projectoperator.c 29.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 * 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/>.
 */

#include "executorimpl.h"
L
Liu Jicong 已提交
17
#include "filter.h"
18 19
#include "functionMgt.h"

H
Haojun Liao 已提交
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
typedef struct SProjectOperatorInfo {
  SOptrBasicInfo binfo;
  SAggSupporter  aggSup;
  SArray*        pPseudoColInfo;
  SLimitInfo     limitInfo;
  bool           mergeDataBlocks;
  SSDataBlock*   pFinalRes;
} SProjectOperatorInfo;

typedef struct SIndefOperatorInfo {
  SOptrBasicInfo binfo;
  SAggSupporter  aggSup;
  SArray*        pPseudoColInfo;
  SExprSupp      scalarSup;
  uint64_t       groupId;
  SSDataBlock*   pNextGroupRes;
} SIndefOperatorInfo;

38 39 40 41 42 43 44
static SSDataBlock* doGenerateSourceData(SOperatorInfo* pOperator);
static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator);
static SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator);
static SArray*      setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols);
static void setFunctionResultOutput(SOperatorInfo* pOperator, SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage,
                                    int32_t numOfExprs);

45
static void destroyProjectOperatorInfo(void* param) {
46 47 48 49 50 51 52 53 54 55 56 57 58
  if (NULL == param) {
    return;
  }

  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
  cleanupBasicInfo(&pInfo->binfo);
  cleanupAggSup(&pInfo->aggSup);
  taosArrayDestroy(pInfo->pPseudoColInfo);

  blockDataDestroy(pInfo->pFinalRes);
  taosMemoryFreeClear(param);
}

59
static void destroyIndefinitOperatorInfo(void* param) {
60
  SIndefOperatorInfo* pInfo = (SIndefOperatorInfo*)param;
61 62 63
  if (pInfo == NULL) {
    return;
  }
64

65
  cleanupBasicInfo(&pInfo->binfo);
66 67 68 69 70 71 72 73 74
  taosArrayDestroy(pInfo->pPseudoColInfo);
  cleanupAggSup(&pInfo->aggSup);
  cleanupExprSupp(&pInfo->scalarSup);

  taosMemoryFreeClear(param);
}

SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SProjectPhysiNode* pProjPhyNode,
                                         SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
75
  int32_t               code = TSDB_CODE_SUCCESS;
76 77 78
  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
  if (pInfo == NULL || pOperator == NULL) {
79
    code = TSDB_CODE_OUT_OF_MEMORY;
80 81 82
    goto _error;
  }

83 84
  pOperator->pTaskInfo = pTaskInfo;

85 86 87
  int32_t    numOfCols = 0;
  SExprInfo* pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &numOfCols);

H
Haojun Liao 已提交
88
  SSDataBlock* pResBlock = createDataBlockFromDescNode(pProjPhyNode->node.pOutputDataBlockDesc);
89 90 91
  initLimitInfo(pProjPhyNode->node.pLimit, pProjPhyNode->node.pSlimit, &pInfo->limitInfo);

  pInfo->binfo.pRes = pResBlock;
92
  pInfo->pFinalRes = createOneDataBlock(pResBlock, false);
93 94 95 96 97 98 99 100 101 102

  if (pTaskInfo->execModel == OPTR_EXEC_MODEL_STREAM) {
    pInfo->mergeDataBlocks = false;
  } else {
    if (!pProjPhyNode->ignoreGroupId) {
      pInfo->mergeDataBlocks = false;
    } else {
      pInfo->mergeDataBlocks = pProjPhyNode->mergeDataBlock;
    }
  }
H
Haojun Liao 已提交
103

104 105 106 107 108 109 110 111
  int32_t numOfRows = 4096;
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;

  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
  int32_t TWOMB = 2 * 1024 * 1024;
  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
    numOfRows = TWOMB / pResBlock->info.rowSize;
  }
112

113
  initResultSizeInfo(&pOperator->resultInfo, numOfRows);
L
Liu Jicong 已提交
114 115
  code = initAggSup(&pOperator->exprSupp, &pInfo->aggSup, pExprInfo, numOfCols, keyBufSize, pTaskInfo->id.str,
                    pTaskInfo->streamInfo.pState);
116 117 118
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
119 120 121 122

  initBasicInfo(&pInfo->binfo, pResBlock);
  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfCols);

H
Haojun Liao 已提交
123 124 125 126 127
  code = filterInitFromNode((SNode*)pProjPhyNode->node.pConditions, &pOperator->exprSupp.pFilterInfo, 0);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }

128 129
  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pOperator->exprSupp.pCtx, numOfCols);

L
Liu Jicong 已提交
130 131 132 133
  setOperatorInfo(pOperator, "ProjectOperator", QUERY_NODE_PHYSICAL_PLAN_PROJECT, false, OP_NOT_OPENED, pInfo,
                  pTaskInfo);
  pOperator->fpSet = createOperatorFpSet(optrDummyOpenFn, doProjectOperation, NULL, destroyProjectOperatorInfo,
                                         optrDefaultBufFn, NULL);
134

135
  code = appendDownstream(pOperator, &downstream, 1);
136 137 138 139 140 141
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }

  return pOperator;

142
_error:
143
  destroyProjectOperatorInfo(pInfo);
144 145
  taosMemoryFree(pOperator);
  pTaskInfo->code = code;
146 147 148 149 150 151
  return NULL;
}

static int32_t discardGroupDataBlock(SSDataBlock* pBlock, SLimitInfo* pLimitInfo) {
  if (pLimitInfo->remainGroupOffset > 0) {
    // it is the first group
H
Haojun Liao 已提交
152 153
    if (pLimitInfo->currentGroupId == 0 || pLimitInfo->currentGroupId == pBlock->info.id.groupId) {
      pLimitInfo->currentGroupId = pBlock->info.id.groupId;
154
      return PROJECT_RETRIEVE_CONTINUE;
H
Haojun Liao 已提交
155
    } else if (pLimitInfo->currentGroupId != pBlock->info.id.groupId) {
156 157
      // now it is the data from a new group
      pLimitInfo->remainGroupOffset -= 1;
H
Haojun Liao 已提交
158
      pLimitInfo->currentGroupId = pBlock->info.id.groupId;
159 160 161 162 163 164 165 166

      // ignore data block in current group
      if (pLimitInfo->remainGroupOffset > 0) {
        return PROJECT_RETRIEVE_CONTINUE;
      }
    }

    // set current group id of the project operator
H
Haojun Liao 已提交
167
    pLimitInfo->currentGroupId = pBlock->info.id.groupId;
168 169 170 171 172 173 174 175
  }

  return PROJECT_RETRIEVE_DONE;
}

static int32_t setInfoForNewGroup(SSDataBlock* pBlock, SLimitInfo* pLimitInfo, SOperatorInfo* pOperator) {
  // remainGroupOffset == 0
  // here check for a new group data, we need to handle the data of the previous group.
H
Haojun Liao 已提交
176
  ASSERT(pLimitInfo->remainGroupOffset == 0 || pLimitInfo->remainGroupOffset == -1);
177

H
Haojun Liao 已提交
178
  if (pLimitInfo->currentGroupId != 0 && pLimitInfo->currentGroupId != pBlock->info.id.groupId) {
179 180
    pLimitInfo->numOfOutputGroups += 1;
    if ((pLimitInfo->slimit.limit > 0) && (pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
H
Haojun Liao 已提交
181
      setOperatorCompleted(pOperator);
182 183 184 185 186
      return PROJECT_RETRIEVE_DONE;
    }

    // reset the value for a new group data
    // existing rows that belongs to previous group.
187
    resetLimitInfoForNextGroup(pLimitInfo);
188 189 190 191 192
  }

  return PROJECT_RETRIEVE_DONE;
}

193
// todo refactor
194 195
static int32_t doIngroupLimitOffset(SLimitInfo* pLimitInfo, uint64_t groupId, SSDataBlock* pBlock,
                                    SOperatorInfo* pOperator) {
196 197
  // set current group id
  pLimitInfo->currentGroupId = groupId;
198 199
  bool limitReached = applyLimitOffset(pLimitInfo, pBlock, pOperator->pTaskInfo);
  if (pBlock->info.rows == 0) {
200
    return PROJECT_RETRIEVE_CONTINUE;
201 202
  } else {
    if (limitReached && (pLimitInfo->slimit.limit > 0 && pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
H
Haojun Liao 已提交
203
      setOperatorCompleted(pOperator);
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220
    }
  }

  return PROJECT_RETRIEVE_DONE;
}

SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;

  SExprSupp*   pSup = &pOperator->exprSupp;
  SSDataBlock* pRes = pInfo->pRes;
  SSDataBlock* pFinalRes = pProjectInfo->pFinalRes;

  blockDataCleanup(pFinalRes);

  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
L
Liu Jicong 已提交
221
  if (pTaskInfo->streamInfo.submit.msgStr) {
L
Liu Jicong 已提交
222 223 224
    pOperator->status = OP_OPENED;
  }

225 226 227
  if (pOperator->status == OP_EXEC_DONE) {
    if (pTaskInfo->execModel == OPTR_EXEC_MODEL_QUEUE) {
      pOperator->status = OP_OPENED;
L
Liu Jicong 已提交
228
      qDebug("projection in queue model, set status open and return null");
229 230 231 232 233 234 235 236 237 238 239 240 241 242 243
      return NULL;
    }

    return NULL;
  }

  int64_t st = 0;
  int32_t order = 0;
  int32_t scanFlag = 0;

  if (pOperator->cost.openCost == 0) {
    st = taosGetTimestampUs();
  }

  SOperatorInfo* downstream = pOperator->pDownstream[0];
244
  SLimitInfo*    pLimitInfo = &pProjectInfo->limitInfo;
245 246 247 248 249 250 251 252 253 254 255 256

  if (downstream == NULL) {
    return doGenerateSourceData(pOperator);
  }

  while (1) {
    while (1) {
      blockDataCleanup(pRes);

      // The downstream exec may change the value of the newgroup, so use a local variable instead.
      SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
      if (pBlock == NULL) {
L
Liu Jicong 已提交
257
        if (pTaskInfo->execModel == OPTR_EXEC_MODEL_QUEUE && pFinalRes->info.rows == 0) {
L
Liu Jicong 已提交
258
          pOperator->status = OP_OPENED;
L
Liu Jicong 已提交
259 260 261 262 263
          if (pOperator->status == OP_EXEC_RECV) {
            continue;
          } else {
            return NULL;
          }
L
Liu Jicong 已提交
264 265 266
        }
        qDebug("set op close, exec %d, status %d rows %d", pTaskInfo->execModel, pOperator->status,
               pFinalRes->info.rows);
H
Haojun Liao 已提交
267
        setOperatorCompleted(pOperator);
268 269
        break;
      }
L
Liu Jicong 已提交
270 271 272 273
      if (pTaskInfo->execModel == OPTR_EXEC_MODEL_QUEUE) {
        qDebug("set status recv");
        pOperator->status = OP_EXEC_RECV;
      }
274 275

      // for stream interval
5
54liuyao 已提交
276
      if (pBlock->info.type == STREAM_RETRIEVE || pBlock->info.type == STREAM_DELETE_RESULT ||
5
54liuyao 已提交
277
          pBlock->info.type == STREAM_DELETE_DATA || pBlock->info.type == STREAM_CREATE_CHILD_TABLE) {
278 279 280 281 282 283 284 285 286 287 288 289 290 291
        return pBlock;
      }

      int32_t status = discardGroupDataBlock(pBlock, pLimitInfo);
      if (status == PROJECT_RETRIEVE_CONTINUE) {
        continue;
      }

      setInfoForNewGroup(pBlock, pLimitInfo, pOperator);
      if (pOperator->status == OP_EXEC_DONE) {
        break;
      }

      // the pDataBlock are always the same one, no need to call this again
292
      int32_t code = getTableScanInfo(downstream, &order, &scanFlag, false);
293
      if (code != TSDB_CODE_SUCCESS) {
294
        T_LONG_JMP(pTaskInfo->env, code);
295 296
      }

297
      setInputDataBlock(pSup, pBlock, order, scanFlag, false);
298 299 300 301 302
      blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);

      code = projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs,
                                   pProjectInfo->pPseudoColInfo);
      if (code != TSDB_CODE_SUCCESS) {
303
        T_LONG_JMP(pTaskInfo->env, code);
304 305
      }

H
Haojun Liao 已提交
306
      status = doIngroupLimitOffset(pLimitInfo, pBlock->info.id.groupId, pInfo->pRes, pOperator);
307 308 309 310 311 312 313 314 315
      if (status == PROJECT_RETRIEVE_CONTINUE) {
        continue;
      }

      break;
    }

    if (pProjectInfo->mergeDataBlocks) {
      if (pRes->info.rows > 0) {
L
Liu Jicong 已提交
316
        pFinalRes->info.id.groupId = 0;  // clear groupId
317 318 319 320 321 322 323 324 325 326
        pFinalRes->info.version = pRes->info.version;

        // continue merge data, ignore the group id
        blockDataMerge(pFinalRes, pRes);
        if (pFinalRes->info.rows + pRes->info.rows <= pOperator->resultInfo.threshold) {
          continue;
        }
      }

      // do apply filter
H
Haojun Liao 已提交
327
      doFilter(pFinalRes, pOperator->exprSupp.pFilterInfo, NULL);
328 329 330

      // when apply the limit/offset for each group, pRes->info.rows may be 0, due to limit constraint.
      if (pFinalRes->info.rows > 0 || (pOperator->status == OP_EXEC_DONE)) {
L
Liu Jicong 已提交
331
        qDebug("project return %d rows, status %d", pFinalRes->info.rows, pOperator->status);
332 333 334 335 336
        break;
      }
    } else {
      // do apply filter
      if (pRes->info.rows > 0) {
H
Haojun Liao 已提交
337
        doFilter(pRes, pOperator->exprSupp.pFilterInfo, NULL);
338 339 340 341 342 343 344 345 346 347 348 349
        if (pRes->info.rows == 0) {
          continue;
        }
      }

      // no results generated
      break;
    }
  }

  SSDataBlock* p = pProjectInfo->mergeDataBlocks ? pFinalRes : pRes;
  pOperator->resultInfo.totalRows += p->info.rows;
D
dapan1121 已提交
350
  p->info.dataLoad = 1;
351 352 353 354

  if (pOperator->cost.openCost == 0) {
    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
  }
355

356 357 358 359 360 361 362 363 364 365 366
  return (p->info.rows > 0) ? p : NULL;
}

SOperatorInfo* createIndefinitOutputOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pNode,
                                                 SExecTaskInfo* pTaskInfo) {
  SIndefOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SIndefOperatorInfo));
  SOperatorInfo*      pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }

367 368
  pOperator->pTaskInfo = pTaskInfo;

369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384
  SExprSupp* pSup = &pOperator->exprSupp;

  SIndefRowsFuncPhysiNode* pPhyNode = (SIndefRowsFuncPhysiNode*)pNode;

  int32_t    numOfExpr = 0;
  SExprInfo* pExprInfo = createExprInfo(pPhyNode->pFuncs, NULL, &numOfExpr);

  if (pPhyNode->pExprs != NULL) {
    int32_t    num = 0;
    SExprInfo* pSExpr = createExprInfo(pPhyNode->pExprs, NULL, &num);
    int32_t    code = initExprSupp(&pInfo->scalarSup, pSExpr, num);
    if (code != TSDB_CODE_SUCCESS) {
      goto _error;
    }
  }

H
Haojun Liao 已提交
385
  SSDataBlock* pResBlock = createDataBlockFromDescNode(pPhyNode->node.pOutputDataBlockDesc);
386 387 388 389 390 391 392 393 394 395

  int32_t numOfRows = 4096;
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;

  // Make sure the size of SSDataBlock will never exceed the size of 2MB.
  int32_t TWOMB = 2 * 1024 * 1024;
  if (numOfRows * pResBlock->info.rowSize > TWOMB) {
    numOfRows = TWOMB / pResBlock->info.rowSize;
  }

H
Haojun Liao 已提交
396
  initBasicInfo(&pInfo->binfo, pResBlock);
397
  initResultSizeInfo(&pOperator->resultInfo, numOfRows);
398
  blockDataEnsureCapacity(pResBlock, numOfRows);
399

L
Liu Jicong 已提交
400 401
  int32_t code = initAggSup(pSup, &pInfo->aggSup, pExprInfo, numOfExpr, keyBufSize, pTaskInfo->id.str,
                            pTaskInfo->streamInfo.pState);
402 403 404
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
405 406

  setFunctionResultOutput(pOperator, &pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfExpr);
H
Haojun Liao 已提交
407 408 409 410
  code = filterInitFromNode((SNode*)pPhyNode->node.pConditions, &pOperator->exprSupp.pFilterInfo, 0);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
411 412 413 414

  pInfo->binfo.pRes = pResBlock;
  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pSup->pCtx, numOfExpr);

L
Liu Jicong 已提交
415 416 417 418
  setOperatorInfo(pOperator, "IndefinitOperator", QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC, false, OP_NOT_OPENED, pInfo,
                  pTaskInfo);
  pOperator->fpSet = createOperatorFpSet(optrDummyOpenFn, doApplyIndefinitFunction, NULL, destroyIndefinitOperatorInfo,
                                         optrDefaultBufFn, NULL);
419

420
  code = appendDownstream(pOperator, &downstream, 1);
421 422 423 424 425 426
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }

  return pOperator;

427
_error:
428
  destroyIndefinitOperatorInfo(pInfo);
429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
  taosMemoryFree(pOperator);
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
}

static void doHandleDataBlock(SOperatorInfo* pOperator, SSDataBlock* pBlock, SOperatorInfo* downstream,
                              SExecTaskInfo* pTaskInfo) {
  int32_t order = 0;
  int32_t scanFlag = 0;

  SIndefOperatorInfo* pIndefInfo = pOperator->info;
  SOptrBasicInfo*     pInfo = &pIndefInfo->binfo;
  SExprSupp*          pSup = &pOperator->exprSupp;

  // the pDataBlock are always the same one, no need to call this again
444
  int32_t code = getTableScanInfo(downstream, &order, &scanFlag, false);
445
  if (code != TSDB_CODE_SUCCESS) {
446
    T_LONG_JMP(pTaskInfo->env, code);
447 448 449 450 451 452 453 454
  }

  // there is an scalar expression that needs to be calculated before apply the group aggregation.
  SExprSupp* pScalarSup = &pIndefInfo->scalarSup;
  if (pScalarSup->pExprInfo != NULL) {
    code = projectApplyFunctions(pScalarSup->pExprInfo, pBlock, pBlock, pScalarSup->pCtx, pScalarSup->numOfExprs,
                                 pIndefInfo->pPseudoColInfo);
    if (code != TSDB_CODE_SUCCESS) {
455
      T_LONG_JMP(pTaskInfo->env, code);
456 457 458
    }
  }

459
  setInputDataBlock(pSup, pBlock, order, scanFlag, false);
460 461 462 463 464
  blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);

  code = projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs,
                               pIndefInfo->pPseudoColInfo);
  if (code != TSDB_CODE_SUCCESS) {
465
    T_LONG_JMP(pTaskInfo->env, code);
466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509
  }
}

SSDataBlock* doApplyIndefinitFunction(SOperatorInfo* pOperator) {
  SIndefOperatorInfo* pIndefInfo = pOperator->info;
  SOptrBasicInfo*     pInfo = &pIndefInfo->binfo;
  SExprSupp*          pSup = &pOperator->exprSupp;

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

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

  int64_t st = 0;

  if (pOperator->cost.openCost == 0) {
    st = taosGetTimestampUs();
  }

  SOperatorInfo* downstream = pOperator->pDownstream[0];

  while (1) {
    // here we need to handle the existsed group results
    if (pIndefInfo->pNextGroupRes != NULL) {  // todo extract method
      for (int32_t k = 0; k < pSup->numOfExprs; ++k) {
        SqlFunctionCtx* pCtx = &pSup->pCtx[k];

        SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
        pResInfo->initialized = false;
        pCtx->pOutput = NULL;
      }

      doHandleDataBlock(pOperator, pIndefInfo->pNextGroupRes, downstream, pTaskInfo);
      pIndefInfo->pNextGroupRes = NULL;
    }

    if (pInfo->pRes->info.rows < pOperator->resultInfo.threshold) {
      while (1) {
        // The downstream exec may change the value of the newgroup, so use a local variable instead.
        SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
        if (pBlock == NULL) {
H
Haojun Liao 已提交
510
          setOperatorCompleted(pOperator);
511 512 513
          break;
        }

H
Haojun Liao 已提交
514 515
        if (pIndefInfo->groupId == 0 && pBlock->info.id.groupId != 0) {
          pIndefInfo->groupId = pBlock->info.id.groupId;  // this is the initial group result
516
        } else {
H
Haojun Liao 已提交
517 518
          if (pIndefInfo->groupId != pBlock->info.id.groupId) {  // reset output buffer and computing status
            pIndefInfo->groupId = pBlock->info.id.groupId;
519 520 521 522 523 524 525 526 527 528 529 530
            pIndefInfo->pNextGroupRes = pBlock;
            break;
          }
        }

        doHandleDataBlock(pOperator, pBlock, downstream, pTaskInfo);
        if (pInfo->pRes->info.rows >= pOperator->resultInfo.threshold) {
          break;
        }
      }
    }

H
Haojun Liao 已提交
531
    doFilter(pInfo->pRes, pOperator->exprSupp.pFilterInfo, NULL);
532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582
    size_t rows = pInfo->pRes->info.rows;
    if (rows > 0 || pOperator->status == OP_EXEC_DONE) {
      break;
    } else {
      blockDataCleanup(pInfo->pRes);
    }
  }

  size_t rows = pInfo->pRes->info.rows;
  pOperator->resultInfo.totalRows += rows;

  if (pOperator->cost.openCost == 0) {
    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
  }

  return (rows > 0) ? pInfo->pRes : NULL;
}

void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
  for (int32_t j = 0; j < size; ++j) {
    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
        fmIsScalarFunc(pCtx[j].functionId)) {
      continue;
    }

    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
  }
}

/*
 * 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.
 * +------------+-----------------result column 1------------+------------------result column 2-----------+
 * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
 * +------------+--------------------------------------------+--------------------------------------------+
 *           offset[0]                                  offset[1]                                   offset[2]
 */
// TODO refactor: some function move away
void setFunctionResultOutput(SOperatorInfo* pOperator, SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage,
                             int32_t numOfExprs) {
  SExecTaskInfo*  pTaskInfo = pOperator->pTaskInfo;
  SqlFunctionCtx* pCtx = pOperator->exprSupp.pCtx;
  int32_t*        rowEntryInfoOffset = pOperator->exprSupp.rowEntryInfoOffset;

  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
  initResultRowInfo(pResultRowInfo);

  int64_t     tid = 0;
  int64_t     groupId = 0;
  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
D
dapan1121 已提交
583
                                            pTaskInfo, false, pSup, true);
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625

  for (int32_t i = 0; i < numOfExprs; ++i) {
    struct SResultRowEntryInfo* pEntry = getResultEntryInfo(pRow, i, rowEntryInfoOffset);
    cleanupResultRowEntry(pEntry);

    pCtx[i].resultInfo = pEntry;
    pCtx[i].scanFlag = stage;
  }

  initCtxOutputBuffer(pCtx, numOfExprs);
}

SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
  SArray* pList = taosArrayInit(4, sizeof(int32_t));
  for (int32_t i = 0; i < numOfCols; ++i) {
    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
      taosArrayPush(pList, &i);
    }
  }

  return pList;
}

SSDataBlock* doGenerateSourceData(SOperatorInfo* pOperator) {
  SProjectOperatorInfo* pProjectInfo = pOperator->info;

  SExprSupp*   pSup = &pOperator->exprSupp;
  SSDataBlock* pRes = pProjectInfo->binfo.pRes;

  blockDataEnsureCapacity(pRes, pOperator->resultInfo.capacity);
  SExprInfo* pExpr = pSup->pExprInfo;

  int64_t st = taosGetTimestampUs();

  for (int32_t k = 0; k < pSup->numOfExprs; ++k) {
    int32_t outputSlotId = pExpr[k].base.resSchema.slotId;

    ASSERT(pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE);
    SColumnInfoData* pColInfoData = taosArrayGet(pRes->pDataBlock, outputSlotId);

    int32_t type = pExpr[k].base.pParam[0].param.nType;
    if (TSDB_DATA_TYPE_NULL == type) {
626
      colDataSetNNULL(pColInfoData, 0, 1);
627
    } else {
628
      colDataSetVal(pColInfoData, 0, taosVariantGet(&pExpr[k].base.pParam[0].param, type), false);
629 630 631 632
    }
  }

  pRes->info.rows = 1;
H
Haojun Liao 已提交
633
  doFilter(pRes, pOperator->exprSupp.pFilterInfo, NULL);
634

635
  /*int32_t status = */ doIngroupLimitOffset(&pProjectInfo->limitInfo, 0, pRes, pOperator);
636 637 638

  pOperator->resultInfo.totalRows += pRes->info.rows;

H
Haojun Liao 已提交
639
  setOperatorCompleted(pOperator);
640 641 642 643 644
  if (pOperator->cost.openCost == 0) {
    pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
  }

  return (pRes->info.rows > 0) ? pRes : NULL;
L
Liu Jicong 已提交
645
}
H
Haojun Liao 已提交
646 647 648 649 650 651 652 653 654 655 656

static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, SArray* pPseudoList) {
  size_t num = (pPseudoList != NULL) ? taosArrayGetSize(pPseudoList) : 0;
  for (int32_t i = 0; i < num; ++i) {
    pCtx[i].pOutput = taosArrayGet(pResult->pDataBlock, i);
  }
}

int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
                              int32_t numOfOutput, SArray* pPseudoList) {
  setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
H
Haojun Liao 已提交
657
  pResult->info.dataLoad = 1;
H
Haojun Liao 已提交
658 659 660 661 662 663 664 665 666 667

  if (pSrcBlock == NULL) {
    for (int32_t k = 0; k < numOfOutput; ++k) {
      int32_t outputSlotId = pExpr[k].base.resSchema.slotId;

      ASSERT(pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE);
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);

      int32_t type = pExpr[k].base.pParam[0].param.nType;
      if (TSDB_DATA_TYPE_NULL == type) {
668
        colDataSetNNULL(pColInfoData, 0, 1);
H
Haojun Liao 已提交
669
      } else {
670
        colDataSetVal(pColInfoData, 0, taosVariantGet(&pExpr[k].base.pParam[0].param, type), false);
H
Haojun Liao 已提交
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699
      }
    }

    pResult->info.rows = 1;
    return TSDB_CODE_SUCCESS;
  }

  if (pResult != pSrcBlock) {
    pResult->info.id.groupId = pSrcBlock->info.id.groupId;
    memcpy(pResult->info.parTbName, pSrcBlock->info.parTbName, TSDB_TABLE_NAME_LEN);
  }

  // if the source equals to the destination, it is to create a new column as the result of scalar
  // function or some operators.
  bool createNewColModel = (pResult == pSrcBlock);
  if (createNewColModel) {
    blockDataEnsureCapacity(pResult, pResult->info.rows);
  }

  int32_t numOfRows = 0;

  for (int32_t k = 0; k < numOfOutput; ++k) {
    int32_t               outputSlotId = pExpr[k].base.resSchema.slotId;
    SqlFunctionCtx*       pfCtx = &pCtx[k];
    SInputColumnInfoData* pInputData = &pfCtx->input;

    if (pExpr[k].pExpr->nodeType == QUERY_NODE_COLUMN) {  // it is a project query
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
      if (pResult->info.rows > 0 && !createNewColModel) {
L
Liu Jicong 已提交
700 701 702 703 704 705 706 707 708 709 710
        if (pInputData->pData[0] == NULL) {
          int32_t slotId = pfCtx->param[0].pCol->slotId;

          SColumnInfoData* pInput = taosArrayGet(pSrcBlock->pDataBlock, slotId);

          colDataMergeCol(pColInfoData, pResult->info.rows, (int32_t*)&pResult->info.capacity, pInput,
                          pSrcBlock->info.rows);
        } else {
          colDataMergeCol(pColInfoData, pResult->info.rows, (int32_t*)&pResult->info.capacity, pInputData->pData[0],
                          pInputData->numOfRows);
        }
H
Haojun Liao 已提交
711
      } else {
L
Liu Jicong 已提交
712 713
        if (pInputData->pData[0] == NULL) {
          int32_t slotId = pfCtx->param[0].pCol->slotId;
H
Haojun Liao 已提交
714

L
Liu Jicong 已提交
715 716
          SColumnInfoData* pInput = taosArrayGet(pSrcBlock->pDataBlock, slotId);
          colDataAssign(pColInfoData, pInput, pSrcBlock->info.rows, &pResult->info);
H
Haojun Liao 已提交
717

L
Liu Jicong 已提交
718 719 720 721 722 723
          numOfRows = pSrcBlock->info.rows;
        } else {
          colDataAssign(pColInfoData, pInputData->pData[0], pInputData->numOfRows, &pResult->info);
          numOfRows = pInputData->numOfRows;
        }
      }
H
Haojun Liao 已提交
724 725 726 727 728 729 730
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE) {
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);

      int32_t offset = createNewColModel ? 0 : pResult->info.rows;

      int32_t type = pExpr[k].base.pParam[0].param.nType;
      if (TSDB_DATA_TYPE_NULL == type) {
731
        colDataSetNNULL(pColInfoData, offset, pSrcBlock->info.rows);
H
Haojun Liao 已提交
732 733
      } else {
        char* p = taosVariantGet(&pExpr[k].base.pParam[0].param, type);
734
        for (int32_t i = 0; i < pSrcBlock->info.rows; ++i) {
735
          colDataSetVal(pColInfoData, i + offset, p, false);
736
        }
H
Haojun Liao 已提交
737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802
      }

      numOfRows = pSrcBlock->info.rows;
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_OPERATOR) {
      SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
      taosArrayPush(pBlockList, &pSrcBlock);

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

      SScalarParam dest = {.columnData = &idata};
      int32_t      code = scalarCalculate(pExpr[k].pExpr->_optrRoot.pRootNode, pBlockList, &dest);
      if (code != TSDB_CODE_SUCCESS) {
        taosArrayDestroy(pBlockList);
        return code;
      }

      int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
      ASSERT(pResult->info.capacity > 0);

      colDataMergeCol(pResColData, startOffset, (int32_t*)&pResult->info.capacity, &idata, dest.numOfRows);
      colDataDestroy(&idata);

      numOfRows = dest.numOfRows;
      taosArrayDestroy(pBlockList);
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_FUNCTION) {
      // _rowts/_c0, not tbname column
      if (fmIsPseudoColumnFunc(pfCtx->functionId) && (!fmIsScanPseudoColumnFunc(pfCtx->functionId))) {
        // do nothing
      } else if (fmIsIndefiniteRowsFunc(pfCtx->functionId)) {
        SResultRowEntryInfo* pResInfo = GET_RES_INFO(pfCtx);
        pfCtx->fpSet.init(pfCtx, pResInfo);

        pfCtx->pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
        pfCtx->offset = createNewColModel ? 0 : pResult->info.rows;  // set the start offset

        // set the timestamp(_rowts) output buffer
        if (taosArrayGetSize(pPseudoList) > 0) {
          int32_t* outputColIndex = taosArrayGet(pPseudoList, 0);
          pfCtx->pTsOutput = (SColumnInfoData*)pCtx[*outputColIndex].pOutput;
        }

        // link pDstBlock to set selectivity value
        if (pfCtx->subsidiaries.num > 0) {
          pfCtx->pDstBlock = pResult;
        }

        int32_t code = pfCtx->fpSet.process(pfCtx);
        if (code != TSDB_CODE_SUCCESS) {
          return code;
        }
        numOfRows = pResInfo->numOfRes;
      } else if (fmIsAggFunc(pfCtx->functionId)) {
        // selective value output should be set during corresponding function execution
        if (fmIsSelectValueFunc(pfCtx->functionId)) {
          continue;
        }
        // _group_key function for "partition by tbname" + csum(col_name) query
        SColumnInfoData* pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
        int32_t          slotId = pfCtx->param[0].pCol->slotId;

        // todo handle the json tag
        SColumnInfoData* pInput = taosArrayGet(pSrcBlock->pDataBlock, slotId);
        for (int32_t f = 0; f < pSrcBlock->info.rows; ++f) {
          bool isNull = colDataIsNull_s(pInput, f);
          if (isNull) {
803
            colDataSetNULL(pOutput, pResult->info.rows + f);
H
Haojun Liao 已提交
804 805
          } else {
            char* data = colDataGetData(pInput, f);
806
            colDataSetVal(pOutput, pResult->info.rows + f, data, isNull);
H
Haojun Liao 已提交
807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842
          }
        }

      } else {
        SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
        taosArrayPush(pBlockList, &pSrcBlock);

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

        SScalarParam dest = {.columnData = &idata};
        int32_t      code = scalarCalculate((SNode*)pExpr[k].pExpr->_function.pFunctNode, pBlockList, &dest);
        if (code != TSDB_CODE_SUCCESS) {
          taosArrayDestroy(pBlockList);
          return code;
        }

        int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
        ASSERT(pResult->info.capacity > 0);
        colDataMergeCol(pResColData, startOffset, (int32_t*)&pResult->info.capacity, &idata, dest.numOfRows);
        colDataDestroy(&idata);

        numOfRows = dest.numOfRows;
        taosArrayDestroy(pBlockList);
      }
    } else {
      return TSDB_CODE_OPS_NOT_SUPPORT;
    }
  }

  if (!createNewColModel) {
    pResult->info.rows += numOfRows;
  }

  return TSDB_CODE_SUCCESS;
}