ppMid)->info.rows < (*ppFin)->info.rows) {
54
    pLess = (*ppMid);
55
    pMore = (*ppFin);
56
  } else {
57
    pLess = (*ppFin);
58
    pMore = (*ppMid);
59
  }
60
*/
61

62
  int32_t totalRows = pMore->info.rows + pLess->info.rows;
×
63
  if (totalRows <= pMore->info.capacity) {
×
64
    HJ_ERR_RET(blockDataMerge(pMore, pLess));
×
65
    blockDataCleanup(pLess);
×
66
    pCtx->midRemains = false;
×
67
  } else {
68
    int32_t copyRows = pMore->info.capacity - pMore->info.rows;
×
69
    HJ_ERR_RET(blockDataMergeNRows(pMore, pLess, pLess->info.rows - copyRows, copyRows));
×
70
    blockDataShrinkNRows(pLess, copyRows);
×
71
    pCtx->midRemains = true;
×
72
  }
73

74
/*
75
  if (pMore != (*ppFin)) {
76
    TSWAP(*ppMid, *ppFin);
77
  }
78
*/
79

80
  return TSDB_CODE_SUCCESS;
×
81
}
82

83
int32_t hJoinSetImplFp(SHJoinOperatorInfo* pJoin) {
1,197,646✔
84
  switch (pJoin->joinType) {
1,197,646✔
85
    case JOIN_TYPE_INNER:
1,197,646✔
86
      pJoin->joinFp = hInnerJoinDo;
1,197,646✔
87
      break;
1,197,646✔
88
    case JOIN_TYPE_LEFT:
×
89
    case JOIN_TYPE_RIGHT: {
90
      switch (pJoin->subType) {
×
91
        case JOIN_STYPE_OUTER:          
×
92
          //pJoin->joinFp = hLeftJoinDo; TOOPEN
93
          break;
×
94
        default:
×
95
          break;
×
96
      }
97
      break;
×
98
    }      
99
    default:
×
100
      break;
×
101
  }
102

103
  return TSDB_CODE_SUCCESS;
1,197,646✔
104
}
105

106

107
int32_t hJoinLaunchPrimExpr(SSDataBlock* pBlock, SHJoinTableCtx* pTable, int32_t startIdx, int32_t endIdx) {
4,670,465✔
108
  if (NULL == pTable->primExpr) {
4,670,465✔
109
    return TSDB_CODE_SUCCESS;
4,670,465✔
110
  }
111

112
  SHJoinPrimExprCtx* pCtx = &pTable->primCtx;
×
113
  SColumnInfoData* pPrimIn = taosArrayGet(pBlock->pDataBlock, pTable->primCol->srcSlot);
×
114
  SColumnInfoData* pPrimOut = taosArrayGet(pBlock->pDataBlock, pTable->primCtx.targetSlotId);
×
115
  if (0 != pCtx->timezoneUnit) {
×
116
    for (int32_t i = startIdx; i <= endIdx; ++i) {
×
117
      ((int64_t*)pPrimOut->pData)[i] = ((int64_t*)pPrimIn->pData)[i] - (((int64_t*)pPrimIn->pData)[i] + pCtx->timezoneUnit) % pCtx->truncateUnit;
×
118
    }
119
  } else {
120
    for (int32_t i = startIdx; i <= endIdx; ++i) {
×
121
      ((int64_t*)pPrimOut->pData)[i] = ((int64_t*)pPrimIn->pData)[i] / pCtx->truncateUnit * pCtx->truncateUnit;
×
122
    }
123
  }
124

125
  return TSDB_CODE_SUCCESS;
×
126
}
127

128

129
static int64_t hJoinGetSingleKeyRowsNum(SBufRowInfo* pRow) {
×
130
  int64_t rows = 0;
×
131
  while (pRow) {
×
132
    rows++;
×
133
    pRow = pRow->next;
×
134
  }
135
  return rows;
×
136
}
137

138
static int64_t hJoinGetRowsNumOfKeyHash(SSHashObj* pHash) {
×
139
  SGroupData* pGroup = NULL;
×
140
  int32_t iter = 0;
×
141
  int64_t rowsNum = 0;
×
142
  
143
  while (NULL != (pGroup = tSimpleHashIterate(pHash, pGroup, &iter))) {
×
144
    int32_t* pKey = tSimpleHashGetKey(pGroup, NULL);
×
145
    int64_t rows = hJoinGetSingleKeyRowsNum(pGroup->rows);
×
146
    //qTrace("build_key:%d, rows:%" PRId64, *pKey, rows);
147
    rowsNum += rows;
×
148
  }
149

150
  return rowsNum;
×
151
}
152

153
static int32_t hJoinInitKeyColsInfo(SHJoinTableCtx* pTable, SNodeList* pList) {
2,395,292✔
154
  pTable->keyNum = LIST_LENGTH(pList);
2,395,292✔
155
  
156
  pTable->keyCols = taosMemoryMalloc(pTable->keyNum * sizeof(SHJoinColInfo));
2,395,292✔
157
  if (NULL == pTable->keyCols) {
2,395,292✔
158
    return terrno;
×
159
  }
160

161
  int64_t bufSize = 0;
2,395,292✔
162
  int32_t i = 0;
2,395,292✔
163
  SNode* pNode = NULL;
2,395,292✔
164
  FOREACH(pNode, pList) {
4,803,184✔
165
    SColumnNode* pColNode = (SColumnNode*)pNode;
2,407,892✔
166
    pTable->keyCols[i].srcSlot = pColNode->slotId;
2,407,892✔
167
    pTable->keyCols[i].vardata = IS_VAR_DATA_TYPE(pColNode->node.resType.type);
2,407,892✔
168
    pTable->keyCols[i].bytes = pColNode->node.resType.bytes;
2,407,892✔
169
    bufSize += pColNode->node.resType.bytes;
2,407,892✔
170
    ++i;
2,407,892✔
171
  }  
172

173
  if (pTable->keyNum > 1) {
2,395,292✔
174
    pTable->keyBuf = taosMemoryMalloc(bufSize);
12,600✔
175
    if (NULL == pTable->keyBuf) {
12,600✔
176
      return terrno;
×
177
    }
178
  }
179

180
  return TSDB_CODE_SUCCESS;
2,395,292✔
181
}
182

183
static void hJoinGetValColsNum(SNodeList* pList, int64_t blkId, int32_t* colNum) {
2,395,292✔
184
  *colNum = 0;
2,395,292✔
185
  
186
  SNode* pNode = NULL;
2,395,292✔
187
  FOREACH(pNode, pList) {
11,989,036✔
188
    STargetNode* pTarget = (STargetNode*)pNode;
9,593,744✔
189
    SColumnNode* pCol = (SColumnNode*)pTarget->pExpr;
9,593,744✔
190
    if (pCol->dataBlockId == blkId) {
9,593,744✔
191
      (*colNum)++;
4,796,872✔
192
    }
193
  }
194
}
2,395,292✔
195

196
static bool hJoinIsValColInKeyCols(int16_t slotId, int32_t keyNum, SHJoinColInfo* pKeys, int32_t* pKeyIdx) {
4,796,872✔
197
  for (int32_t i = 0; i < keyNum; ++i) {
9,623,660✔
198
    if (pKeys[i].srcSlot == slotId) {
4,828,360✔
199
      *pKeyIdx = i;
1,572✔
200
      return true;
1,572✔
201
    }
202
  }
203

204
  *pKeyIdx = -1;
4,795,300✔
205
  return false;
4,795,300✔
206
}
207

208
static int32_t hJoinInitValColsInfo(SHJoinTableCtx* pTable, SNodeList* pList) {
2,395,292✔
209
  hJoinGetValColsNum(pList, pTable->blkId, &pTable->valNum);
2,395,292✔
210
  if (pTable->valNum == 0) {
2,395,292✔
211
    return TSDB_CODE_SUCCESS;
×
212
  }
213
  
214
  pTable->valCols = taosMemoryMalloc(pTable->valNum * sizeof(SHJoinColInfo));
2,395,292✔
215
  if (NULL == pTable->valCols) {
2,395,292✔
216
    return terrno;
×
217
  }
218

219
  int32_t i = 0;
2,395,292✔
220
  int32_t colNum = 0;
2,395,292✔
221
  SNode* pNode = NULL;
2,395,292✔
222
  FOREACH(pNode, pList) {
11,989,036✔
223
    STargetNode* pTarget = (STargetNode*)pNode;
9,593,744✔
224
    SColumnNode* pColNode = (SColumnNode*)pTarget->pExpr;
9,593,744✔
225
    if (pColNode->dataBlockId == pTable->blkId) {
9,593,744✔
226
      if (!hJoinIsValColInKeyCols(pColNode->slotId, pTable->keyNum, pTable->keyCols, &pTable->valCols[i].keyColIdx)) {
4,796,872✔
227
        pTable->valColExist = true;
4,795,300✔
228
        colNum++;
4,795,300✔
229
      }
230
      pTable->valCols[i].srcSlot = pColNode->slotId;
4,796,872✔
231
      pTable->valCols[i].dstSlot = pTarget->slotId;
4,796,872✔
232
      pTable->valCols[i].vardata = IS_VAR_DATA_TYPE(pColNode->node.resType.type);
4,796,872✔
233
      if (pTable->valCols[i].vardata && !IS_HASH_JOIN_KEY_COL(pTable->valCols[i].keyColIdx)) {
4,796,872✔
234
        if (NULL == pTable->valVarCols) {
2,358✔
235
          pTable->valVarCols = taosArrayInit(pTable->valNum, sizeof(int32_t));
786✔
236
          if (NULL == pTable->valVarCols) {
786✔
237
            return terrno;
×
238
          }
239
        }
240
        if (NULL == taosArrayPush(pTable->valVarCols, &i)) {
4,716✔
241
          return terrno;
×
242
        }
243
      }
244
      pTable->valCols[i].bytes = pColNode->node.resType.bytes;
4,796,872✔
245
      if (!IS_HASH_JOIN_KEY_COL(pTable->valCols[i].keyColIdx) && !pTable->valCols[i].vardata) {
4,796,872✔
246
        pTable->valBufSize += pColNode->node.resType.bytes;
4,792,942✔
247
      }
248
      i++;
4,796,872✔
249
    }
250
  }
251

252
  pTable->valBitMapSize = BitmapLen(colNum);
2,395,292✔
253
  pTable->valBufSize += pTable->valBitMapSize;
2,395,292✔
254

255
  return TSDB_CODE_SUCCESS;
2,395,292✔
256
}
257

258
static int32_t hJoinInitPrimKeyInfo(SHJoinTableCtx* pTable, int32_t slotId) {
2,395,292✔
259
  pTable->primCol = taosMemoryMalloc(sizeof(SHJoinColMap));
2,395,292✔
260
  if (NULL == pTable->primCol) {
2,395,292✔
261
    return terrno;
×
262
  }
263

264
  pTable->primCol->srcSlot = slotId;
2,395,292✔
265

266
  return TSDB_CODE_SUCCESS;
2,395,292✔
267
}
268

269

270
static int32_t hJoinInitPrimExprCtx(SNode* pNode, SHJoinPrimExprCtx* pCtx, SHJoinTableCtx* pTable) {
2,395,292✔
271
  if (NULL == pNode) {
2,395,292✔
272
    pCtx->targetSlotId = pTable->primCol->srcSlot;
2,395,292✔
273
    return TSDB_CODE_SUCCESS;
2,395,292✔
274
  }
275
  
276
  if (QUERY_NODE_TARGET != nodeType(pNode)) {
×
277
    return TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR;
×
278
  }  
279

280
  STargetNode* pTarget = (STargetNode*)pNode;
×
281
  if (QUERY_NODE_FUNCTION != nodeType(pTarget->pExpr)) {
×
282
    return TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR;
×
283
  }
284

285
  SFunctionNode* pFunc = (SFunctionNode*)pTarget->pExpr;
×
286
  if (FUNCTION_TYPE_TIMETRUNCATE != pFunc->funcType) {
×
287
    return TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR;
×
288
  }
289

290
  if (4 != pFunc->pParameterList->length && 5 != pFunc->pParameterList->length) {
×
291
    return TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR;
×
292
  }
293

294
  SValueNode* pUnit = (SValueNode*)nodesListGetNode(pFunc->pParameterList, 1);
×
295
  SValueNode* pCurrTz = (5 == pFunc->pParameterList->length) ? (SValueNode*)nodesListGetNode(pFunc->pParameterList, 2) : NULL;
×
296
  SValueNode* pTimeZone = (5 == pFunc->pParameterList->length) ? (SValueNode*)nodesListGetNode(pFunc->pParameterList, 4) : (SValueNode*)nodesListGetNode(pFunc->pParameterList, 3);
×
297

298
  pCtx->truncateUnit = pUnit->typeData;
×
299
  if ((NULL == pCurrTz || 1 == pCurrTz->typeData) && pCtx->truncateUnit >= (86400 * TSDB_TICK_PER_SECOND(pFunc->node.resType.precision))) {
×
300
    pCtx->timezoneUnit = offsetFromTz(varDataVal(pTimeZone->datum.p), TSDB_TICK_PER_SECOND(pFunc->node.resType.precision));
×
301
  }
302

303
  pCtx->targetSlotId = pTarget->slotId;
×
304

305
  return TSDB_CODE_SUCCESS;
×
306
}
307

308

309
static int32_t hJoinInitTableInfo(SHJoinOperatorInfo* pJoin, SHashJoinPhysiNode* pJoinNode, SOperatorInfo** pDownstream, int32_t idx, SQueryStat* pStat) {
2,395,292✔
310
  SNodeList* pKeyList = NULL;
2,395,292✔
311
  SHJoinTableCtx* pTable = &pJoin->tbs[idx];
2,395,292✔
312
  pTable->downStream = pDownstream[idx];
2,395,292✔
313
  pTable->blkId = pDownstream[idx]->resultDataBlockId;
2,395,292✔
314
  if (0 == idx) {
2,395,292✔
315
    pKeyList = pJoinNode->pOnLeft;
1,197,646✔
316
    pTable->hasTimeRange = pJoinNode->timeRangeTarget & 0x1;
1,197,646✔
317
  } else {
318
    pKeyList = pJoinNode->pOnRight;
1,197,646✔
319
    pTable->hasTimeRange = pJoinNode->timeRangeTarget & 0x2;
1,197,646✔
320
  }
321

322
  HJ_ERR_RET(hJoinInitPrimKeyInfo(pTable, (0 == idx) ? pJoinNode->leftPrimSlotId : pJoinNode->rightPrimSlotId));
2,395,292✔
323
  
324
  int32_t code = hJoinInitKeyColsInfo(pTable, pKeyList);
2,395,292✔
325
  if (code) {
2,395,292✔
326
    return code;
×
327
  }
328
  code = hJoinInitValColsInfo(pTable, pJoinNode->pTargets);
2,395,292✔
329
  if (code) {
2,395,292✔
330
    return code;
×
331
  }
332

333
  TAOS_MEMCPY(&pTable->inputStat, pStat, sizeof(*pStat));
2,395,292✔
334

335
  HJ_ERR_RET(hJoinInitPrimExprCtx(pTable->primExpr, &pTable->primCtx, pTable));
2,395,292✔
336

337
  return TSDB_CODE_SUCCESS;
2,395,292✔
338
}
339

340
static void hJoinSetBuildAndProbeTable(SHJoinOperatorInfo* pInfo, SHashJoinPhysiNode* pJoinNode) {
1,197,646✔
341
  int32_t buildIdx = 0;
1,197,646✔
342
  int32_t probeIdx = 1;
1,197,646✔
343

344
  pInfo->joinType = pJoinNode->joinType;
1,197,646✔
345
  pInfo->subType = pJoinNode->subType;
1,197,646✔
346
  
347
  switch (pInfo->joinType) {
1,197,646✔
348
    case JOIN_TYPE_INNER:
1,197,646✔
349
    case JOIN_TYPE_FULL:
350
      if (pInfo->tbs[0].inputStat.inputRowNum <= pInfo->tbs[1].inputStat.inputRowNum) {
1,197,646✔
351
        buildIdx = 0;
1,197,646✔
352
        probeIdx = 1;
1,197,646✔
353
      } else {
354
        buildIdx = 1;
×
355
        probeIdx = 0;
×
356
      }
357
      break;
1,197,646✔
358
    case JOIN_TYPE_LEFT:
×
359
      buildIdx = 1;
×
360
      probeIdx = 0;
×
361
      break;
×
362
    case JOIN_TYPE_RIGHT:
×
363
      buildIdx = 0;
×
364
      probeIdx = 1;
×
365
      break;
×
366
    default:
×
367
      break;
×
368
  } 
369
  
370
  pInfo->pBuild = &pInfo->tbs[buildIdx];
1,197,646✔
371
  pInfo->pProbe = &pInfo->tbs[probeIdx];
1,197,646✔
372
  
373
  pInfo->pBuild->downStreamIdx = buildIdx;
1,197,646✔
374
  pInfo->pProbe->downStreamIdx = probeIdx;
1,197,646✔
375

376
  if (0 == buildIdx) {
1,197,646✔
377
    pInfo->pBuild->primExpr = pJoinNode->leftPrimExpr;
1,197,646✔
378
    pInfo->pProbe->primExpr = pJoinNode->rightPrimExpr;
1,197,646✔
379
  } else {
380
    pInfo->pBuild->primExpr = pJoinNode->rightPrimExpr;
×
381
    pInfo->pProbe->primExpr = pJoinNode->leftPrimExpr;
×
382
  }
383
}
1,197,646✔
384

385
static int32_t hJoinBuildResColsMap(SHJoinOperatorInfo* pInfo, SHashJoinPhysiNode* pJoinNode) {
1,197,646✔
386
  pInfo->pResColNum = pJoinNode->pTargets->length;
1,197,646✔
387
  pInfo->pResColMap = taosMemoryCalloc(pJoinNode->pTargets->length, sizeof(int8_t));
1,197,646✔
388
  if (NULL == pInfo->pResColMap) {
1,197,646✔
389
    return terrno;
×
390
  }
391
  
392
  SNode* pNode = NULL;
1,197,646✔
393
  int32_t i = 0;
1,197,646✔
394
  FOREACH(pNode, pJoinNode->pTargets) {
5,994,518✔
395
    STargetNode* pTarget = (STargetNode*)pNode;
4,796,872✔
396
    SColumnNode* pCol = (SColumnNode*)pTarget->pExpr;
4,796,872✔
397
    if (pCol->dataBlockId == pInfo->pBuild->blkId) {
4,796,872✔
398
      pInfo->pResColMap[i] = 1;
2,398,436✔
399
    }
400
    
401
    i++;
4,796,872✔
402
  }
403

404
  return TSDB_CODE_SUCCESS;
1,197,646✔
405
}
406

407

408
static FORCE_INLINE int32_t hJoinAddPageToBufs(SArray* pRowBufs) {
409
  SBufPageInfo page;
×
410
  page.pageSize = HASH_JOIN_DEFAULT_PAGE_SIZE;
1,198,462✔
411
  page.offset = 0;
1,198,462✔
412
  page.data = taosMemoryMalloc(page.pageSize);
1,198,462✔
413
  if (NULL == page.data) {
1,198,462✔
414
    return terrno;
×
415
  }
416

417
  if (NULL == taosArrayPush(pRowBufs, &page)) {
1,198,462✔
418
    return terrno;
×
419
  }
420
  return TSDB_CODE_SUCCESS;
1,198,462✔
421
}
422

423
static int32_t hJoinInitBufPages(SHJoinOperatorInfo* pInfo) {
1,198,462✔
424
  pInfo->pRowBufs = taosArrayInit(32, sizeof(SBufPageInfo));
1,198,462✔
425
  if (NULL == pInfo->pRowBufs) {
1,198,462✔
426
    return terrno;
×
427
  }
428

429
  return hJoinAddPageToBufs(pInfo->pRowBufs);
2,396,924✔
430
}
431

432
static void hJoinFreeTableInfo(SHJoinTableCtx* pTable) {
2,395,292✔
433
  taosMemoryFreeClear(pTable->keyCols);
2,395,292✔
434
  taosMemoryFreeClear(pTable->keyBuf);
2,395,292✔
435
  taosMemoryFreeClear(pTable->valCols);
2,395,292✔
436
  taosArrayDestroy(pTable->valVarCols);
2,395,292✔
437
  taosMemoryFree(pTable->primCol);
2,395,292✔
438
}
2,395,292✔
439

440
static void hJoinFreeBufPage(void* param) {
1,198,462✔
441
  SBufPageInfo* pInfo = (SBufPageInfo*)param;
1,198,462✔
442
  taosMemoryFree(pInfo->data);
1,198,462✔
443
}
1,198,462✔
444

445
static void hJoinDestroyKeyHash(SSHashObj** ppHash) {
2,395,836✔
446
  if (NULL == ppHash || NULL == (*ppHash)) {
2,395,836✔
447
    return;
1,197,374✔
448
  }
449

450
  void*   pIte = NULL;
1,198,462✔
451
  int32_t iter = 0;
1,198,462✔
452
  while ((pIte = tSimpleHashIterate(*ppHash, pIte, &iter)) != NULL) {
5,828,372✔
453
    SGroupData* pGroup = pIte;
4,629,910✔
454
    SBufRowInfo* pRow = pGroup->rows;
4,629,910✔
455
    SBufRowInfo* pNext = NULL;
4,629,910✔
456
    while (pRow) {
9,377,852✔
457
      pNext = pRow->next;
4,747,942✔
458
      taosMemoryFree(pRow);
4,747,942✔
459
      pRow = pNext;
4,747,942✔
460
    }
461
  }
462

463
  tSimpleHashCleanup(*ppHash);
1,198,462✔
464
  *ppHash = NULL;
1,198,462✔
465
}
466

467
static FORCE_INLINE int32_t hJoinRetrieveColDataFromRowBufs(SArray* pRowBufs, SBufRowInfo* pRow, char** ppData) {
468
  *ppData = NULL;
7,630,842✔
469
  
470
  if ((uint16_t)-1 == pRow->pageId) {
3,815,421✔
471
    return TSDB_CODE_SUCCESS;
×
472
  }
473
  SBufPageInfo *pPage = taosArrayGet(pRowBufs, pRow->pageId);
3,815,421✔
474
  if (NULL == pPage) {
3,815,421✔
475
    qError("fail to get %d page, total:%d", pRow->pageId, (int32_t)taosArrayGetSize(pRowBufs));
×
476
    QRY_ERR_RET(TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR);
×
477
  }
478
  
479
  *ppData = pPage->data + pRow->offset;
3,815,421✔
480

481
  return TSDB_CODE_SUCCESS;
3,815,421✔
482
}
483

484
static int32_t hJoinCopyResRowsToBlock(SHJoinOperatorInfo* pJoin, int32_t rowNum, SBufRowInfo* pStart, SSDataBlock* pRes) {
296,652✔
485
  SHJoinTableCtx* pBuild = pJoin->pBuild;
296,652✔
486
  SHJoinTableCtx* pProbe = pJoin->pProbe;
296,652✔
487
  int32_t buildIdx = 0, buildValIdx = 0;
296,652✔
488
  int32_t probeIdx = 0;
296,652✔
489
  SBufRowInfo* pRow = pStart;
296,652✔
490
  int32_t code = 0;
296,652✔
491
  char* pData = NULL;
296,652✔
492

493
  for (int32_t r = 0; r < rowNum; ++r) {
4,112,073✔
494
    HJ_ERR_RET(hJoinRetrieveColDataFromRowBufs(pJoin->pRowBufs, pRow, &pData));
7,630,842✔
495
    
496
    char* pValData = pData + pBuild->valBitMapSize;
3,815,421✔
497
    char* pKeyData = pProbe->keyData;
3,815,421✔
498
    buildIdx = buildValIdx = probeIdx = 0;
3,815,421✔
499
    for (int32_t i = 0; i < pJoin->pResColNum; ++i) {
19,095,969✔
500
      if (pJoin->pResColMap[i]) {
15,280,548✔
501
        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, pBuild->valCols[buildIdx].dstSlot);
7,640,274✔
502
        if (IS_HASH_JOIN_KEY_COL(pBuild->valCols[buildIdx].keyColIdx)) {
7,640,274✔
503
          int32_t bufOffset = pProbe->keyCols[pBuild->valCols[buildIdx].keyColIdx].bufOffset;
2,358✔
504
          code = colDataSetVal(pDst, pRes->info.rows + r, pKeyData + bufOffset, false);
2,358✔
505
          if (code) {
2,358✔
506
            return code;
×
507
          }
508
        } else {
509
          if (BMIsNull(pData, buildValIdx)) {
7,637,916✔
510
            code = colDataSetVal(pDst, pRes->info.rows + r, NULL, true);
4,716✔
511
            if (code) {
4,716✔
512
              return code;
×
513
            }
514
          } else {
515
            code = colDataSetVal(pDst, pRes->info.rows + r, pValData, false);
7,633,200✔
516
            if (code) {
7,633,200✔
517
              return code;
×
518
            }
519
            pValData += pBuild->valCols[buildIdx].vardata ? varDataTLen(pValData) : pBuild->valCols[buildIdx].bytes;
7,633,200✔
520
          }
521
          buildValIdx++;
7,637,916✔
522
        }
523
        buildIdx++;
7,640,274✔
524
      } else if (0 == r) {
7,640,274✔
525
        SColumnInfoData* pSrc = taosArrayGet(pJoin->ctx.pProbeData->pDataBlock, pProbe->valCols[probeIdx].srcSlot);
602,736✔
526
        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, pProbe->valCols[probeIdx].dstSlot);
602,736✔
527

528
        if (colDataIsNull_s(pSrc, pJoin->ctx.probeStartIdx)) {
1,205,472✔
529
          code = colDataCopyNItems(pDst, pRes->info.rows, NULL, rowNum, true);
4,716✔
530
        } else {
531
          code = colDataCopyNItems(pDst, pRes->info.rows, colDataGetData(pSrc, pJoin->ctx.probeStartIdx), rowNum, false);
598,020✔
532
        }
533
        if (code) {
602,736✔
534
          return code;
×
535
        }
536
        probeIdx++;
602,736✔
537
      }
538
    }
539
    pRow = pRow->next;
3,815,421✔
540
  }
541

542
  return TSDB_CODE_SUCCESS;
296,652✔
543
}
544

545
int32_t hJoinCopyNMatchRowsToBlock(SHJoinOperatorInfo* pJoin, SSDataBlock* pRes, int32_t startIdx, int32_t rows) {
×
546
  SHJoinTableCtx* pBuild = pJoin->pBuild;
×
547
  SHJoinTableCtx* pProbe = pJoin->pProbe;
×
548
  int32_t buildIdx = 0;
×
549
  int32_t probeIdx = 0;
×
550
  int32_t code = 0;
×
551

552
  for (int32_t i = 0; i < pJoin->pResColNum; ++i) {
×
553
    if (pJoin->pResColMap[i]) {
×
554
      SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, pBuild->valCols[buildIdx].dstSlot);
×
555
      colDataSetNItemsNull(pDst, pRes->info.rows, rows);
×
556

557
      buildIdx++;
×
558
    } else {
559
      SColumnInfoData* pSrc = taosArrayGet(pJoin->ctx.pProbeData->pDataBlock, pProbe->valCols[probeIdx].srcSlot);
×
560
      SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, pProbe->valCols[probeIdx].dstSlot);
×
561

562
      QRY_ERR_RET(colDataAssignNRows(pDst, pRes->info.rows, pSrc, startIdx, rows));
×
563

564
      probeIdx++;
×
565
    }
566
  }
567

568
  pRes->info.rows += rows;
×
569

570
  return TSDB_CODE_SUCCESS;
×
571
}
572

573

574

575
void hJoinAppendResToBlock(struct SOperatorInfo* pOperator, SSDataBlock* pRes, bool* allFetched) {
296,652✔
576
  SHJoinOperatorInfo* pJoin = pOperator->info;
296,652✔
577
  SHJoinCtx* pCtx = &pJoin->ctx;
296,652✔
578
  SBufRowInfo* pStart = pCtx->pBuildRow;
296,652✔
579
  int32_t rowNum = 0;
296,652✔
580
  int32_t resNum = pRes->info.rows;
296,652✔
581
  
582
  while (pCtx->pBuildRow && (resNum < pRes->info.capacity)) {
4,112,073✔
583
    rowNum++;
3,815,421✔
584
    resNum++;
3,815,421✔
585
    pCtx->pBuildRow = pCtx->pBuildRow->next;
3,815,421✔
586
  }
587

588
  pJoin->execInfo.resRows += rowNum;
296,652✔
589

590
  int32_t code = hJoinCopyResRowsToBlock(pJoin, rowNum, pStart, pRes);
296,652✔
591
  if (code) {
296,652✔
592
    pOperator->pTaskInfo->code = code;
×
593
    T_LONG_JMP(pOperator->pTaskInfo->env, code);
×
594
  }
595

596
  pRes->info.rows = resNum;
296,652✔
597
  *allFetched = pCtx->pBuildRow ? false : true;
296,652✔
598
}
296,652✔
599

600

601
bool hJoinCopyKeyColsDataToBuf(SHJoinTableCtx* pTable, int32_t rowIdx, size_t *pBufLen) {
7,290,045✔
602
  char *pData = NULL;
7,290,045✔
603
  size_t bufLen = 0;
7,290,045✔
604
  
605
  if (1 == pTable->keyNum) {
7,290,045✔
606
    if (colDataIsNull_s(pTable->keyCols[0].colData, rowIdx)) {
14,508,862✔
607
      return true;
4,945✔
608
    }
609
    if (pTable->keyCols[0].vardata) {
7,249,486✔
610
      pData = pTable->keyCols[0].data + pTable->keyCols[0].offset[rowIdx];
1,743,807✔
611
      bufLen = varDataTLen(pData);
1,743,807✔
612
    } else {
613
      pData = pTable->keyCols[0].data + pTable->keyCols[0].bytes * rowIdx;
5,505,679✔
614
      bufLen = pTable->keyCols[0].bytes;
5,505,679✔
615
    }
616
    pTable->keyData = pData;
7,249,486✔
617
  } else {
618
    for (int32_t i = 0; i < pTable->keyNum; ++i) {
106,842✔
619
      if (colDataIsNull_s(pTable->keyCols[i].colData, rowIdx)) {
142,456✔
620
        return true;
×
621
      }
622
      
623
      pTable->keyCols[i].bufOffset = bufLen;
71,228✔
624
      
625
      if (pTable->keyCols[i].vardata) {
71,228✔
626
        pData = pTable->keyCols[i].data + pTable->keyCols[i].offset[rowIdx];
28,176✔
627
        TAOS_MEMCPY(pTable->keyBuf + bufLen, pData, varDataTLen(pData));
28,176✔
628
        bufLen += varDataTLen(pData);
28,176✔
629
      } else {
630
        pData = pTable->keyCols[i].data + pTable->keyCols[i].bytes * rowIdx;
43,052✔
631
        TAOS_MEMCPY(pTable->keyBuf + bufLen, pData, pTable->keyCols[i].bytes);
43,052✔
632
        bufLen += pTable->keyCols[i].bytes;
43,052✔
633
      }
634
    }
635
    pTable->keyData = pTable->keyBuf;
35,614✔
636
  }
637

638
  if (pBufLen) {
7,285,100✔
639
    *pBufLen = bufLen;
7,285,100✔
640
  }
641

642
  return false;
7,285,100✔
643
}
644

645
static int32_t hJoinSetKeyColsData(SSDataBlock* pBlock, SHJoinTableCtx* pTable) {
4,670,465✔
646
  for (int32_t i = 0; i < pTable->keyNum; ++i) {
9,359,483✔
647
    SColumnInfoData* pCol = taosArrayGet(pBlock->pDataBlock, pTable->keyCols[i].srcSlot);
4,689,018✔
648
    if (NULL == pCol) {
4,689,018✔
649
      qError("fail to get %d col, total:%d", pTable->keyCols[i].srcSlot, (int32_t)taosArrayGetSize(pBlock->pDataBlock));
×
650
      QRY_ERR_RET(TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR);
×
651
    }
652
    if (pTable->keyCols[i].vardata != IS_VAR_DATA_TYPE(pCol->info.type))  {
4,689,018✔
653
      qError("column type mismatch, idx:%d, slotId:%d, type:%d, vardata:%d", i, pTable->keyCols[i].srcSlot, pCol->info.type, pTable->keyCols[i].vardata);
×
654
      return TSDB_CODE_INVALID_PARA;
×
655
    }
656
    if (pTable->keyCols[i].bytes != pCol->info.bytes)  {
4,689,018✔
657
      qError("column bytes mismatch, idx:%d, slotId:%d, bytes:%d, %d", i, pTable->keyCols[i].srcSlot, pCol->info.bytes, pTable->keyCols[i].bytes);
×
658
      return TSDB_CODE_INVALID_PARA;
×
659
    }
660
    pTable->keyCols[i].data = pCol->pData;
4,689,018✔
661
    if (pTable->keyCols[i].vardata) {
4,689,018✔
662
      pTable->keyCols[i].offset = pCol->varmeta.offset;
1,175,740✔
663
    }
664
    pTable->keyCols[i].colData = pCol;
4,689,018✔
665
  }
666

667
  return TSDB_CODE_SUCCESS;
4,670,465✔
668
}
669

670
static int32_t hJoinSetValColsData(SSDataBlock* pBlock, SHJoinTableCtx* pTable) {
7,081,240✔
671
  if (!pTable->valColExist) {
7,081,240✔
672
    return TSDB_CODE_SUCCESS;
×
673
  }
674
  for (int32_t i = 0; i < pTable->valNum; ++i) {
21,257,868✔
675
    if (IS_HASH_JOIN_KEY_COL(pTable->valCols[i].keyColIdx)) {
14,176,628✔
676
      continue;
3,930✔
677
    }
678
    SColumnInfoData* pCol = taosArrayGet(pBlock->pDataBlock, pTable->valCols[i].srcSlot);
14,172,698✔
679
    if (NULL == pCol) {
14,172,698✔
680
      qError("fail to get %d col, total:%d", pTable->valCols[i].srcSlot, (int32_t)taosArrayGetSize(pBlock->pDataBlock));
×
681
      QRY_ERR_RET(TSDB_CODE_QRY_EXECUTOR_INTERNAL_ERROR);
×
682
    }
683
    if (pTable->valCols[i].vardata != IS_VAR_DATA_TYPE(pCol->info.type))  {
14,172,698✔
684
      qError("column type mismatch, idx:%d, slotId:%d, type:%d, vardata:%d", i, pTable->valCols[i].srcSlot, pCol->info.type, pTable->valCols[i].vardata);
×
685
      return TSDB_CODE_INVALID_PARA;
×
686
    }
687
    if (pTable->valCols[i].bytes != pCol->info.bytes)  {
14,172,698✔
688
      qError("column bytes mismatch, idx:%d, slotId:%d, bytes:%d, %d", i, pTable->valCols[i].srcSlot, pCol->info.bytes, pTable->valCols[i].bytes);
×
689
      return TSDB_CODE_INVALID_PARA;
×
690
    }
691
    if (!pTable->valCols[i].vardata) {
14,172,698✔
692
      pTable->valCols[i].bitMap = pCol->nullbitmap;
14,167,982✔
693
    }
694
    pTable->valCols[i].data = pCol->pData;
14,172,698✔
695
    if (pTable->valCols[i].vardata) {
14,172,698✔
696
      pTable->valCols[i].offset = pCol->varmeta.offset;
4,716✔
697
    }
698
  }
699

700
  return TSDB_CODE_SUCCESS;
7,081,240✔
701
}
702

703

704

705
static FORCE_INLINE void hJoinCopyValColsDataToBuf(SHJoinTableCtx* pTable, int32_t rowIdx) {
706
  if (!pTable->valColExist) {
4,747,942✔
707
    return;
×
708
  }
709

710
  char *pData = NULL;
4,747,942✔
711
  size_t bufLen = pTable->valBitMapSize;
4,747,942✔
712
  TAOS_MEMSET(pTable->valData, 0, pTable->valBitMapSize);
4,747,942✔
713
  for (int32_t i = 0, m = 0; i < pTable->valNum; ++i) {
14,253,258✔
714
    if (IS_HASH_JOIN_KEY_COL(pTable->valCols[i].keyColIdx)) {
9,505,316✔
715
      continue;
2,358✔
716
    }
717
    if (pTable->valCols[i].vardata) {
9,502,958✔
718
      if (-1 == pTable->valCols[i].offset[rowIdx]) {
3,537✔
719
        colDataSetNull_f(pTable->valData, m);
3,537✔
720
      } else {
721
        pData = pTable->valCols[i].data + pTable->valCols[i].offset[rowIdx];
×
722
        TAOS_MEMCPY(pTable->valData + bufLen, pData, varDataTLen(pData));
×
723
        bufLen += varDataTLen(pData);
×
724
      }
725
    } else {
726
      if (BMIsNull(pTable->valCols[i].bitMap, rowIdx)) {
9,499,421✔
727
        colDataSetNull_f(pTable->valData, m);
1,179✔
728
      } else {
729
        pData = pTable->valCols[i].data + pTable->valCols[i].bytes * rowIdx;
9,498,242✔
730
        TAOS_MEMCPY(pTable->valData + bufLen, pData, pTable->valCols[i].bytes);
9,498,242✔
731
        bufLen += pTable->valCols[i].bytes;
9,498,242✔
732
      }
733
    }
734
    m++;
9,502,958✔
735
  }
736
}
737

738

739
static FORCE_INLINE int32_t hJoinGetValBufFromPages(SArray* pPages, int32_t bufSize, char** pBuf, SBufRowInfo* pRow) {
740
  if (0 == bufSize) {
4,747,942✔
741
    pRow->pageId = -1;
×
742
    return TSDB_CODE_SUCCESS;
×
743
  }
744

745
  if (bufSize > HASH_JOIN_DEFAULT_PAGE_SIZE) {
4,747,942✔
746
    qError("invalid join value buf size:%d", bufSize);
×
747
    return TSDB_CODE_INVALID_PARA;
×
748
  }
749
  
750
  do {
×
751
    SBufPageInfo* page = taosArrayGetLast(pPages);
4,747,942✔
752
    if ((page->pageSize - page->offset) >= bufSize) {
4,747,942✔
753
      *pBuf = page->data + page->offset;
4,747,942✔
754
      pRow->pageId = taosArrayGetSize(pPages) - 1;
4,747,942✔
755
      pRow->offset = page->offset;
4,747,942✔
756
      page->offset += bufSize;
4,747,942✔
757
      return TSDB_CODE_SUCCESS;
4,747,942✔
758
    }
759

760
    int32_t code = hJoinAddPageToBufs(pPages);
×
761
    if (code) {
×
762
      return code;
×
763
    }
764
  } while (true);
765
}
766

767
static FORCE_INLINE int32_t hJoinGetValBufSize(SHJoinTableCtx* pTable, int32_t rowIdx) {
768
  if (NULL == pTable->valVarCols) {
4,747,942✔
769
    return pTable->valBufSize;
4,746,763✔
770
  }
771

772
  int32_t* varColIdx = NULL;
1,179✔
773
  int32_t bufLen = pTable->valBufSize;
1,179✔
774
  int32_t varColNum = taosArrayGetSize(pTable->valVarCols);
1,179✔
775
  for (int32_t i = 0; i < varColNum; ++i) {
4,716✔
776
    varColIdx = taosArrayGet(pTable->valVarCols, i);
3,537✔
777
    if (-1 == pTable->valCols[*varColIdx].offset[rowIdx]) {
3,537✔
778
      continue;
3,537✔
779
    }
780
    char* pData = pTable->valCols[*varColIdx].data + pTable->valCols[*varColIdx].offset[rowIdx];
×
781
    bufLen += varDataTLen(pData);
×
782
  }
783

784
  return bufLen;
1,179✔
785
}
786

787

788
static int32_t hJoinAddRowToHashImpl(SHJoinOperatorInfo* pJoin, SGroupData* pGroup, SHJoinTableCtx* pTable, size_t keyLen, int32_t rowIdx) {
4,747,942✔
789
  SGroupData group = {0};
4,747,942✔
790
  SBufRowInfo* pRow = NULL;
4,747,942✔
791

792
  if (NULL == pGroup) {
4,747,942✔
793
    group.rows = taosMemoryMalloc(sizeof(SBufRowInfo));
4,629,910✔
794
    if (NULL == group.rows) {
4,629,910✔
795
      return terrno;
×
796
    }
797
    pRow = group.rows;
4,629,910✔
798
  } else {
799
    pRow = taosMemoryMalloc(sizeof(SBufRowInfo));
118,032✔
800
    if (NULL == pRow) {
118,032✔
801
      return terrno;
×
802
    }
803
  }
804

805
  int32_t code = hJoinGetValBufFromPages(pJoin->pRowBufs, hJoinGetValBufSize(pTable, rowIdx), &pTable->valData, pRow);
9,495,884✔
806
  if (code) {
4,747,942✔
807
    taosMemoryFree(pRow);
×
808
    return code;
×
809
  }
810

811
  if (NULL == pGroup) {
4,747,942✔
812
    pRow->next = NULL;
4,629,910✔
813
    if (tSimpleHashPut(pJoin->pKeyHash, pTable->keyData, keyLen, &group, sizeof(group))) {
4,629,910✔
814
      taosMemoryFree(pRow);
×
815
      return terrno;
×
816
    }
817
  } else {
818
    pRow->next = pGroup->rows;
118,032✔
819
    pGroup->rows = pRow;
118,032✔
820
  }
821

822
  return TSDB_CODE_SUCCESS;
4,747,942✔
823
}
824

825
static int32_t hJoinAddRowToHash(SHJoinOperatorInfo* pJoin, SSDataBlock* pBlock, size_t keyLen, int32_t rowIdx) {
4,747,942✔
826
  SHJoinTableCtx* pBuild = pJoin->pBuild;
4,747,942✔
827
  int32_t code = hJoinSetValColsData(pBlock, pBuild);
4,747,942✔
828
  if (code) {
4,747,942✔
829
    return code;
×
830
  }
831

832
  SGroupData* pGroup = tSimpleHashGet(pJoin->pKeyHash, pBuild->keyData, keyLen);
4,747,942✔
833
  code = hJoinAddRowToHashImpl(pJoin, pGroup, pBuild, keyLen, rowIdx);
4,747,942✔
834
  if (code) {
4,747,942✔
835
    return code;
×
836
  }
837
  
838
  hJoinCopyValColsDataToBuf(pBuild, rowIdx);
839

840
  return TSDB_CODE_SUCCESS;
4,747,942✔
841
}
842

843
static bool hJoinFilterTimeRange(SSDataBlock* pBlock, STimeWindow* pRange, int32_t primSlot, int32_t* startIdx, int32_t* endIdx) {
×
844
  SColumnInfoData* pCol = taosArrayGet(pBlock->pDataBlock, primSlot);
×
845
  if (NULL == pCol) {
×
846
    qError("hash join can't get prim col, slot:%d, slotNum:%d", primSlot, (int32_t)taosArrayGetSize(pBlock->pDataBlock));
×
847
    return false;
×
848
  }
849

850
  TSKEY skey = *(TSKEY*)colDataGetData(pCol, 0);
×
851
  TSKEY ekey = *(TSKEY*)colDataGetData(pCol, (pBlock->info.rows - 1));
×
852

853
  if (ekey < pRange->skey || skey > pRange->ekey) {
×
854
    return false;
×
855
  }
856

857
  if (skey >= pRange->skey && ekey <= pRange->ekey) {
×
858
    *startIdx = 0;
×
859
    *endIdx = pBlock->info.rows - 1;
×
860
    return true;
×
861
  }
862

863
  if (skey < pRange->skey && ekey > pRange->ekey) {
×
864
    TSKEY *pStart = (TSKEY*)taosbsearch(&pRange->skey, pCol->pData, pBlock->info.rows, sizeof(TSKEY), compareInt64Val, TD_GE);
×
865
    TSKEY *pEnd = (TSKEY*)taosbsearch(&pRange->ekey, pCol->pData, pBlock->info.rows, sizeof(TSKEY), compareInt64Val, TD_LE);
×
866
    *startIdx = ((uint64_t)pStart - (uint64_t)pCol->pData) / sizeof(int64_t);
×
867
    *endIdx = ((uint64_t)pEnd - (uint64_t)pCol->pData) / sizeof(int64_t);
×
868
    return true;
×
869
  }
870

871
  if (skey >= pRange->skey) {
×
872
    TSKEY *pEnd = (TSKEY*)taosbsearch(&pRange->ekey, pCol->pData, pBlock->info.rows, sizeof(TSKEY), compareInt64Val, TD_LE);
×
873
    *startIdx = 0;
×
874
    *endIdx = ((uint64_t)pEnd - (uint64_t)pCol->pData) / sizeof(int64_t);
×
875
    return true;
×
876
  }
877

878
  TSKEY *pStart = (TSKEY*)taosbsearch(&pRange->skey, pCol->pData, pBlock->info.rows, sizeof(TSKEY), compareInt64Val, TD_GE);
×
879
  *startIdx = ((uint64_t)pStart - (uint64_t)pCol->pData) / sizeof(int64_t);
×
880
  *endIdx = pBlock->info.rows - 1;
×
881
  
882
  return true;
×
883
}
884

885
static int32_t hJoinAddBlockRowsToHash(SSDataBlock* pBlock, SHJoinOperatorInfo* pJoin) {
2,337,167✔
886
  SHJoinTableCtx* pBuild = pJoin->pBuild;
2,337,167✔
887
  int32_t startIdx = 0, endIdx = pBlock->info.rows - 1;
2,337,167✔
888
  if (pBuild->hasTimeRange && !hJoinFilterTimeRange(pBlock, &pJoin->tblTimeRange, pBuild->primCol->srcSlot, &startIdx, &endIdx)) {
2,337,167✔
889
    return TSDB_CODE_SUCCESS;
×
890
  }
891

892
  HJ_ERR_RET(hJoinLaunchPrimExpr(pBlock, pBuild, startIdx, endIdx));
2,337,167✔
893

894
  int32_t code = hJoinSetKeyColsData(pBlock, pBuild);
2,337,167✔
895
  if (code) {
2,337,167✔
896
    return code;
×
897
  }
898

899
  size_t bufLen = 0;
2,337,167✔
900
  for (int32_t i = startIdx; i <= endIdx; ++i) {
7,088,994✔
901
    if (hJoinCopyKeyColsDataToBuf(pBuild, i, &bufLen)) {
4,751,827✔
902
      continue;
3,885✔
903
    }
904
    code = hJoinAddRowToHash(pJoin, pBlock, bufLen, i);
4,747,942✔
905
    if (code) {
4,747,942✔
906
      return code;
×
907
    }
908
  }
909

910
  return code;
2,337,167✔
911
}
912

913
static int32_t hJoinBuildHash(struct SOperatorInfo* pOperator, bool* queryDone) {
1,198,190✔
914
  SHJoinOperatorInfo* pJoin = pOperator->info;
1,198,190✔
915
  SSDataBlock*        pBlock = NULL;
1,198,190✔
916
  int32_t             code = TSDB_CODE_SUCCESS;
1,198,190✔
917

918
  while (true) {
919
    pBlock = getNextBlockFromDownstream(pOperator, pJoin->pBuild->downStreamIdx);
3,535,357✔
920
    if (NULL == pBlock) {
3,535,357✔
921
      break;
1,198,190✔
922
    }
923

924
    pJoin->execInfo.buildBlkNum++;
2,337,167✔
925
    pJoin->execInfo.buildBlkRows += pBlock->info.rows;
2,337,167✔
926

927
    code = hJoinAddBlockRowsToHash(pBlock, pJoin);
2,337,167✔
928
    if (code) {
2,337,167✔
929
      return code;
×
930
    }
931
  }
932

933
  if (IS_INNER_NONE_JOIN(pJoin->joinType, pJoin->subType) && tSimpleHashGetSize(pJoin->pKeyHash) <= 0) {
1,198,190✔
934
    hJoinSetDone(pOperator);
18,045✔
935
    *queryDone = true;
18,045✔
936
  }
937
  
938
  //qTrace("build table rows:%" PRId64, hJoinGetRowsNumOfKeyHash(pJoin->pKeyHash));
939

940
  return TSDB_CODE_SUCCESS;
1,198,190✔
941
}
942

943
static int32_t hJoinPrepareStart(struct SOperatorInfo* pOperator, SSDataBlock* pBlock) {
2,333,298✔
944
  SHJoinOperatorInfo* pJoin = pOperator->info;
2,333,298✔
945
  SHJoinTableCtx* pProbe = pJoin->pProbe;
2,333,298✔
946
  int32_t startIdx = 0, endIdx = pBlock->info.rows - 1;
2,333,298✔
947
  if (pProbe->hasTimeRange && !hJoinFilterTimeRange(pBlock, &pJoin->tblTimeRange, pProbe->primCol->srcSlot, &startIdx, &endIdx)) {
2,333,298✔
948
    if (!IS_INNER_NONE_JOIN(pJoin->joinType, pJoin->subType)) {
×
949
      pJoin->ctx.probeEndIdx = -1;
×
950
      pJoin->ctx.probePostIdx = 0;
×
951
      pJoin->ctx.pProbeData = pBlock;
×
952
      pJoin->ctx.rowRemains = true;
×
953
      pJoin->ctx.probePhase = E_JOIN_PHASE_POST;
×
954
      
955
      HJ_ERR_RET((*pJoin->joinFp)(pOperator));
×
956
    }
957
    
958
    return TSDB_CODE_SUCCESS;
×
959
  }
960

961
  HJ_ERR_RET(hJoinLaunchPrimExpr(pBlock, pProbe, startIdx, endIdx));
2,333,298✔
962

963
  int32_t code = hJoinSetKeyColsData(pBlock, pProbe);
2,333,298✔
964
  if (code) {
2,333,298✔
965
    return code;
×
966
  }
967
  code = hJoinSetValColsData(pBlock, pProbe);
2,333,298✔
968
  if (code) {
2,333,298✔
969
    return code;
×
970
  }
971

972
  pJoin->ctx.probeStartIdx = startIdx;
2,333,298✔
973
  pJoin->ctx.probeEndIdx = endIdx;
2,333,298✔
974
  pJoin->ctx.pBuildRow = NULL;
2,333,298✔
975
  pJoin->ctx.pProbeData = pBlock;
2,333,298✔
976
  pJoin->ctx.rowRemains = true;
2,333,298✔
977
  pJoin->ctx.probePreIdx = 0;
2,333,298✔
978
  pJoin->ctx.probePostIdx = endIdx + 1;
2,333,298✔
979

980
  if (!IS_INNER_NONE_JOIN(pJoin->joinType, pJoin->subType) && startIdx > 0) {
2,333,298✔
981
    pJoin->ctx.probePhase = E_JOIN_PHASE_PRE;
×
982
  } else {
983
    pJoin->ctx.probePhase = E_JOIN_PHASE_CUR;
2,333,298✔
984
  }
985

986
  HJ_ERR_RET((*pJoin->joinFp)(pOperator));
2,333,298✔
987

988
  return TSDB_CODE_SUCCESS;
2,333,298✔
989
}
990

991
void hJoinSetDone(struct SOperatorInfo* pOperator) {
1,198,190✔
992
  setOperatorCompleted(pOperator);
1,198,190✔
993

994
  SHJoinOperatorInfo* pInfo = pOperator->info;
1,198,190✔
995
  hJoinDestroyKeyHash(&pInfo->pKeyHash);
1,198,190✔
996

997
  qDebug("hash Join done");  
1,198,190✔
998
}
1,198,190✔
999

1000
static int32_t hJoinMainProcess(struct SOperatorInfo* pOperator, SSDataBlock** pResBlock) {
1,274,990✔
1001
  SHJoinOperatorInfo* pJoin = pOperator->info;
1,274,990✔
1002
  SExecTaskInfo*      pTaskInfo = pOperator->pTaskInfo;
1,274,990✔
1003
  int32_t             code = TSDB_CODE_SUCCESS;
1,274,990✔
1004
  int32_t             lino = 0;
1,274,990✔
1005
  SSDataBlock*        pRes = pJoin->finBlk;
1,274,990✔
1006

1007
  QRY_PARAM_CHECK(pResBlock);
1,274,990✔
1008

1009
  if (pOperator->status == OP_EXEC_DONE) {
1,274,990✔
1010
    pRes->info.rows = 0;
76,800✔
1011
    goto _end;
76,800✔
1012
  }
1013

1014
  if (!pJoin->keyHashBuilt) {
1,198,190✔
1015
    pJoin->keyHashBuilt = true;
1,198,190✔
1016

1017
    bool queryDone = false;
1,198,190✔
1018
    code = hJoinBuildHash(pOperator, &queryDone);
1,198,190✔
1019
    QUERY_CHECK_CODE(code, lino, _end);
1,198,190✔
1020

1021
    if (queryDone) {
1,198,190✔
1022
      goto _end;
18,045✔
1023
    }
1024
  }
1025

1026
  blockDataCleanup(pRes);
1,180,145✔
1027

1028
  if (pJoin->ctx.rowRemains) {
1,180,145✔
UNCOV
1029
    code = (*pJoin->joinFp)(pOperator);
×
UNCOV
1030
    QUERY_CHECK_CODE(code, lino, _end);
×
1031

UNCOV
1032
    if (pRes->info.rows > 0 && pJoin->pFinFilter != NULL) {
×
1033
      code = doFilter(pRes, pJoin->pFinFilter, NULL, NULL);
×
1034
      QUERY_CHECK_CODE(code, lino, _end);
×
1035
    }
1036

1037
    if (pRes->info.rows > 0) {
×
1038
      *pResBlock = pRes;
×
UNCOV
1039
      return code;
×
1040
    }
1041
  }
1042

1043
  while (true) {
2,333,298✔
1044
    SSDataBlock* pBlock = getNextBlockFromDownstream(pOperator, pJoin->pProbe->downStreamIdx);
3,513,443✔
1045
    if (NULL == pBlock) {
3,513,443✔
1046
      hJoinSetDone(pOperator);
1,180,145✔
1047
      break;
1,180,145✔
1048
    }
1049

1050
    pJoin->execInfo.probeBlkNum++;
2,333,298✔
1051
    pJoin->execInfo.probeBlkRows += pBlock->info.rows;
2,333,298✔
1052

1053
    code = hJoinPrepareStart(pOperator, pBlock);
2,333,298✔
1054
    QUERY_CHECK_CODE(code, lino, _end);
2,333,298✔
1055

1056
    if (!hJoinBlkReachThreshold(pJoin, pRes->info.rows)) {
2,333,298✔
1057
      continue;
2,333,298✔
1058
    }
1059

UNCOV
1060
    if (pRes->info.rows > 0 && pJoin->pFinFilter != NULL) {
×
UNCOV
1061
      code = doFilter(pRes, pJoin->pFinFilter, NULL, NULL);
×
UNCOV
1062
      QUERY_CHECK_CODE(code, lino, _end);
×
1063
    }
1064

1065
    if (pRes->info.rows > 0) {
×
1066
      break;
×
1067
    }
1068
  }
1069

1070
_end:
1,274,990✔
1071
  if (code != TSDB_CODE_SUCCESS) {
1,274,990✔
UNCOV
1072
    qError("%s failed at line %d since %s", __func__, lino, tstrerror(code));
×
UNCOV
1073
    pTaskInfo->code = code;
×
UNCOV
1074
    T_LONG_JMP(pTaskInfo->env, code);
×
1075
  }
1076
  if (pRes->info.rows > 0) {
1,274,990✔
1077
    *pResBlock = pRes;
76,800✔
1078
    qDebug("%s %s output %" PRId64 " rows final res", GET_TASKID(pTaskInfo), __func__, pRes->info.rows);
76,800✔
1079
  }
1080

1081
  return code;
1,274,990✔
1082
}
1083

1084
static void destroyHashJoinOperator(void* param) {
1,197,646✔
1085
  SHJoinOperatorInfo* pJoinOperator = (SHJoinOperatorInfo*)param;
1,197,646✔
1086
  qDebug("hashJoin exec info, buildBlk:%" PRId64 ", buildRows:%" PRId64 ", probeBlk:%" PRId64 ", probeRows:%" PRId64 ", resRows:%" PRId64, 
1,197,646✔
1087
         pJoinOperator->execInfo.buildBlkNum, pJoinOperator->execInfo.buildBlkRows, pJoinOperator->execInfo.probeBlkNum, 
1088
         pJoinOperator->execInfo.probeBlkRows, pJoinOperator->execInfo.resRows);
1089

1090
  hJoinDestroyKeyHash(&pJoinOperator->pKeyHash);
1,197,646✔
1091

1092
  hJoinFreeTableInfo(&pJoinOperator->tbs[0]);
1,197,646✔
1093
  hJoinFreeTableInfo(&pJoinOperator->tbs[1]);
1,197,646✔
1094
  blockDataDestroy(pJoinOperator->finBlk);
1,197,646✔
1095
  pJoinOperator->finBlk = NULL;
1,197,646✔
1096
  taosMemoryFreeClear(pJoinOperator->pResColMap);
1,197,646✔
1097
  taosArrayDestroyEx(pJoinOperator->pRowBufs, hJoinFreeBufPage);
1,197,646✔
1098

1099
  taosMemoryFreeClear(param);
1,197,646✔
1100
}
1,197,646✔
1101

1102
int32_t hJoinHandleConds(SHJoinOperatorInfo* pJoin, SHashJoinPhysiNode* pJoinNode, SExecTaskInfo* pTaskInfo) {
1,197,646✔
1103
  switch (pJoin->joinType) {
1,197,646✔
1104
    case JOIN_TYPE_INNER: {
1,197,646✔
1105
      SNode* pCond = NULL;
1,197,646✔
1106
      if (pJoinNode->pFullOnCond != NULL) {
1,197,646✔
UNCOV
1107
        if (pJoinNode->node.pConditions != NULL) {
×
UNCOV
1108
          HJ_ERR_RET(mergeJoinConds(&pJoinNode->pFullOnCond, &pJoinNode->node.pConditions));
×
1109
        }
UNCOV
1110
        pCond = pJoinNode->pFullOnCond;
×
1111
      } else if (pJoinNode->node.pConditions != NULL) {
1,197,646✔
UNCOV
1112
        pCond = pJoinNode->node.pConditions;
×
1113
      }
1114

1115
      HJ_ERR_RET(filterInitFromNode(pCond, &pJoin->pFinFilter, 0, pTaskInfo->pStreamRuntimeInfo));
1,197,646✔
1116
      break;
1,197,646✔
1117
    }
UNCOV
1118
    case JOIN_TYPE_LEFT:
×
1119
    case JOIN_TYPE_RIGHT:
1120
    case JOIN_TYPE_FULL:
UNCOV
1121
      if (pJoinNode->pFullOnCond != NULL) {
×
UNCOV
1122
        HJ_ERR_RET(filterInitFromNode(pJoinNode->pFullOnCond, &pJoin->pPreFilter, 0,
×
1123
                                      pTaskInfo->pStreamRuntimeInfo));
1124
      }
1125
      if (pJoinNode->node.pConditions != NULL) {
×
UNCOV
1126
        HJ_ERR_RET(filterInitFromNode(pJoinNode->node.pConditions, &pJoin->pFinFilter, 0,
×
1127
                                      pTaskInfo->pStreamRuntimeInfo));
1128
      }
1129
      break;
×
UNCOV
1130
    default:
×
UNCOV
1131
      break;
×
1132
  }
1133

1134
  return TSDB_CODE_SUCCESS;
1,197,646✔
1135
}
1136

1137
static uint32_t hJoinGetFinBlkCapacity(SHJoinOperatorInfo* pJoin, SHashJoinPhysiNode* pJoinNode) {
1,197,646✔
1138
  uint32_t maxRows = TMAX(HJOIN_DEFAULT_BLK_ROWS_NUM, HJOIN_BLK_SIZE_LIMIT/pJoinNode->node.pOutputDataBlockDesc->totalRowSize);
1,197,646✔
1139
  if (INT64_MAX != pJoin->ctx.limit && NULL == pJoin->pFinFilter) {
1,197,646✔
UNCOV
1140
    uint32_t limitMaxRows = pJoin->ctx.limit / HJOIN_BLK_THRESHOLD_RATIO + 1;
×
UNCOV
1141
    return (maxRows > limitMaxRows) ? limitMaxRows : maxRows;
×
1142
  }
1143

1144
  return maxRows;
1,197,646✔
1145
}
1146

1147

1148
int32_t hJoinInitResBlocks(SHJoinOperatorInfo* pJoin, SHashJoinPhysiNode* pJoinNode) {
1,197,646✔
1149
  pJoin->finBlk = createDataBlockFromDescNode(pJoinNode->node.pOutputDataBlockDesc);
1,197,646✔
1150
  if (NULL == pJoin->finBlk) {
1,197,646✔
UNCOV
1151
    QRY_ERR_RET(terrno);
×
1152
  }
1153

1154
  int32_t code = blockDataEnsureCapacity(pJoin->finBlk, hJoinGetFinBlkCapacity(pJoin, pJoinNode));
1,197,646✔
1155
  if (TSDB_CODE_SUCCESS != code) {
1,197,646✔
UNCOV
1156
    QRY_ERR_RET(code);
×
1157
  }
1158
  
1159
  if (NULL != pJoin->pPreFilter) {
1,197,646✔
UNCOV
1160
    pJoin->midBlk = NULL;
×
UNCOV
1161
    code = createOneDataBlock(pJoin->finBlk, false, &pJoin->midBlk);
×
UNCOV
1162
    if (code) {
×
1163
      QRY_ERR_RET(code);
×
1164
    }
1165

UNCOV
1166
    code = blockDataEnsureCapacity(pJoin->midBlk, pJoin->finBlk->info.capacity);
×
1167
    if (TSDB_CODE_SUCCESS != code) {
×
1168
      QRY_ERR_RET(code);
×
1169
    }
1170
  }
1171

1172
  pJoin->blkThreshold = pJoin->finBlk->info.capacity * HJOIN_BLK_THRESHOLD_RATIO;
1,197,646✔
1173
  return TSDB_CODE_SUCCESS;
1,197,646✔
1174
}
1175

1176
static int32_t resetHashJoinOperState(SOperatorInfo* pOper) {
816✔
1177
  SHJoinOperatorInfo* pHjOper = pOper->info;
816✔
1178
  pHjOper->keyHashBuilt = false;
816✔
1179
  blockDataCleanup(pHjOper->midBlk);
816✔
1180
  blockDataCleanup(pHjOper->finBlk);
816✔
1181
  pOper->status = OP_NOT_OPENED;
816✔
1182

1183
  pHjOper->execInfo = (SHJoinExecInfo){0};
816✔
1184

1185
  void*   pIte = NULL;
816✔
1186
  int32_t iter = 0;
816✔
1187
  while ((pIte = tSimpleHashIterate(pHjOper->pKeyHash, pIte, &iter)) != NULL) {
816✔
UNCOV
1188
    SGroupData* pGroup = pIte;
×
UNCOV
1189
    SBufRowInfo* pRow = pGroup->rows;
×
UNCOV
1190
    SBufRowInfo* pNext = NULL;
×
UNCOV
1191
    while (pRow) {
×
UNCOV
1192
      pNext = pRow->next;
×
UNCOV
1193
      taosMemoryFree(pRow);
×
UNCOV
1194
      pRow = pNext;
×
1195
    }
1196
  }
1197
  tSimpleHashCleanup(pHjOper->pKeyHash);
816✔
1198
  size_t hashCap = pHjOper->pBuild->inputStat.inputRowNum > 0 ? (pHjOper->pBuild->inputStat.inputRowNum * 1.5) : 1024;
816✔
1199
  pHjOper->pKeyHash = tSimpleHashInit(hashCap, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY));
816✔
1200
  if (pHjOper->pKeyHash == NULL) {
816✔
1201
    return terrno; 
×
1202
  }
1203
  taosArrayDestroyEx(pHjOper->pRowBufs, hJoinFreeBufPage);
816✔
1204
  int32_t code = hJoinInitBufPages(pHjOper);
816✔
1205
  int64_t limit = pHjOper->ctx.limit;
816✔
1206
  pHjOper->ctx = (SHJoinCtx){0};
816✔
1207
  pHjOper->ctx.limit = limit;
816✔
1208
  return code;
816✔
1209
}
1210

1211
int32_t createHashJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream,
1,197,646✔
1212
                                           SHashJoinPhysiNode* pJoinNode, SExecTaskInfo* pTaskInfo, SOperatorInfo** pOptrInfo) {
1213
  QRY_PARAM_CHECK(pOptrInfo);
1,197,646✔
1214

1215
  int32_t             code = TSDB_CODE_SUCCESS;
1,197,646✔
1216
  SHJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SHJoinOperatorInfo));
1,197,646✔
1217
  SOperatorInfo*      pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
1,197,646✔
1218
  if (pOperator == NULL || pInfo == NULL) {
1,197,646✔
UNCOV
1219
    code = terrno;
×
UNCOV
1220
    goto _return;
×
1221
  }
1222
  initOperatorCostInfo(pOperator);
1,197,646✔
1223

1224
  pInfo->tblTimeRange.skey = pJoinNode->timeRange.skey;
1,197,646✔
1225
  pInfo->tblTimeRange.ekey = pJoinNode->timeRange.ekey;
1,197,646✔
1226
  
1227
  pInfo->ctx.limit = (pJoinNode->node.pLimit && ((SLimitNode*)pJoinNode->node.pLimit)->limit) ? ((SLimitNode*)pJoinNode->node.pLimit)->limit->datum.i : INT64_MAX;
1,197,646✔
1228

1229
  setOperatorInfo(pOperator, "HashJoinOperator", QUERY_NODE_PHYSICAL_PLAN_HASH_JOIN, false, OP_NOT_OPENED, pInfo, pTaskInfo);
1,197,646✔
1230

1231
  HJ_ERR_JRET(hJoinInitTableInfo(pInfo, pJoinNode, pDownstream, 0, &pJoinNode->inputStat[0]));
1,197,646✔
1232
  HJ_ERR_JRET(hJoinInitTableInfo(pInfo, pJoinNode, pDownstream, 1, &pJoinNode->inputStat[1]));
1,197,646✔
1233

1234
  hJoinSetBuildAndProbeTable(pInfo, pJoinNode);
1,197,646✔
1235
  
1236
  HJ_ERR_JRET(hJoinBuildResColsMap(pInfo, pJoinNode));
1,197,646✔
1237

1238
  HJ_ERR_JRET(hJoinInitBufPages(pInfo));
1,197,646✔
1239

1240
  size_t hashCap = pInfo->pBuild->inputStat.inputRowNum > 0 ? (pInfo->pBuild->inputStat.inputRowNum * 1.5) : 1024;
1,197,646✔
1241
  pInfo->pKeyHash = tSimpleHashInit(hashCap, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY));
1,197,646✔
1242
  if (pInfo->pKeyHash == NULL) {
1,197,646✔
UNCOV
1243
    code = terrno;
×
UNCOV
1244
    goto _return;
×
1245
  }
1246

1247
  HJ_ERR_JRET(hJoinHandleConds(pInfo, pJoinNode, pTaskInfo));
1,197,646✔
1248

1249
  HJ_ERR_JRET(hJoinInitResBlocks(pInfo, pJoinNode));
1,197,646✔
1250

1251
  HJ_ERR_JRET(hJoinSetImplFp(pInfo));
1,197,646✔
1252

1253
  HJ_ERR_JRET(appendDownstream(pOperator, pDownstream, numOfDownstream));
1,197,646✔
1254

1255
  pOperator->fpSet = createOperatorFpSet(optrDummyOpenFn, hJoinMainProcess, NULL, destroyHashJoinOperator, optrDefaultBufFn, NULL, optrDefaultGetNextExtFn, NULL);
1,197,646✔
1256
  setOperatorResetStateFn(pOperator, resetHashJoinOperState);
1,197,646✔
1257

1258
  qDebug("create hash Join operator done");
1,197,646✔
1259

1260
  *pOptrInfo = pOperator;
1,197,646✔
1261
  return TSDB_CODE_SUCCESS;
1,197,646✔
1262

UNCOV
1263
_return:
×
1264

UNCOV
1265
  if (pInfo != NULL) {
×
UNCOV
1266
    destroyHashJoinOperator(pInfo);
×
1267
  }
UNCOV
1268
  destroyOperatorAndDownstreams(pOperator, pDownstream, numOfDownstream);
×
1269
  pTaskInfo->code = code;
×
UNCOV
1270
  return code;
×
1271
}
1272

1273

STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc