executorimpl.c 153.3 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

H
Haojun Liao 已提交
16 17
#include "filter.h"
#include "function.h"
18 19
#include "functionMgt.h"
#include "os.h"
H
Haojun Liao 已提交
20
#include "querynodes.h"
21
#include "tfill.h"
dengyihao's avatar
dengyihao 已提交
22
#include "tname.h"
X
Xiaoyu Wang 已提交
23
#include "tref.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"
dengyihao's avatar
dengyihao 已提交
32
#include "index.h"
33
#include "query.h"
34 35
#include "tcompare.h"
#include "tcompression.h"
H
Haojun Liao 已提交
36
#include "thash.h"
37
#include "ttypes.h"
dengyihao's avatar
dengyihao 已提交
38
#include "vnode.h"
39

H
Haojun Liao 已提交
40
#define IS_MAIN_SCAN(runtime)          ((runtime)->scanFlag == MAIN_SCAN)
41 42 43 44 45 46
#define SET_REVERSE_SCAN_FLAG(runtime) ((runtime)->scanFlag = REVERSE_SCAN)

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

#if 0
static UNUSED_FUNC void *u_malloc (size_t __size) {
wafwerar's avatar
wafwerar 已提交
47
  uint32_t v = taosRand();
48 49 50 51

  if (v % 1000 <= 0) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
52
    return taosMemoryMalloc(__size);
53 54 55 56
  }
}

static UNUSED_FUNC void* u_calloc(size_t num, size_t __size) {
wafwerar's avatar
wafwerar 已提交
57
  uint32_t v = taosRand();
58 59 60
  if (v % 1000 <= 0) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
61
    return taosMemoryCalloc(num, __size);
62 63 64 65
  }
}

static UNUSED_FUNC void* u_realloc(void* p, size_t __size) {
wafwerar's avatar
wafwerar 已提交
66
  uint32_t v = taosRand();
67 68 69
  if (v % 5 <= 1) {
    return NULL;
  } else {
wafwerar's avatar
wafwerar 已提交
70
    return taosMemoryRealloc(p, __size);
71 72 73 74 75 76 77 78
  }
}

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

X
Xiaoyu Wang 已提交
79
#define CLEAR_QUERY_STATUS(q, st)   ((q)->status &= (~(st)))
80 81
#define QUERY_IS_INTERVAL_QUERY(_q) ((_q)->interval.interval > 0)

L
Liu Jicong 已提交
82 83
int32_t getMaximumIdleDurationSec() { return tsShellActivityTimer * 2; }

84
static void setBlockSMAInfo(SqlFunctionCtx* pCtx, SExprInfo* pExpr, SSDataBlock* pBlock);
85

X
Xiaoyu Wang 已提交
86
static void releaseQueryBuf(size_t numOfTables);
87

88 89 90 91
static void destroyFillOperatorInfo(void* param);
static void destroyProjectOperatorInfo(void* param);
static void destroyOrderOperatorInfo(void* param);
static void destroyAggOperatorInfo(void* param);
X
Xiaoyu Wang 已提交
92

93 94
static void destroyIntervalOperatorInfo(void* param);
static void destroyExchangeOperatorInfo(void* param);
H
Haojun Liao 已提交
95

96 97
static void destroyOperatorInfo(SOperatorInfo* pOperator);

98
void doSetOperatorCompleted(SOperatorInfo* pOperator) {
99
  pOperator->status = OP_EXEC_DONE;
100

101
  pOperator->cost.totalCost = (taosGetTimestampUs() - pOperator->pTaskInfo->cost.start * 1000) / 1000.0;
H
Haojun Liao 已提交
102
  if (pOperator->pTaskInfo != NULL) {
103
    setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
104 105
  }
}
106

H
Haojun Liao 已提交
107
int32_t operatorDummyOpenFn(SOperatorInfo* pOperator) {
108
  OPTR_SET_OPENED(pOperator);
109
  pOperator->cost.openCost = 0;
H
Haojun Liao 已提交
110
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
111 112
}

113
SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t streamFn,
L
Liu Jicong 已提交
114
                                   __optr_fn_t cleanup, __optr_close_fn_t closeFn, __optr_encode_fn_t encode,
115
                                   __optr_decode_fn_t decode, __optr_explain_fn_t explain) {
116 117 118 119 120 121 122 123 124 125 126 127 128 129
  SOperatorFpSet fpSet = {
      ._openFn = openFn,
      .getNextFn = nextFn,
      .getStreamResFn = streamFn,
      .cleanupFn = cleanup,
      .closeFn = closeFn,
      .encodeResultRow = encode,
      .decodeResultRow = decode,
      .getExplainFn = explain,
  };

  return fpSet;
}

130 131
static int32_t doCopyToSDataBlock(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock, SExprSupp* pSup, SDiskbasedBuf* pBuf,
                                  SGroupResInfo* pGroupResInfo);
H
Haojun Liao 已提交
132

133
static void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size);
134
static void doSetTableGroupOutputBuf(SOperatorInfo* pOperator, int32_t numOfOutput, uint64_t groupId);
135

136
#if 0
L
Liu Jicong 已提交
137 138
static bool chkResultRowFromKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, char* pData,
                                int16_t bytes, bool masterscan, uint64_t uid) {
139 140 141
  bool existed = false;
  SET_RES_WINDOW_KEY(pRuntimeEnv->keyBuf, pData, bytes, uid);

L
Liu Jicong 已提交
142 143
  SResultRow** p1 =
      (SResultRow**)taosHashGet(pRuntimeEnv->pResultRowHashTable, pRuntimeEnv->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
144 145 146 147 148 149 150 151 152 153 154

  // 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
155
        //        existed = (pResultRowInfo->pResult[0] == (*p1));
156 157
      } 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
158 159
        int64_t* index =
            taosHashGet(pRuntimeEnv->pResultRowListSet, pRuntimeEnv->keyBuf, GET_RES_EXT_WINDOW_KEY_LEN(bytes));
160 161 162 163 164 165 166 167 168 169 170 171 172
        if (index != NULL) {
          existed = true;
        } else {
          existed = false;
        }
      }
    }

    return existed;
  }

  return p1 != NULL;
}
173
#endif
174

175
SResultRow* getNewResultRow(SDiskbasedBuf* pResultBuf, int32_t* currentPageId, int32_t interBufSize) {
L
Liu Jicong 已提交
176
  SFilePage* pData = NULL;
177 178 179

  // in the first scan, new space needed for results
  int32_t pageId = -1;
180
  if (*currentPageId == -1) {
181
    pData = getNewBufPage(pResultBuf, &pageId);
182 183
    pData->num = sizeof(SFilePage);
  } else {
184 185
    pData = getBufPage(pResultBuf, *currentPageId);
    pageId = *currentPageId;
186

wmmhello's avatar
wmmhello 已提交
187
    if (pData->num + interBufSize > getBufPageSize(pResultBuf)) {
188
      // release current page first, and prepare the next one
189
      releaseBufPage(pResultBuf, pData);
190

191
      pData = getNewBufPage(pResultBuf, &pageId);
192 193 194 195 196 197 198 199 200 201
      if (pData != NULL) {
        pData->num = sizeof(SFilePage);
      }
    }
  }

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

202 203
  setBufPageDirty(pData, true);

204 205 206 207
  // 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;
208
  *currentPageId = pageId;
209

wmmhello's avatar
wmmhello 已提交
210
  pData->num += interBufSize;
211 212 213
  return pResultRow;
}

214 215 216 217 218 219 220
/**
 * the struct of key in hash table
 * +----------+---------------+
 * | group id |   key data    |
 * | 8 bytes  | actual length |
 * +----------+---------------+
 */
221 222 223
SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pResultRowInfo, char* pData,
                                   int16_t bytes, bool masterscan, uint64_t groupId, SExecTaskInfo* pTaskInfo,
                                   bool isIntervalQuery, SAggSupporter* pSup) {
224
  SET_RES_WINDOW_KEY(pSup->keyBuf, pData, bytes, groupId);
H
Haojun Liao 已提交
225

dengyihao's avatar
dengyihao 已提交
226
  SResultRowPosition* p1 =
227
      (SResultRowPosition*)tSimpleHashGet(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
H
Haojun Liao 已提交
228

229 230
  SResultRow* pResult = NULL;

H
Haojun Liao 已提交
231 232
  // in case of repeat scan/reverse scan, no new time window added.
  if (isIntervalQuery) {
233
    if (masterscan && p1 != NULL) {  // the *p1 may be NULL in case of sliding+offset exists.
234
      pResult = getResultRowByPos(pResultBuf, p1, true);
235
      ASSERT(pResult->pageId == p1->pageId && pResult->offset == p1->offset);
H
Haojun Liao 已提交
236 237
    }
  } else {
dengyihao's avatar
dengyihao 已提交
238 239
    // In case of group by column query, the required SResultRow object must be existInCurrentResusltRowInfo in the
    // pResultRowInfo object.
H
Haojun Liao 已提交
240
    if (p1 != NULL) {
241
      // todo
242
      pResult = getResultRowByPos(pResultBuf, p1, true);
243
      ASSERT(pResult->pageId == p1->pageId && pResult->offset == p1->offset);
H
Haojun Liao 已提交
244 245 246
    }
  }

L
Liu Jicong 已提交
247
  // 1. close current opened time window
248
  if (pResultRowInfo->cur.pageId != -1 && ((pResult == NULL) || (pResult->pageId != pResultRowInfo->cur.pageId))) {
249
    SResultRowPosition pos = pResultRowInfo->cur;
X
Xiaoyu Wang 已提交
250
    SFilePage*         pPage = getBufPage(pResultBuf, pos.pageId);
251 252 253 254 255
    releaseBufPage(pResultBuf, pPage);
  }

  // allocate a new buffer page
  if (pResult == NULL) {
H
Haojun Liao 已提交
256
    ASSERT(pSup->resultRowSize > 0);
257
    pResult = getNewResultRow(pResultBuf, &pSup->currentPageId, pSup->resultRowSize);
258

259 260
    // add a new result set for a new group
    SResultRowPosition pos = {.pageId = pResult->pageId, .offset = pResult->offset};
261
    tSimpleHashPut(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes), &pos,
L
Liu Jicong 已提交
262
                   sizeof(SResultRowPosition));
H
Haojun Liao 已提交
263 264
  }

265 266 267
  // 2. set the new time window to be the new active time window
  pResultRowInfo->cur = (SResultRowPosition){.pageId = pResult->pageId, .offset = pResult->offset};

H
Haojun Liao 已提交
268
  // too many time window in query
269
  if (pTaskInfo->execModel == OPTR_EXEC_MODEL_BATCH &&
270
      tSimpleHashGetSize(pSup->pResultRowHashTable) > MAX_INTERVAL_TIME_WINDOW) {
271
    T_LONG_JMP(pTaskInfo->env, TSDB_CODE_QRY_TOO_MANY_TIMEWINDOW);
H
Haojun Liao 已提交
272 273
  }

H
Haojun Liao 已提交
274
  return pResult;
H
Haojun Liao 已提交
275 276
}

277
// a new buffer page for each table. Needs to opt this design
L
Liu Jicong 已提交
278
static int32_t addNewWindowResultBuf(SResultRow* pWindowRes, SDiskbasedBuf* pResultBuf, int32_t tid, uint32_t size) {
279 280 281 282
  if (pWindowRes->pageId != -1) {
    return 0;
  }

L
Liu Jicong 已提交
283
  SFilePage* pData = NULL;
284 285 286

  // in the first scan, new space needed for results
  int32_t pageId = -1;
287
  SIDList list = getDataBufPagesIdList(pResultBuf);
288 289

  if (taosArrayGetSize(list) == 0) {
290
    pData = getNewBufPage(pResultBuf, &pageId);
291
    pData->num = sizeof(SFilePage);
292 293
  } else {
    SPageInfo* pi = getLastPageInfo(list);
294
    pData = getBufPage(pResultBuf, getPageId(pi));
295
    pageId = getPageId(pi);
296

297
    if (pData->num + size > getBufPageSize(pResultBuf)) {
298
      // release current page first, and prepare the next one
299
      releaseBufPageInfo(pResultBuf, pi);
300

301
      pData = getNewBufPage(pResultBuf, &pageId);
302
      if (pData != NULL) {
303
        pData->num = sizeof(SFilePage);
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
      }
    }
  }

  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;
}

324
//  query_range_start, query_range_end, window_duration, window_start, window_end
325
void initExecTimeWindowInfo(SColumnInfoData* pColData, STimeWindow* pQueryWindow) {
326 327 328
  pColData->info.type = TSDB_DATA_TYPE_TIMESTAMP;
  pColData->info.bytes = sizeof(int64_t);

329
  colInfoDataEnsureCapacity(pColData, 5);
330 331 332 333 334 335 336 337 338
  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);
}

L
Liu Jicong 已提交
339
void cleanupExecTimeWindowInfo(SColumnInfoData* pColData) { colDataDestroy(pColData); }
H
Haojun Liao 已提交
340

341 342 343 344 345 346 347 348 349 350 351 352 353 354
typedef struct {
  bool    hasAgg;
  int32_t numOfRows;
  int32_t startOffset;
} SFunctionCtxStatus;

static void functionCtxSave(SqlFunctionCtx* pCtx, SFunctionCtxStatus* pStatus) {
  pStatus->hasAgg = pCtx->input.colDataAggIsSet;
  pStatus->numOfRows = pCtx->input.numOfRows;
  pStatus->startOffset = pCtx->input.startRowIndex;
}

static void functionCtxRestore(SqlFunctionCtx* pCtx, SFunctionCtxStatus* pStatus) {
  pCtx->input.colDataAggIsSet = pStatus->hasAgg;
H
Haojun Liao 已提交
355
  pCtx->input.numOfRows = pStatus->numOfRows;
356 357 358 359 360
  pCtx->input.startRowIndex = pStatus->startOffset;
}

void doApplyFunctions(SExecTaskInfo* taskInfo, SqlFunctionCtx* pCtx, SColumnInfoData* pTimeWindowData, int32_t offset,
                      int32_t forwardStep, int32_t numOfTotal, int32_t numOfOutput) {
361
  for (int32_t k = 0; k < numOfOutput; ++k) {
H
Haojun Liao 已提交
362
    // keep it temporarily
363 364
    SFunctionCtxStatus status = {0};
    functionCtxSave(&pCtx[k], &status);
365

366
    pCtx[k].input.startRowIndex = offset;
367
    pCtx[k].input.numOfRows = forwardStep;
368 369 370

    // not a whole block involved in query processing, statistics data can not be used
    // NOTE: the original value of isSet have been changed here
371 372
    if (pCtx[k].input.colDataAggIsSet && forwardStep < numOfTotal) {
      pCtx[k].input.colDataAggIsSet = false;
373 374
    }

375 376
    if (fmIsWindowPseudoColumnFunc(pCtx[k].functionId)) {
      SResultRowEntryInfo* pEntryInfo = GET_RES_INFO(&pCtx[k]);
377 378

      char* p = GET_ROWCELL_INTERBUF(pEntryInfo);
379

380
      SColumnInfoData idata = {0};
dengyihao's avatar
dengyihao 已提交
381
      idata.info.type = TSDB_DATA_TYPE_BIGINT;
382
      idata.info.bytes = tDataTypes[TSDB_DATA_TYPE_BIGINT].bytes;
dengyihao's avatar
dengyihao 已提交
383
      idata.pData = p;
384 385 386 387

      SScalarParam out = {.columnData = &idata};
      SScalarParam tw = {.numOfRows = 5, .columnData = pTimeWindowData};
      pCtx[k].sfp.process(&tw, 1, &out);
388
      pEntryInfo->numOfRes = 1;
389 390 391 392 393 394 395 396
    } else {
      int32_t code = TSDB_CODE_SUCCESS;
      if (functionNeedToExecute(&pCtx[k]) && pCtx[k].fpSet.process != NULL) {
        code = pCtx[k].fpSet.process(&pCtx[k]);

        if (code != TSDB_CODE_SUCCESS) {
          qError("%s apply functions error, code: %s", GET_TASKID(taskInfo), tstrerror(code));
          taskInfo->code = code;
397
          T_LONG_JMP(taskInfo->env, code);
398
        }
399
      }
400

401
      // restore it
402
      functionCtxRestore(&pCtx[k], &status);
403
    }
404 405 406
  }
}

dengyihao's avatar
dengyihao 已提交
407
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
408
                                   int32_t scanFlag, bool createDummyCol);
409

dengyihao's avatar
dengyihao 已提交
410 411
static void doSetInputDataBlockInfo(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock,
                                    int32_t order) {
412
  for (int32_t i = 0; i < pOperator->exprSupp.numOfExprs; ++i) {
413
    pCtx[i].order = order;
414
    pCtx[i].input.numOfRows = pBlock->info.rows;
415
    setBlockSMAInfo(&pCtx[i], &pOperator->exprSupp.pExprInfo[i], pBlock);
416
    pCtx[i].pSrcBlock = pBlock;
417 418 419
  }
}

X
Xiaoyu Wang 已提交
420 421
void setInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
                       int32_t scanFlag, bool createDummyCol) {
422
  if (pBlock->pBlockAgg != NULL) {
H
Haojun Liao 已提交
423
    doSetInputDataBlockInfo(pOperator, pCtx, pBlock, order);
424
  } else {
425
    doSetInputDataBlock(pOperator, pCtx, pBlock, order, scanFlag, createDummyCol);
H
Haojun Liao 已提交
426
  }
427 428
}

L
Liu Jicong 已提交
429 430
static int32_t doCreateConstantValColumnInfo(SInputColumnInfoData* pInput, SFunctParam* pFuncParam, int32_t paramIndex,
                                             int32_t numOfRows) {
431 432 433 434 435 436 437 438
  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)
439 440
    pColInfo->info.type = pFuncParam->param.nType;
    pColInfo->info.bytes = pFuncParam->param.nLen;
441 442

    pInput->pData[paramIndex] = pColInfo;
443 444
  } else {
    pColInfo = pInput->pData[paramIndex];
445 446
  }

447
  colInfoDataEnsureCapacity(pColInfo, numOfRows);
448

449
  int8_t type = pFuncParam->param.nType;
450 451
  if (type == TSDB_DATA_TYPE_BIGINT || type == TSDB_DATA_TYPE_UBIGINT) {
    int64_t v = pFuncParam->param.i;
dengyihao's avatar
dengyihao 已提交
452
    for (int32_t i = 0; i < numOfRows; ++i) {
453 454 455 456
      colDataAppendInt64(pColInfo, i, &v);
    }
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
dengyihao's avatar
dengyihao 已提交
457
    for (int32_t i = 0; i < numOfRows; ++i) {
458 459
      colDataAppendDouble(pColInfo, i, &v);
    }
460
  } else if (type == TSDB_DATA_TYPE_VARCHAR) {
L
Liu Jicong 已提交
461
    char* tmp = taosMemoryMalloc(pFuncParam->param.nLen + VARSTR_HEADER_SIZE);
462
    STR_WITH_SIZE_TO_VARSTR(tmp, pFuncParam->param.pz, pFuncParam->param.nLen);
L
Liu Jicong 已提交
463
    for (int32_t i = 0; i < numOfRows; ++i) {
464 465
      colDataAppend(pColInfo, i, tmp, false);
    }
466 467 468 469 470
  }

  return TSDB_CODE_SUCCESS;
}

dengyihao's avatar
dengyihao 已提交
471
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
X
Xiaoyu Wang 已提交
472
                                   int32_t scanFlag, bool createDummyCol) {
473 474
  int32_t code = TSDB_CODE_SUCCESS;

475
  for (int32_t i = 0; i < pOperator->exprSupp.numOfExprs; ++i) {
L
Liu Jicong 已提交
476
    pCtx[i].order = order;
477 478
    pCtx[i].input.numOfRows = pBlock->info.rows;

L
Liu Jicong 已提交
479
    pCtx[i].pSrcBlock = pBlock;
X
Xiaoyu Wang 已提交
480
    pCtx[i].scanFlag = scanFlag;
H
Haojun Liao 已提交
481

482
    SInputColumnInfoData* pInput = &pCtx[i].input;
483
    pInput->uid = pBlock->info.uid;
C
Cary Xu 已提交
484
    pInput->colDataAggIsSet = false;
485

486
    SExprInfo* pOneExpr = &pOperator->exprSupp.pExprInfo[i];
487
    for (int32_t j = 0; j < pOneExpr->base.numOfParams; ++j) {
dengyihao's avatar
dengyihao 已提交
488
      SFunctParam* pFuncParam = &pOneExpr->base.pParam[j];
G
Ganlin Zhao 已提交
489 490
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
dengyihao's avatar
dengyihao 已提交
491
        pInput->pData[j] = taosArrayGet(pBlock->pDataBlock, slotId);
492 493 494
        pInput->totalRows = pBlock->info.rows;
        pInput->numOfRows = pBlock->info.rows;
        pInput->startRowIndex = 0;
495

496
        // NOTE: the last parameter is the primary timestamp column
H
Haojun Liao 已提交
497
        // todo: refactor this
498
        if (fmIsImplicitTsFunc(pCtx[i].functionId) && (j == pOneExpr->base.numOfParams - 1)) {
L
Liu Jicong 已提交
499 500
          pInput->pPTS = pInput->pData[j];  // in case of merge function, this is not always the ts column data.
                                            //          ASSERT(pInput->pPTS->info.type == TSDB_DATA_TYPE_TIMESTAMP);
501
        }
502 503
        ASSERT(pInput->pData[j] != NULL);
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
504 505 506
        // 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) {
507 508 509 510
          pInput->totalRows = pBlock->info.rows;
          pInput->numOfRows = pBlock->info.rows;
          pInput->startRowIndex = 0;

511
          code = doCreateConstantValColumnInfo(pInput, pFuncParam, j, pBlock->info.rows);
512 513 514
          if (code != TSDB_CODE_SUCCESS) {
            return code;
          }
515
        }
G
Ganlin Zhao 已提交
516 517
      }
    }
H
Haojun Liao 已提交
518
  }
519 520

  return code;
H
Haojun Liao 已提交
521 522
}

523
static int32_t doAggregateImpl(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx) {
524
  for (int32_t k = 0; k < pOperator->exprSupp.numOfExprs; ++k) {
H
Haojun Liao 已提交
525
    if (functionNeedToExecute(&pCtx[k])) {
526
      // todo add a dummy funtion to avoid process check
527 528 529
      if (pCtx[k].fpSet.process == NULL) {
        continue;
      }
H
Haojun Liao 已提交
530

531 532 533 534
      int32_t code = pCtx[k].fpSet.process(&pCtx[k]);
      if (code != TSDB_CODE_SUCCESS) {
        qError("%s aggregate function error happens, code: %s", GET_TASKID(pOperator->pTaskInfo), tstrerror(code));
        return code;
535
      }
536 537
    }
  }
538 539

  return TSDB_CODE_SUCCESS;
540 541
}

H
Haojun Liao 已提交
542
static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, SArray* pPseudoList) {
dengyihao's avatar
dengyihao 已提交
543
  size_t num = (pPseudoList != NULL) ? taosArrayGetSize(pPseudoList) : 0;
H
Haojun Liao 已提交
544 545 546 547 548
  for (int32_t i = 0; i < num; ++i) {
    pCtx[i].pOutput = taosArrayGet(pResult->pDataBlock, i);
  }
}

549
int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
X
Xiaoyu Wang 已提交
550
                              int32_t numOfOutput, SArray* pPseudoList) {
H
Haojun Liao 已提交
551
  setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571

  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) {
        colDataAppendNNULL(pColInfoData, 0, 1);
      } else {
        colDataAppend(pColInfoData, 0, taosVariantGet(&pExpr[k].base.pParam[0].param, type), false);
      }
    }

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

H
Haojun Liao 已提交
572
  pResult->info.groupId = pSrcBlock->info.groupId;
H
Haojun Liao 已提交
573

574 575
  // if the source equals to the destination, it is to create a new column as the result of scalar
  // function or some operators.
576 577
  bool createNewColModel = (pResult == pSrcBlock);

578 579
  int32_t numOfRows = 0;

580
  for (int32_t k = 0; k < numOfOutput; ++k) {
581 582
    int32_t               outputSlotId = pExpr[k].base.resSchema.slotId;
    SqlFunctionCtx*       pfCtx = &pCtx[k];
583
    SInputColumnInfoData* pInputData = &pfCtx->input;
584

L
Liu Jicong 已提交
585
    if (pExpr[k].pExpr->nodeType == QUERY_NODE_COLUMN) {  // it is a project query
586
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
587
      if (pResult->info.rows > 0 && !createNewColModel) {
588
        colDataMergeCol(pColInfoData, pResult->info.rows, (int32_t*)&pResult->info.capacity, pInputData->pData[0],
589
                        pInputData->numOfRows);
590
      } else {
591
        colDataAssign(pColInfoData, pInputData->pData[0], pInputData->numOfRows, &pResult->info);
592
      }
593

594
      numOfRows = pInputData->numOfRows;
595
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE) {
596
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
597

dengyihao's avatar
dengyihao 已提交
598
      int32_t offset = createNewColModel ? 0 : pResult->info.rows;
599 600 601 602 603 604 605 606

      int32_t type = pExpr[k].base.pParam[0].param.nType;
      if (TSDB_DATA_TYPE_NULL == type) {
        colDataAppendNNULL(pColInfoData, offset, pSrcBlock->info.rows);
      } else {
        for (int32_t i = 0; i < pSrcBlock->info.rows; ++i) {
          colDataAppend(pColInfoData, i + offset, taosVariantGet(&pExpr[k].base.pParam[0].param, type), false);
        }
607
      }
608 609

      numOfRows = pSrcBlock->info.rows;
H
Haojun Liao 已提交
610
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_OPERATOR) {
611 612 613
      SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
      taosArrayPush(pBlockList, &pSrcBlock);

614
      SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
615
      SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
616

617
      SScalarParam dest = {.columnData = &idata};
X
Xiaoyu Wang 已提交
618
      int32_t      code = scalarCalculate(pExpr[k].pExpr->_optrRoot.pRootNode, pBlockList, &dest);
619 620 621 622
      if (code != TSDB_CODE_SUCCESS) {
        taosArrayDestroy(pBlockList);
        return code;
      }
623

dengyihao's avatar
dengyihao 已提交
624
      int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
625
      ASSERT(pResult->info.capacity > 0);
626
      colDataMergeCol(pResColData, startOffset, (int32_t*)&pResult->info.capacity, &idata, dest.numOfRows);
D
dapan1121 已提交
627
      colDataDestroy(&idata);
L
Liu Jicong 已提交
628

629
      numOfRows = dest.numOfRows;
630 631
      taosArrayDestroy(pBlockList);
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_FUNCTION) {
632 633
      // _rowts/_c0, not tbname column
      if (fmIsPseudoColumnFunc(pfCtx->functionId) && (!fmIsScanPseudoColumnFunc(pfCtx->functionId))) {
H
Haojun Liao 已提交
634
        // do nothing
635
      } else if (fmIsIndefiniteRowsFunc(pfCtx->functionId)) {
636 637
        SResultRowEntryInfo* pResInfo = GET_RES_INFO(pfCtx);
        pfCtx->fpSet.init(pfCtx, pResInfo);
638 639 640 641 642 643 644 645 646 647

        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;
        }

648 649 650 651 652
        // link pDstBlock to set selectivity value
        if (pfCtx->subsidiaries.num > 0) {
          pfCtx->pDstBlock = pResult;
        }

653
        numOfRows = pfCtx->fpSet.process(pfCtx);
H
Haojun Liao 已提交
654
      } else if (fmIsAggFunc(pfCtx->functionId)) {
G
Ganlin Zhao 已提交
655
        // selective value output should be set during corresponding function execution
656 657 658
        if (fmIsSelectValueFunc(pfCtx->functionId)) {
          continue;
        }
659 660
        // _group_key function for "partition by tbname" + csum(col_name) query
        SColumnInfoData* pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
661
        int32_t          slotId = pfCtx->param[0].pCol->slotId;
662 663 664

        // todo handle the json tag
        SColumnInfoData* pInput = taosArrayGet(pSrcBlock->pDataBlock, slotId);
665
        for (int32_t f = 0; f < pSrcBlock->info.rows; ++f) {
666 667 668 669 670 671 672 673 674
          bool isNull = colDataIsNull_s(pInput, f);
          if (isNull) {
            colDataAppendNULL(pOutput, pResult->info.rows + f);
          } else {
            char* data = colDataGetData(pInput, f);
            colDataAppend(pOutput, pResult->info.rows + f, data, isNull);
          }
        }

H
Haojun Liao 已提交
675 676 677
      } else {
        SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
        taosArrayPush(pBlockList, &pSrcBlock);
G
Ganlin Zhao 已提交
678

679
        SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
680
        SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
H
Haojun Liao 已提交
681

682
        SScalarParam dest = {.columnData = &idata};
X
Xiaoyu Wang 已提交
683
        int32_t      code = scalarCalculate((SNode*)pExpr[k].pExpr->_function.pFunctNode, pBlockList, &dest);
684 685 686 687
        if (code != TSDB_CODE_SUCCESS) {
          taosArrayDestroy(pBlockList);
          return code;
        }
688

dengyihao's avatar
dengyihao 已提交
689
        int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
690
        ASSERT(pResult->info.capacity > 0);
691
        colDataMergeCol(pResColData, startOffset, (int32_t*)&pResult->info.capacity, &idata, dest.numOfRows);
D
dapan1121 已提交
692
        colDataDestroy(&idata);
693 694

        numOfRows = dest.numOfRows;
H
Haojun Liao 已提交
695 696
        taosArrayDestroy(pBlockList);
      }
697
    } else {
698
      return TSDB_CODE_OPS_NOT_SUPPORT;
699 700
    }
  }
701

702 703 704
  if (!createNewColModel) {
    pResult->info.rows += numOfRows;
  }
705 706

  return TSDB_CODE_SUCCESS;
707 708
}

5
54liuyao 已提交
709
bool functionNeedToExecute(SqlFunctionCtx* pCtx) {
710
  struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
711

712 713 714 715 716
  // in case of timestamp column, always generated results.
  int32_t functionId = pCtx->functionId;
  if (functionId == -1) {
    return false;
  }
717

718 719
  if (pCtx->scanFlag == REPEAT_SCAN) {
    return fmIsRepeatScanFunc(pCtx->functionId);
720 721
  }

722 723
  if (isRowEntryCompleted(pResInfo)) {
    return false;
724 725
  }

726 727 728
  return true;
}

729 730 731 732 733 734 735
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;
    }
736

737 738 739
    // Set the correct column info (data type and bytes)
    pInput->pData[paramIndex]->info.type = type;
    pInput->pData[paramIndex]->info.bytes = tDataTypes[type].bytes;
740
  }
H
Haojun Liao 已提交
741

742 743 744 745 746 747
  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;
748 749
    }
  } else {
750
    da = pInput->pColumnDataAgg[paramIndex];
751 752
  }

753
  ASSERT(!IS_VAR_DATA_TYPE(type));
754

755 756
  if (type == TSDB_DATA_TYPE_BIGINT) {
    int64_t v = pFuncParam->param.i;
757
    *da = (SColumnDataAgg){.numOfNull = 0, .min = v, .max = v, .sum = v * numOfRows};
758 759
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
760
    *da = (SColumnDataAgg){.numOfNull = 0};
761

762 763 764 765 766 767
    *(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;

768
    *da = (SColumnDataAgg){.numOfNull = 0};
769 770 771 772 773
    *(bool*)&da->min = 0;
    *(bool*)&da->max = v;
    *(bool*)&da->sum = v * numOfRows;
  } else if (type == TSDB_DATA_TYPE_TIMESTAMP) {
    // do nothing
774
  } else {
775
    ASSERT(0);
776 777
  }

778 779
  return TSDB_CODE_SUCCESS;
}
780

781
void setBlockSMAInfo(SqlFunctionCtx* pCtx, SExprInfo* pExprInfo, SSDataBlock* pBlock) {
782 783 784 785 786 787 788 789 790
  int32_t numOfRows = pBlock->info.rows;

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

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

791 792
    for (int32_t j = 0; j < pExprInfo->base.numOfParams; ++j) {
      SFunctParam* pFuncParam = &pExprInfo->base.pParam[j];
793

794 795
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
796 797 798 799
        pInput->pColumnDataAgg[j] = pBlock->pBlockAgg[slotId];
        if (pInput->pColumnDataAgg[j] == NULL) {
          pInput->colDataAggIsSet = false;
        }
800 801 802 803

        // 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);
804 805
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
        doCreateConstantValColumnAggInfo(pInput, pFuncParam, pFuncParam->param.nType, j, pBlock->info.rows);
806 807
      }
    }
808
  } else {
809
    pInput->colDataAggIsSet = false;
810 811 812
  }
}

L
Liu Jicong 已提交
813
bool isTaskKilled(SExecTaskInfo* pTaskInfo) {
814 815
  // query has been executed more than tsShellActivityTimer, and the retrieve has not arrived
  // abort current query execution.
L
Liu Jicong 已提交
816 817
  if (pTaskInfo->owner != 0 &&
      ((taosGetTimestampSec() - pTaskInfo->cost.start / 1000) > 10 * getMaximumIdleDurationSec())
818 819
      /*(!needBuildResAfterQueryComplete(pTaskInfo))*/) {
    assert(pTaskInfo->cost.start != 0);
L
Liu Jicong 已提交
820 821 822
    //    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;
823 824 825 826 827
  }

  return false;
}

L
Liu Jicong 已提交
828
void setTaskKilled(SExecTaskInfo* pTaskInfo) { pTaskInfo->code = TSDB_CODE_TSC_QUERY_CANCELLED; }
829 830

/////////////////////////////////////////////////////////////////////////////////////////////
831
STimeWindow getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int64_t key) {
L
Liu Jicong 已提交
832
  STimeWindow win = {0};
833
  win.skey = taosTimeTruncate(key, pInterval, precision);
834 835

  /*
H
Haojun Liao 已提交
836
   * if the realSkey > INT64_MAX - pInterval->interval, the query duration between
837 838
   * realSkey and realEkey must be less than one interval.Therefore, no need to adjust the query ranges.
   */
839 840 841
  win.ekey = taosTimeAdd(win.skey, pInterval->interval, pInterval->intervalUnit, precision) - 1;
  if (win.ekey < win.skey) {
    win.ekey = INT64_MAX;
842
  }
843 844

  return win;
845 846
}

847
#if 0
H
Haojun Liao 已提交
848
static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
849 850
  STimeWindow w = {0};

dengyihao's avatar
dengyihao 已提交
851 852
  TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
  TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
853

854
  if (true) {
L
Liu Jicong 已提交
855
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
856 857 858 859 860 861
    assert(w.ekey >= pBlockInfo->window.skey);

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

L
Liu Jicong 已提交
862 863
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
864 865 866 867 868 869 870 871 872 873
      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
874
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
875 876 877 878 879 880
    assert(w.skey <= pBlockInfo->window.ekey);

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

L
Liu Jicong 已提交
881 882
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
883 884 885 886 887 888 889 890 891 892 893 894 895
      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;
}
896
#endif
897 898

static uint32_t doFilterByBlockTimeWindow(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock) {
H
Haojun Liao 已提交
899
#if 0
H
Haojun Liao 已提交
900
  SqlFunctionCtx* pCtx = pTableScanInfo->pCtx;
901
  uint32_t        status = BLK_DATA_NOT_LOAD;
902

L
Liu Jicong 已提交
903
  int32_t numOfOutput = 0;  // pTableScanInfo->numOfOutput;
904 905
  for (int32_t i = 0; i < numOfOutput; ++i) {
    int32_t functionId = pCtx[i].functionId;
H
Haojun Liao 已提交
906
    int32_t colId = pTableScanInfo->pExpr[i].base.pParam[0].pCol->colId;
907 908 909

    // group by + first/last should not apply the first/last block filter
    if (functionId < 0) {
910
      status |= BLK_DATA_DATA_LOAD;
911 912
      return status;
    } else {
L
Liu Jicong 已提交
913
      //      status |= aAggs[functionId].dataReqFunc(&pTableScanInfo->pCtx[i], &pBlock->info.window, colId);
914
      //      if ((status & BLK_DATA_DATA_LOAD) == BLK_DATA_DATA_LOAD) {
L
Liu Jicong 已提交
915 916
      //        return status;
      //      }
917 918 919 920
    }
  }

  return status;
H
Haojun Liao 已提交
921 922
#endif
  return 0;
923 924
}

L
Liu Jicong 已提交
925 926
int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                              uint32_t* status) {
927
  *status = BLK_DATA_NOT_LOAD;
928

H
Haojun Liao 已提交
929
  pBlock->pDataBlock = NULL;
L
Liu Jicong 已提交
930
  pBlock->pBlockAgg = NULL;
H
Haojun Liao 已提交
931

L
Liu Jicong 已提交
932 933
  //  int64_t groupId = pRuntimeEnv->current->groupIndex;
  //  bool    ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
934

H
Haojun Liao 已提交
935
  STaskCostInfo* pCost = &pTaskInfo->cost;
936

937 938
//  pCost->totalBlocks += 1;
//  pCost->totalRows += pBlock->info.rows;
H
Haojun Liao 已提交
939
#if 0
940 941 942
  // 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
943
      (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && overlapWithTimeWindow(pTaskInfo, &pBlock->info))) {
944
    (*status) = BLK_DATA_DATA_LOAD;
945 946 947
  }

  // check if this data block is required to load
948
  if ((*status) != BLK_DATA_DATA_LOAD) {
949 950 951 952 953 954 955
    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
956
      bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
957 958 959 960 961 962
      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,
963
                                    pTableScanInfo->rowEntryInfoOffset);
964 965 966
      } else {
        if (setResultOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pBlock->info.uid, &win, masterScan, &pResult, groupId,
                                    pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
967
                                    pTableScanInfo->rowEntryInfoOffset) != TSDB_CODE_SUCCESS) {
968
          T_LONG_JMP(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
969 970 971 972
        }
      }
    } else if (pQueryAttr->stableQuery && (!pQueryAttr->tsCompQuery) && (!pQueryAttr->diffQuery)) { // stable aggregate, not interval aggregate or normal column aggregate
      doSetTableGroupOutputBuf(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pTableScanInfo->pCtx,
973
                               pTableScanInfo->rowEntryInfoOffset, pTableScanInfo->numOfOutput,
974 975 976 977 978 979
                               pRuntimeEnv->current->groupIndex);
    }

    if (needFilter) {
      (*status) = doFilterByBlockTimeWindow(pTableScanInfo, pBlock);
    } else {
980
      (*status) = BLK_DATA_DATA_LOAD;
981 982 983 984
    }
  }

  SDataBlockInfo* pBlockInfo = &pBlock->info;
H
Haojun Liao 已提交
985
//  *status = updateBlockLoadStatus(pRuntimeEnv->pQueryAttr, *status);
986

987
  if ((*status) == BLK_DATA_NOT_LOAD || (*status) == BLK_DATA_FILTEROUT) {
988 989
    //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//           pBlockInfo->window.ekey, pBlockInfo->rows);
990
    pCost->skipBlocks += 1;
991
  } else if ((*status) == BLK_DATA_SMA_LOAD) {
992 993
    // this function never returns error?
    pCost->loadBlockStatis += 1;
994
//    tsdbRetrieveDatablockSMA(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
995 996

    if (pBlock->pBlockAgg == NULL) {  // data block statistics does not exist, load data block
997
//      pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
998 999 1000
      pCost->totalCheckedRows += pBlock->info.rows;
    }
  } else {
1001
    assert((*status) == BLK_DATA_DATA_LOAD);
1002 1003 1004

    // load the data block statistics to perform further filter
    pCost->loadBlockStatis += 1;
1005
//    tsdbRetrieveDatablockSMA(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1006 1007 1008 1009 1010 1011

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

H
Haojun Liao 已提交
1012
          bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1013 1014 1015 1016 1017
          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,
1018
                                      pTableScanInfo->rowEntryInfoOffset) != TSDB_CODE_SUCCESS) {
1019
            T_LONG_JMP(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
          }
        }
      }
      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
1030
            pCost->skipBlocks += 1;
1031 1032
            //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId,
//                   pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
1033
            (*status) = BLK_DATA_FILTEROUT;
1034 1035 1036 1037 1038 1039 1040
            return TSDB_CODE_SUCCESS;
          }
        }
      }
    }

    // current block has been discard due to filter applied
H
Haojun Liao 已提交
1041
//    if (!doFilterByBlockSMA(pRuntimeEnv, pBlock->pBlockAgg, pTableScanInfo->pCtx, pBlockInfo->rows)) {
1042
//      pCost->skipBlocks += 1;
1043 1044
//      qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//             pBlockInfo->window.ekey, pBlockInfo->rows);
1045
//      (*status) = BLK_DATA_FILTEROUT;
1046 1047 1048 1049 1050
//      return TSDB_CODE_SUCCESS;
//    }

    pCost->totalCheckedRows += pBlockInfo->rows;
    pCost->loadBlocks += 1;
1051
//    pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1052 1053 1054 1055 1056
//    if (pBlock->pDataBlock == NULL) {
//      return terrno;
//    }

//    if (pQueryAttr->pFilters != NULL) {
1057
//      filterSetColFieldData(pQueryAttr->pFilters, taosArrayGetSize(pBlock->pDataBlock), pBlock->pDataBlock);
1058
//    }
1059

1060 1061 1062 1063
//    if (pQueryAttr->pFilters != NULL || pRuntimeEnv->pTsBuf != NULL) {
//      filterColRowsInDataBlock(pRuntimeEnv, pBlock, ascQuery);
//    }
  }
H
Haojun Liao 已提交
1064
#endif
1065 1066 1067
  return TSDB_CODE_SUCCESS;
}

L
Liu Jicong 已提交
1068
static void updateTableQueryInfoForReverseScan(STableQueryInfo* pTableQueryInfo) {
1069 1070 1071 1072 1073
  if (pTableQueryInfo == NULL) {
    return;
  }
}

L
Liu Jicong 已提交
1074
void setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status) {
1075
  if (status == TASK_NOT_COMPLETED) {
H
Haojun Liao 已提交
1076
    pTaskInfo->status = status;
1077 1078
  } else {
    // QUERY_NOT_COMPLETED is not compatible with any other status, so clear its position first
1079
    CLEAR_QUERY_STATUS(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
1080
    pTaskInfo->status |= status;
1081 1082 1083
  }
}

1084
void setResultRowInitCtx(SResultRow* pResult, SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t* rowEntryInfoOffset) {
5
54liuyao 已提交
1085
  bool init = false;
1086
  for (int32_t i = 0; i < numOfOutput; ++i) {
1087
    pCtx[i].resultInfo = getResultEntryInfo(pResult, i, rowEntryInfoOffset);
5
54liuyao 已提交
1088 1089 1090
    if (init) {
      continue;
    }
1091 1092 1093 1094 1095

    struct SResultRowEntryInfo* pResInfo = pCtx[i].resultInfo;
    if (isRowEntryCompleted(pResInfo) && isRowEntryInitialized(pResInfo)) {
      continue;
    }
1096 1097 1098 1099 1100

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

1101 1102 1103 1104 1105 1106
    if (!pResInfo->initialized) {
      if (pCtx[i].functionId != -1) {
        pCtx[i].fpSet.init(&pCtx[i], pResInfo);
      } else {
        pResInfo->initialized = true;
      }
5
54liuyao 已提交
1107 1108
    } else {
      init = true;
1109 1110 1111 1112
    }
  }
}

1113
static void extractQualifiedTupleByFilterResult(SSDataBlock* pBlock, const SColumnInfoData* p, bool keep, int32_t status);
1114

1115
void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock, const SArray* pColMatchInfo) {
1116
  if (pFilterNode == NULL || pBlock->info.rows == 0) {
S
shenglian zhou 已提交
1117 1118
    return;
  }
1119

1120
  SFilterInfo* filter = NULL;
H
Haojun Liao 已提交
1121

H
Haojun Liao 已提交
1122
  // todo move to the initialization function
H
Haojun Liao 已提交
1123
  int32_t code = filterInitFromNode((SNode*)pFilterNode, &filter, 0);
1124
  SFilterColumnParam param1 = {.numOfCols = taosArrayGetSize(pBlock->pDataBlock), .pDataBlock = pBlock->pDataBlock};
1125 1126
  code = filterSetDataFromSlotId(filter, &param1);

1127 1128
  SColumnInfoData* p = NULL;
  int32_t status = 0;
1129

1130
  // todo the keep seems never to be True??
1131
  bool keep = filterExecute(filter, pBlock, &p, NULL, param1.numOfCols, &status);
D
dapan1121 已提交
1132
  filterFreeInfo(filter);
1133

1134
  extractQualifiedTupleByFilterResult(pBlock, p, keep, status);
1135 1136

  if (pColMatchInfo != NULL) {
1137
    for (int32_t i = 0; i < taosArrayGetSize(pColMatchInfo); ++i) {
1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148
      SColMatchInfo* pInfo = taosArrayGet(pColMatchInfo, i);
      if (pInfo->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
        SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, pInfo->targetSlotId);
        if (pColData->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
          blockDataUpdateTsWindow(pBlock, pInfo->targetSlotId);
          break;
        }
      }
    }
  }

1149 1150
  colDataDestroy(p);
  taosMemoryFree(p);
1151 1152
}

1153
void extractQualifiedTupleByFilterResult(SSDataBlock* pBlock, const SColumnInfoData* p, bool keep, int32_t status) {
1154 1155 1156 1157
  if (keep) {
    return;
  }

1158 1159 1160 1161 1162 1163 1164
  int32_t totalRows = pBlock->info.rows;

  if (status == FILTER_RESULT_ALL_QUALIFIED) {
    // here nothing needs to be done
  } else if (status == FILTER_RESULT_NONE_QUALIFIED) {
    pBlock->info.rows = 0;
  } else {
1165
    SSDataBlock* px = createOneDataBlock(pBlock, true);
1166

1167 1168
    size_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
    for (int32_t i = 0; i < numOfCols; ++i) {
1169 1170
      SColumnInfoData* pSrc = taosArrayGet(px->pDataBlock, i);
      SColumnInfoData* pDst = taosArrayGet(pBlock->pDataBlock, i);
1171
      // it is a reserved column for scalar function, and no data in this column yet.
1172
      if (pDst->pData == NULL || pSrc->pData == NULL) {
1173 1174 1175
        continue;
      }

1176 1177
      colInfoDataCleanup(pDst, pBlock->info.rows);

1178
      int32_t numOfRows = 0;
1179
      for (int32_t j = 0; j < totalRows; ++j) {
1180
        if (((int8_t*)p->pData)[j] == 0) {
D
dapan1121 已提交
1181 1182
          continue;
        }
1183

D
dapan1121 已提交
1184
        if (colDataIsNull_s(pSrc, j)) {
1185
          colDataAppendNULL(pDst, numOfRows);
D
dapan1121 已提交
1186
        } else {
1187
          colDataAppend(pDst, numOfRows, colDataGetData(pSrc, j), false);
D
dapan1121 已提交
1188
        }
1189
        numOfRows += 1;
H
Haojun Liao 已提交
1190
      }
1191

1192
      // todo this value can be assigned directly
1193 1194 1195 1196 1197
      if (pBlock->info.rows == totalRows) {
        pBlock->info.rows = numOfRows;
      } else {
        ASSERT(pBlock->info.rows == numOfRows);
      }
1198
    }
1199

dengyihao's avatar
dengyihao 已提交
1200
    blockDataDestroy(px);  // fix memory leak
1201 1202 1203
  }
}

1204
  void doSetTableGroupOutputBuf(SOperatorInfo* pOperator, int32_t numOfOutput, uint64_t groupId) {
1205
  // for simple group by query without interval, all the tables belong to one group result.
1206 1207 1208
  SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
  SAggOperatorInfo* pAggInfo = pOperator->info;

1209
  SResultRowInfo* pResultRowInfo = &pAggInfo->binfo.resultRowInfo;
1210 1211
  SqlFunctionCtx* pCtx = pOperator->exprSupp.pCtx;
  int32_t*        rowEntryInfoOffset = pOperator->exprSupp.rowEntryInfoOffset;
1212

1213
  SResultRow* pResultRow = doSetResultOutBufByKey(pAggInfo->aggSup.pResultBuf, pResultRowInfo, (char*)&groupId,
L
Liu Jicong 已提交
1214
                                                  sizeof(groupId), true, groupId, pTaskInfo, false, &pAggInfo->aggSup);
L
Liu Jicong 已提交
1215
  assert(pResultRow != NULL);
1216 1217 1218 1219 1220 1221

  /*
   * 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
1222 1223
    int32_t ret =
        addNewWindowResultBuf(pResultRow, pAggInfo->aggSup.pResultBuf, groupId, pAggInfo->binfo.pRes->info.rowSize);
1224 1225 1226 1227 1228
    if (ret != TSDB_CODE_SUCCESS) {
      return;
    }
  }

1229
  setResultRowInitCtx(pResultRow, pCtx, numOfOutput, rowEntryInfoOffset);
1230 1231
}

1232 1233 1234
static void setExecutionContext(SOperatorInfo* pOperator, int32_t numOfOutput, uint64_t groupId) {
  SAggOperatorInfo* pAggInfo = pOperator->info;
  if (pAggInfo->groupId != UINT64_MAX && pAggInfo->groupId == groupId) {
1235 1236
    return;
  }
1237 1238

  doSetTableGroupOutputBuf(pOperator, numOfOutput, groupId);
1239 1240

  // record the current active group id
H
Haojun Liao 已提交
1241
  pAggInfo->groupId = groupId;
1242 1243
}

dengyihao's avatar
dengyihao 已提交
1244 1245
static void doUpdateNumOfRows(SqlFunctionCtx* pCtx, SResultRow* pRow, int32_t numOfExprs,
                              const int32_t* rowCellOffset) {
1246
  bool returnNotNull = false;
1247
  for (int32_t j = 0; j < numOfExprs; ++j) {
1248
    struct SResultRowEntryInfo* pResInfo = getResultEntryInfo(pRow, j, rowCellOffset);
1249 1250 1251 1252 1253 1254 1255
    if (!isRowEntryInitialized(pResInfo)) {
      continue;
    }

    if (pRow->numOfRows < pResInfo->numOfRes) {
      pRow->numOfRows = pResInfo->numOfRes;
    }
1256

1257
    if (fmIsNotNullOutputFunc(pCtx[j].functionId)) {
1258 1259
      returnNotNull = true;
    }
1260
  }
S
shenglian zhou 已提交
1261 1262
  // if all expr skips all blocks, e.g. all null inputs for max function, output one row in final result.
  //  except for first/last, which require not null output, output no rows
1263
  if (pRow->numOfRows == 0 && !returnNotNull) {
1264
    pRow->numOfRows = 1;
1265 1266 1267
  }
}

1268 1269
static void doCopyResultToDataBlock(SExprInfo* pExprInfo, int32_t numOfExprs, SResultRow* pRow, SqlFunctionCtx* pCtx,
                                    SSDataBlock* pBlock, const int32_t* rowEntryOffset, SExecTaskInfo* pTaskInfo) {
1270 1271 1272
  for (int32_t j = 0; j < numOfExprs; ++j) {
    int32_t slotId = pExprInfo[j].base.resSchema.slotId;

1273
    pCtx[j].resultInfo = getResultEntryInfo(pRow, j, rowEntryOffset);
1274 1275 1276 1277
    if (pCtx[j].fpSet.finalize) {
      int32_t code = pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
      if (TAOS_FAILED(code)) {
        qError("%s build result data block error, code %s", GET_TASKID(pTaskInfo), tstrerror(code));
1278
        T_LONG_JMP(pTaskInfo->env, code);
1279 1280
      }
    } else if (strcmp(pCtx[j].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
1281
      // do nothing
1282
    } else {
1283 1284
      // expand the result into multiple rows. E.g., _wstart, top(k, 20)
      // the _wstart needs to copy to 20 following rows, since the results of top-k expands to 20 different rows.
1285 1286 1287 1288 1289 1290 1291
      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);
      }
    }
  }
1292 1293
}

1294 1295 1296
// todo refactor. SResultRow has direct pointer in miainfo
int32_t finalizeResultRows(SDiskbasedBuf* pBuf, SResultRowPosition* resultRowPosition, SExprSupp* pSup,
                           SSDataBlock* pBlock, SExecTaskInfo* pTaskInfo) {
1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322
  SFilePage*  page = getBufPage(pBuf, resultRowPosition->pageId);
  SResultRow* pRow = (SResultRow*)((char*)page + resultRowPosition->offset);

  SqlFunctionCtx* pCtx = pSup->pCtx;
  SExprInfo*      pExprInfo = pSup->pExprInfo;
  const int32_t*  rowEntryOffset = pSup->rowEntryInfoOffset;

  doUpdateNumOfRows(pCtx, pRow, pSup->numOfExprs, rowEntryOffset);
  if (pRow->numOfRows == 0) {
    releaseBufPage(pBuf, page);
    return 0;
  }

  int32_t size = pBlock->info.capacity;
  while (pBlock->info.rows + pRow->numOfRows > size) {
    size = size * 1.25;
  }

  int32_t code = blockDataEnsureCapacity(pBlock, size);
  if (TAOS_FAILED(code)) {
    releaseBufPage(pBuf, page);
    qError("%s ensure result data capacity failed, code %s", GET_TASKID(pTaskInfo), tstrerror(code));
    T_LONG_JMP(pTaskInfo->env, code);
  }

  doCopyResultToDataBlock(pExprInfo, pSup->numOfExprs, pRow, pCtx, pBlock, rowEntryOffset, pTaskInfo);
1323 1324

  releaseBufPage(pBuf, page);
1325
  pBlock->info.rows += pRow->numOfRows;
1326 1327 1328
  return 0;
}

1329 1330 1331 1332 1333 1334 1335
int32_t doCopyToSDataBlock(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock, SExprSupp* pSup, SDiskbasedBuf* pBuf,
                           SGroupResInfo* pGroupResInfo) {
  SExprInfo*      pExprInfo = pSup->pExprInfo;
  int32_t         numOfExprs = pSup->numOfExprs;
  int32_t*        rowEntryOffset = pSup->rowEntryInfoOffset;
  SqlFunctionCtx* pCtx = pSup->pCtx;

1336
  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
1337

1338
  for (int32_t i = pGroupResInfo->index; i < numOfRows; i += 1) {
L
Liu Jicong 已提交
1339 1340
    SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
    SFilePage*  page = getBufPage(pBuf, pPos->pos.pageId);
1341

1342
    SResultRow* pRow = (SResultRow*)((char*)page + pPos->pos.offset);
1343

H
Haojun Liao 已提交
1344
    doUpdateNumOfRows(pCtx, pRow, numOfExprs, rowEntryOffset);
1345 1346

    // no results, continue to check the next one
1347 1348
    if (pRow->numOfRows == 0) {
      pGroupResInfo->index += 1;
1349
      releaseBufPage(pBuf, page);
1350 1351 1352
      continue;
    }

1353 1354 1355 1356 1357
    if (pBlock->info.groupId == 0) {
      pBlock->info.groupId = pPos->groupId;
    } else {
      // current value belongs to different group, it can't be packed into one datablock
      if (pBlock->info.groupId != pPos->groupId) {
1358
        releaseBufPage(pBuf, page);
1359 1360 1361 1362
        break;
      }
    }

1363
    if (pBlock->info.rows + pRow->numOfRows > pBlock->info.capacity) {
1364
      ASSERT(pBlock->info.rows > 0);
1365
      releaseBufPage(pBuf, page);
1366 1367 1368 1369
      break;
    }

    pGroupResInfo->index += 1;
1370
    doCopyResultToDataBlock(pExprInfo, numOfExprs, pRow, pCtx, pBlock, rowEntryOffset, pTaskInfo);
1371

1372
    releaseBufPage(pBuf, page);
1373
    pBlock->info.rows += pRow->numOfRows;
1374 1375
  }

X
Xiaoyu Wang 已提交
1376 1377
  qDebug("%s result generated, rows:%d, groupId:%" PRIu64, GET_TASKID(pTaskInfo), pBlock->info.rows,
         pBlock->info.groupId);
1378

1379
  blockDataUpdateTsWindow(pBlock, 0);
1380 1381 1382
  return 0;
}

X
Xiaoyu Wang 已提交
1383 1384
void doBuildResultDatablock(SOperatorInfo* pOperator, SOptrBasicInfo* pbInfo, SGroupResInfo* pGroupResInfo,
                            SDiskbasedBuf* pBuf) {
1385
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
1386
  SSDataBlock*   pBlock = pbInfo->pRes;
1387

1388 1389 1390
  // set output datablock version
  pBlock->info.version = pTaskInfo->version;

1391
  blockDataCleanup(pBlock);
1392
  if (!hasRemainResults(pGroupResInfo)) {
1393 1394 1395
    return;
  }

1396 1397
  // clear the existed group id
  pBlock->info.groupId = 0;
1398 1399 1400
  if (!pbInfo->mergeResultBlock) {
    doCopyToSDataBlock(pTaskInfo, pBlock, &pOperator->exprSupp, pBuf, pGroupResInfo);
  } else {
dengyihao's avatar
dengyihao 已提交
1401
    while (hasRemainResults(pGroupResInfo)) {
1402 1403 1404
      doCopyToSDataBlock(pTaskInfo, pBlock, &pOperator->exprSupp, pBuf, pGroupResInfo);
      if (pBlock->info.rows >= pOperator->resultInfo.threshold) {
        break;
1405 1406
      }

1407 1408
      // clearing group id to continue to merge data that belong to different groups
      pBlock->info.groupId = 0;
1409
    }
1410 1411 1412

    // clear the group id info in SSDataBlock, since the client does not need it
    pBlock->info.groupId = 0;
1413 1414 1415
  }
}

L
Liu Jicong 已提交
1416
static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
1417 1418
  int32_t colSize = pColRes->info.bytes * numOfRows;
  return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
L
Liu Jicong 已提交
1419
                                                      colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
1420 1421
}

L
Liu Jicong 已提交
1422 1423
void queryCostStatis(SExecTaskInfo* pTaskInfo) {
  STaskCostInfo* pSummary = &pTaskInfo->cost;
1424

L
Liu Jicong 已提交
1425 1426 1427
  //  uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
  //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
  //  pSummary->hashSize = hashSize;
1428

L
Liu Jicong 已提交
1429 1430 1431 1432 1433 1434 1435 1436 1437
  //  SResultRowPool* p = pTaskInfo->pool;
  //  if (p != NULL) {
  //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
  //    pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
  //  } else {
  //    pSummary->winInfoSize = 0;
  //    pSummary->numOfTimeWindows = 0;
  //  }

1438 1439
  SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
  if (pSummary->pRecoder != NULL) {
1440
    qDebug(
X
Xiaoyu Wang 已提交
1441 1442 1443 1444
        "%s :cost summary: elapsed time:%.2f ms, total blocks:%d, load block SMA:%d, load data block:%d, total "
        "rows:%" PRId64 ", check rows:%" PRId64,
        GET_TASKID(pTaskInfo), pSummary->elapsedTime / 1000.0, pRecorder->totalBlocks, pRecorder->loadBlockStatis,
        pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
1445
  }
1446

L
Liu Jicong 已提交
1447 1448 1449
  // 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);
1450 1451
}

L
Liu Jicong 已提交
1452 1453 1454
// static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
1455
//
L
Liu Jicong 已提交
1456
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
1457
//
L
Liu Jicong 已提交
1458 1459 1460 1461
//   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;
//   }
1462
//
L
Liu Jicong 已提交
1463 1464 1465 1466 1467
//   if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//     pQueryAttr->pos = (int32_t)pQueryAttr->limit.offset;
//   } else {
//     pQueryAttr->pos = pBlockInfo->rows - (int32_t)pQueryAttr->limit.offset - 1;
//   }
1468
//
L
Liu Jicong 已提交
1469
//   assert(pQueryAttr->pos >= 0 && pQueryAttr->pos <= pBlockInfo->rows - 1);
1470
//
L
Liu Jicong 已提交
1471 1472
//   SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//   SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
1473
//
L
Liu Jicong 已提交
1474 1475
//   // update the pQueryAttr->limit.offset value, and pQueryAttr->pos value
//   TSKEY *keys = (TSKEY *) pColInfoData->pData;
1476
//
L
Liu Jicong 已提交
1477 1478 1479
//   // update the offset value
//   pTableQueryInfo->lastKey = keys[pQueryAttr->pos];
//   pQueryAttr->limit.offset = 0;
1480
//
L
Liu Jicong 已提交
1481
//   int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
1482
//
L
Liu Jicong 已提交
1483 1484 1485 1486
//   //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);
// }
1487

L
Liu Jicong 已提交
1488 1489
// void skipBlocks(STaskRuntimeEnv *pRuntimeEnv) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
1490
//
L
Liu Jicong 已提交
1491 1492 1493
//   if (pQueryAttr->limit.offset <= 0 || pQueryAttr->numOfFilterCols > 0) {
//     return;
//   }
1494
//
L
Liu Jicong 已提交
1495 1496
//   pQueryAttr->pos = 0;
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
1497
//
L
Liu Jicong 已提交
1498 1499
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
//   TsdbQueryHandleT pTsdbReadHandle = pRuntimeEnv->pTsdbReadHandle;
1500
//
L
Liu Jicong 已提交
1501 1502 1503
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pTsdbReadHandle)) {
//     if (isTaskKilled(pRuntimeEnv->qinfo)) {
1504
//       T_LONG_JMP(pRuntimeEnv->env, TSDB_CODE_TSC_QUERY_CANCELLED);
L
Liu Jicong 已提交
1505
//     }
1506
//
L
Liu Jicong 已提交
1507
//     tsdbRetrieveDataBlockInfo(pTsdbReadHandle, &blockInfo);
1508
//
L
Liu Jicong 已提交
1509 1510 1511 1512
//     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;
1513
//
L
Liu Jicong 已提交
1514 1515 1516 1517 1518 1519 1520
//       //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;
//     }
//   }
1521
//
L
Liu Jicong 已提交
1522
//   if (terrno != TSDB_CODE_SUCCESS) {
1523
//     T_LONG_JMP(pRuntimeEnv->env, terrno);
L
Liu Jicong 已提交
1524 1525 1526 1527 1528 1529 1530
//   }
// }

// static TSKEY doSkipIntervalProcess(STaskRuntimeEnv* pRuntimeEnv, STimeWindow* win, SDataBlockInfo* pBlockInfo,
// STableQueryInfo* pTableQueryInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
1531
//
L
Liu Jicong 已提交
1532 1533 1534
//   assert(pQueryAttr->limit.offset == 0);
//   STimeWindow tw = *win;
//   getNextTimeWindow(pQueryAttr, &tw);
1535
//
L
Liu Jicong 已提交
1536 1537
//   if ((tw.skey <= pBlockInfo->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) ||
//       (tw.ekey >= pBlockInfo->window.skey && !QUERY_IS_ASC_QUERY(pQueryAttr))) {
1538
//
L
Liu Jicong 已提交
1539 1540 1541 1542
//     // 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);
1543
//
L
Liu Jicong 已提交
1544 1545 1546 1547
//     tw = *win;
//     int32_t startPos =
//         getNextQualifiedWindow(pQueryAttr, &tw, pBlockInfo, pColInfoData->pData, binarySearchForKey, -1);
//     assert(startPos >= 0);
1548
//
L
Liu Jicong 已提交
1549 1550
//     // set the abort info
//     pQueryAttr->pos = startPos;
1551
//
L
Liu Jicong 已提交
1552 1553 1554 1555
//     // reset the query start timestamp
//     pTableQueryInfo->win.skey = ((TSKEY *)pColInfoData->pData)[startPos];
//     pQueryAttr->window.skey = pTableQueryInfo->win.skey;
//     TSKEY key = pTableQueryInfo->win.skey;
1556
//
L
Liu Jicong 已提交
1557 1558
//     pWindowResInfo->prevSKey = tw.skey;
//     int32_t index = pRuntimeEnv->resultRowInfo.curIndex;
1559
//
L
Liu Jicong 已提交
1560 1561
//     int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
//     pRuntimeEnv->resultRowInfo.curIndex = index;  // restore the window index
1562
//
L
Liu Jicong 已提交
1563 1564 1565 1566
//     //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);
1567
//
L
Liu Jicong 已提交
1568 1569 1570 1571 1572
//     return key;
//   } else {  // do nothing
//     pQueryAttr->window.skey      = tw.skey;
//     pWindowResInfo->prevSKey = tw.skey;
//     pTableQueryInfo->lastKey = tw.skey;
1573
//
L
Liu Jicong 已提交
1574 1575
//     return tw.skey;
//   }
1576
//
L
Liu Jicong 已提交
1577 1578 1579 1580 1581 1582 1583 1584 1585 1586
//   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);
//   }
1587
//
L
Liu Jicong 已提交
1588 1589 1590 1591 1592
//   // 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;
//   }
1593
//
L
Liu Jicong 已提交
1594 1595 1596 1597 1598 1599 1600
//   /*
//    * 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);
1601
//
L
Liu Jicong 已提交
1602 1603
//   STimeWindow w = TSWINDOW_INITIALIZER;
//   bool ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
1604
//
L
Liu Jicong 已提交
1605 1606
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
//   STableQueryInfo *pTableQueryInfo = pRuntimeEnv->current;
1607
//
L
Liu Jicong 已提交
1608 1609 1610
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pRuntimeEnv->pTsdbReadHandle)) {
//     tsdbRetrieveDataBlockInfo(pRuntimeEnv->pTsdbReadHandle, &blockInfo);
1611
//
L
Liu Jicong 已提交
1612 1613 1614 1615 1616 1617 1618 1619 1620
//     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;
//     }
1621
//
L
Liu Jicong 已提交
1622 1623
//     // the first time window
//     STimeWindow win = getActiveTimeWindow(pWindowResInfo, pWindowResInfo->prevSKey, pQueryAttr);
1624
//
L
Liu Jicong 已提交
1625 1626
//     while (pQueryAttr->limit.offset > 0) {
//       STimeWindow tw = win;
1627
//
L
Liu Jicong 已提交
1628 1629 1630
//       if ((win.ekey <= blockInfo.window.ekey && ascQuery) || (win.ekey >= blockInfo.window.skey && !ascQuery)) {
//         pQueryAttr->limit.offset -= 1;
//         pWindowResInfo->prevSKey = win.skey;
1631
//
L
Liu Jicong 已提交
1632 1633 1634 1635 1636 1637
//         // 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;
//         }
//       }
1638
//
L
Liu Jicong 已提交
1639 1640 1641 1642
//       if (pQueryAttr->limit.offset == 0) {
//         *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
//         return true;
//       }
1643
//
L
Liu Jicong 已提交
1644 1645
//       // current window does not ended in current data block, try next data block
//       getNextTimeWindow(pQueryAttr, &tw);
1646
//
L
Liu Jicong 已提交
1647 1648 1649 1650 1651 1652 1653 1654 1655
//       /*
//        * 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)) {
1656
//
L
Liu Jicong 已提交
1657 1658
//         SArray *pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//         SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
1659
//
L
Liu Jicong 已提交
1660 1661 1662
//         if ((win.ekey > blockInfo.window.ekey && ascQuery) || (win.ekey < blockInfo.window.skey && !ascQuery)) {
//           pQueryAttr->limit.offset -= 1;
//         }
1663
//
L
Liu Jicong 已提交
1664 1665 1666 1667 1668 1669 1670 1671
//         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);
1672
//
L
Liu Jicong 已提交
1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683
//           // 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.
//       }
//     }
//   }
1684
//
L
Liu Jicong 已提交
1685 1686
//   // check for error
//   if (terrno != TSDB_CODE_SUCCESS) {
1687
//     T_LONG_JMP(pRuntimeEnv->env, terrno);
L
Liu Jicong 已提交
1688
//   }
1689
//
L
Liu Jicong 已提交
1690 1691
//   return true;
// }
1692

1693
int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t num) {
H
Haojun Liao 已提交
1694
  if (p->pDownstream == NULL) {
H
Haojun Liao 已提交
1695
    assert(p->numOfDownstream == 0);
1696 1697
  }

wafwerar's avatar
wafwerar 已提交
1698
  p->pDownstream = taosMemoryCalloc(1, num * POINTER_BYTES);
1699 1700 1701 1702 1703 1704 1705
  if (p->pDownstream == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

  memcpy(p->pDownstream, pDownstream, num * POINTER_BYTES);
  p->numOfDownstream = num;
  return TSDB_CODE_SUCCESS;
1706 1707
}

wmmhello's avatar
wmmhello 已提交
1708
static void doDestroyTableList(STableListInfo* pTableqinfoList);
1709

1710 1711 1712 1713
typedef struct SFetchRspHandleWrapper {
  uint32_t exchangeId;
  int32_t  sourceIndex;
} SFetchRspHandleWrapper;
1714

D
dapan1121 已提交
1715
int32_t loadRemoteDataCallback(void* param, SDataBuf* pMsg, int32_t code) {
X
Xiaoyu Wang 已提交
1716
  SFetchRspHandleWrapper* pWrapper = (SFetchRspHandleWrapper*)param;
1717 1718 1719 1720

  SExchangeInfo* pExchangeInfo = taosAcquireRef(exchangeObjRefPool, pWrapper->exchangeId);
  if (pExchangeInfo == NULL) {
    qWarn("failed to acquire exchange operator, since it may have been released");
1721
    taosMemoryFree(pMsg->pData);
1722 1723 1724
    return TSDB_CODE_SUCCESS;
  }

X
Xiaoyu Wang 已提交
1725
  int32_t          index = pWrapper->sourceIndex;
1726
  SSourceDataInfo* pSourceDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, index);
1727

H
Haojun Liao 已提交
1728 1729
  if (code == TSDB_CODE_SUCCESS) {
    pSourceDataInfo->pRsp = pMsg->pData;
1730

H
Haojun Liao 已提交
1731 1732
    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
    pRsp->numOfRows = htonl(pRsp->numOfRows);
dengyihao's avatar
dengyihao 已提交
1733
    pRsp->compLen = htonl(pRsp->compLen);
1734
    pRsp->numOfCols = htonl(pRsp->numOfCols);
dengyihao's avatar
dengyihao 已提交
1735
    pRsp->useconds = htobe64(pRsp->useconds);
1736
    pRsp->numOfBlocks = htonl(pRsp->numOfBlocks);
1737

1738
    ASSERT(pRsp != NULL);
H
Haojun Liao 已提交
1739 1740
    qDebug("%s fetch rsp received, index:%d, blocks:%d, rows:%d", pSourceDataInfo->taskId, index, pRsp->numOfBlocks,
           pRsp->numOfRows);
H
Haojun Liao 已提交
1741
  } else {
1742
    taosMemoryFree(pMsg->pData);
H
Haojun Liao 已提交
1743
    pSourceDataInfo->code = code;
D
dapan1121 已提交
1744
    qDebug("%s fetch rsp received, index:%d, error:%d", pSourceDataInfo->taskId, index, tstrerror(code));
H
Haojun Liao 已提交
1745
  }
H
Haojun Liao 已提交
1746

H
Haojun Liao 已提交
1747
  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
1748 1749 1750 1751

  tsem_post(&pExchangeInfo->ready);
  taosReleaseRef(exchangeObjRefPool, pWrapper->exchangeId);

wmmhello's avatar
wmmhello 已提交
1752
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
1753 1754
}

D
dapan1121 已提交
1755
void qProcessRspMsg(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
S
Shengliang Guan 已提交
1756 1757
  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->info.ahandle;
  assert(pMsg->info.ahandle != NULL);
H
Haojun Liao 已提交
1758 1759 1760 1761

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

  if (pMsg->contLen > 0) {
wafwerar's avatar
wafwerar 已提交
1762
    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
H
Haojun Liao 已提交
1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773
    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);
1774 1775
}

L
Liu Jicong 已提交
1776
static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
1777
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
1778

wafwerar's avatar
wafwerar 已提交
1779
  SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
1780 1781 1782 1783
  if (NULL == pMsg) {
    pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return pTaskInfo->code;
  }
1784

L
Liu Jicong 已提交
1785 1786
  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
1787

1788 1789
  ASSERT(pDataInfo->status == EX_SOURCE_DATA_NOT_READY);

1790 1791 1792
  qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", execId:%d, %d/%" PRIzu,
         GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId, pSource->execId,
         sourceIndex, totalSources);
1793 1794 1795 1796 1797

  pMsg->header.vgId = htonl(pSource->addr.nodeId);
  pMsg->sId = htobe64(pSource->schedId);
  pMsg->taskId = htobe64(pSource->taskId);
  pMsg->queryId = htobe64(pTaskInfo->id.queryId);
D
dapan1121 已提交
1798
  pMsg->execId = htonl(pSource->execId);
1799 1800

  // send the fetch remote task result reques
wafwerar's avatar
wafwerar 已提交
1801
  SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
1802
  if (NULL == pMsgSendInfo) {
wafwerar's avatar
wafwerar 已提交
1803
    taosMemoryFreeClear(pMsg);
1804 1805 1806
    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
1807 1808
  }

1809
  SFetchRspHandleWrapper* pWrapper = taosMemoryCalloc(1, sizeof(SFetchRspHandleWrapper));
X
Xiaoyu Wang 已提交
1810
  pWrapper->exchangeId = pExchangeInfo->self;
1811 1812 1813
  pWrapper->sourceIndex = sourceIndex;

  pMsgSendInfo->param = pWrapper;
D
dapan1121 已提交
1814
  pMsgSendInfo->paramFreeFp = taosMemoryFree;
1815 1816
  pMsgSendInfo->msgInfo.pData = pMsg;
  pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
L
Liu Jicong 已提交
1817
  pMsgSendInfo->msgType = pSource->fetchMsgType;
1818
  pMsgSendInfo->fp = loadRemoteDataCallback;
1819

1820
  int64_t transporterId = 0;
L
Liu Jicong 已提交
1821
  int32_t code = asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
1822 1823 1824
  return TSDB_CODE_SUCCESS;
}

1825 1826 1827 1828 1829 1830 1831 1832
void updateLoadRemoteInfo(SLoadRemoteDataInfo* pInfo, int32_t numOfRows, int32_t dataLen, int64_t startTs,
                          SOperatorInfo* pOperator) {
  pInfo->totalRows += numOfRows;
  pInfo->totalSize += dataLen;
  pInfo->totalElapsed += (taosGetTimestampUs() - startTs);
  pOperator->resultInfo.totalRows += numOfRows;
}

H
Haojun Liao 已提交
1833
int32_t extractDataBlockFromFetchRsp(SSDataBlock* pRes, char* pData, SArray* pColList, char** pNextStart) {
H
Haojun Liao 已提交
1834
  if (pColList == NULL) {  // data from other sources
1835
    blockDataCleanup(pRes);
dengyihao's avatar
dengyihao 已提交
1836
    *pNextStart = (char*)blockDecode(pRes, pData);
H
Haojun Liao 已提交
1837
  } else {  // extract data according to pColList
1838 1839 1840 1841 1842
    char* pStart = pData;

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

1843
    // todo refactor:extract method
1844
    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
dengyihao's avatar
dengyihao 已提交
1845
    for (int32_t i = 0; i < numOfCols; ++i) {
1846 1847 1848 1849 1850 1851 1852
      SSysTableSchema* p = (SSysTableSchema*)pStart;

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

1853
    SSDataBlock* pBlock = createDataBlock();
dengyihao's avatar
dengyihao 已提交
1854
    for (int32_t i = 0; i < numOfCols; ++i) {
1855 1856
      SColumnInfoData idata = createColumnInfoData(pSchema[i].type, pSchema[i].bytes, pSchema[i].colId);
      blockDataAppendColInfo(pBlock, &idata);
1857 1858
    }

1859
    blockDecode(pBlock, pStart);
1860
    blockDataEnsureCapacity(pRes, pBlock->info.rows);
1861

H
Haojun Liao 已提交
1862
    // data from mnode
1863
    pRes->info.rows = pBlock->info.rows;
1864 1865
    relocateColumnData(pRes, pColList, pBlock->pDataBlock, false);
    blockDataDestroy(pBlock);
1866
  }
1867

1868 1869
  // todo move this to time window aggregator, since the primary timestamp may not be known by exchange operator.
  blockDataUpdateTsWindow(pRes, 0);
1870 1871
  return TSDB_CODE_SUCCESS;
}
1872

L
Liu Jicong 已提交
1873 1874
static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
1875
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
1876

1877
  int64_t              el = taosGetTimestampUs() - startTs;
H
Haojun Liao 已提交
1878
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
1879

H
Haojun Liao 已提交
1880
  pLoadInfo->totalElapsed += el;
H
Haojun Liao 已提交
1881

1882
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
L
Liu Jicong 已提交
1883 1884 1885
  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);
1886 1887 1888 1889 1890

  doSetOperatorCompleted(pOperator);
  return NULL;
}

1891 1892
static void concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
                                           SExecTaskInfo* pTaskInfo) {
1893 1894 1895 1896 1897 1898 1899 1900
  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);
1901
      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
1902
        completed += 1;
H
Haojun Liao 已提交
1903 1904
        continue;
      }
1905

1906
      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
1907 1908 1909
        continue;
      }

1910 1911 1912 1913 1914
      if (pDataInfo->code != TSDB_CODE_SUCCESS) {
        code = pDataInfo->code;
        goto _error;
      }

L
Liu Jicong 已提交
1915
      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
1916
      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
1917

H
Haojun Liao 已提交
1918
      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
1919
      if (pRsp->numOfRows == 0) {
1920 1921
        qDebug("%s vgId:%d, taskId:0x%" PRIx64 " execId:%d index:%d completed, rowsOfSource:%" PRIu64
               ", totalRows:%" PRIu64 ", completed:%d try next %d/%" PRIzu,
D
dapan1121 已提交
1922
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pDataInfo->totalRows,
1923
               pExchangeInfo->loadInfo.totalRows, completed + 1, i + 1, totalSources);
1924
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
1925
        completed += 1;
D
dapan1121 已提交
1926
        taosMemoryFreeClear(pDataInfo->pRsp);
1927 1928
        continue;
      }
H
Haojun Liao 已提交
1929

1930
      SRetrieveTableRsp* pRetrieveRsp = pDataInfo->pRsp;
dengyihao's avatar
dengyihao 已提交
1931 1932 1933
      int32_t            index = 0;
      char*              pStart = pRetrieveRsp->data;
      while (index++ < pRetrieveRsp->numOfBlocks) {
1934
        SSDataBlock* pb = createOneDataBlock(pExchangeInfo->pDummyBlock, false);
H
Haojun Liao 已提交
1935
        code = extractDataBlockFromFetchRsp(pb, pStart, NULL, &pStart);
1936 1937 1938 1939 1940 1941
        if (code != 0) {
          taosMemoryFreeClear(pDataInfo->pRsp);
          goto _error;
        }

        taosArrayPush(pExchangeInfo->pResultBlockList, &pb);
1942 1943
      }

1944
      updateLoadRemoteInfo(pLoadInfo, pRetrieveRsp->numOfRows, pRetrieveRsp->compLen, startTs, pOperator);
1945

1946
      if (pRsp->completed == 1) {
dengyihao's avatar
dengyihao 已提交
1947 1948 1949 1950
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
               " execId:%d"
               " index:%d completed, blocks:%d, numOfRows:%d, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
               ", total:%.2f Kb,"
1951
               " completed:%d try next %d/%" PRIzu,
H
Haojun Liao 已提交
1952 1953 1954
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, i, pRsp->numOfBlocks,
               pRsp->numOfRows, pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize / 1024.0,
               completed + 1, i + 1, totalSources);
1955
        completed += 1;
1956
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
1957
      } else {
dengyihao's avatar
dengyihao 已提交
1958 1959 1960 1961
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64
               " execId:%d blocks:%d, numOfRows:%d, totalRows:%" PRIu64 ", total:%.2f Kb",
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRsp->numOfBlocks,
               pRsp->numOfRows, pLoadInfo->totalRows, pLoadInfo->totalSize / 1024.0);
1962 1963
      }

1964 1965
      taosMemoryFreeClear(pDataInfo->pRsp);

1966 1967
      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
1968 1969
        code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
        if (code != TSDB_CODE_SUCCESS) {
1970
          taosMemoryFreeClear(pDataInfo->pRsp);
1971 1972 1973 1974
          goto _error;
        }
      }

1975
      return;
1976 1977
    }

1978
    if (completed == totalSources) {
1979 1980
      setAllSourcesCompleted(pOperator, startTs);
      return;
1981
    }
H
Haojun Liao 已提交
1982 1983

    sched_yield();
1984 1985 1986 1987 1988 1989
  }

_error:
  pTaskInfo->code = code;
}

L
Liu Jicong 已提交
1990 1991 1992
static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
1993

L
Liu Jicong 已提交
1994
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
1995 1996 1997
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
1998
  for (int32_t i = 0; i < totalSources; ++i) {
1999 2000
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
2001 2002
      pTaskInfo->code = code;
      return code;
2003 2004 2005 2006
    }
  }

  int64_t endTs = taosGetTimestampUs();
2007
  qDebug("%s send all fetch requests to %" PRIzu " sources completed, elapsed:%.2fms", GET_TASKID(pTaskInfo),
X
Xiaoyu Wang 已提交
2008
         totalSources, (endTs - startTs) / 1000.0);
2009

2010
  pOperator->status = OP_RES_TO_RETURN;
H
Haojun Liao 已提交
2011
  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
2012

2013
  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
2014
  return TSDB_CODE_SUCCESS;
2015 2016
}

2017
static int32_t seqLoadRemoteData(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
2018 2019
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
2020

L
Liu Jicong 已提交
2021
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
2022
  int64_t startTs = taosGetTimestampUs();
2023

L
Liu Jicong 已提交
2024
  while (1) {
2025
    if (pExchangeInfo->current >= totalSources) {
2026 2027
      setAllSourcesCompleted(pOperator, startTs);
      return TSDB_CODE_SUCCESS;
2028
    }
2029

2030 2031 2032
    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
    tsem_wait(&pExchangeInfo->ready);

dengyihao's avatar
dengyihao 已提交
2033
    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
X
Xiaoyu Wang 已提交
2034
    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
2035

H
Haojun Liao 已提交
2036
    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
2037 2038
      qError("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d error happens, code:%s", GET_TASKID(pTaskInfo),
             pSource->addr.nodeId, pSource->taskId, pSource->execId, tstrerror(pDataInfo->code));
H
Haojun Liao 已提交
2039
      pOperator->pTaskInfo->code = pDataInfo->code;
2040
      return pOperator->pTaskInfo->code;
H
Haojun Liao 已提交
2041 2042
    }

L
Liu Jicong 已提交
2043
    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
H
Haojun Liao 已提交
2044
    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
2045
    if (pRsp->numOfRows == 0) {
2046 2047
      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " execId:%d %d of total completed, rowsOfSource:%" PRIu64
             ", totalRows:%" PRIu64 " try next",
D
dapan1121 已提交
2048
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pExchangeInfo->current + 1,
H
Haojun Liao 已提交
2049
             pDataInfo->totalRows, pLoadInfo->totalRows);
H
Haojun Liao 已提交
2050

2051
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
2052
      pExchangeInfo->current += 1;
D
dapan1121 已提交
2053
      taosMemoryFreeClear(pDataInfo->pRsp);
2054 2055
      continue;
    }
H
Haojun Liao 已提交
2056

2057 2058 2059
    SRetrieveTableRsp* pRetrieveRsp = pDataInfo->pRsp;

    char*   pStart = pRetrieveRsp->data;
H
Haojun Liao 已提交
2060
    int32_t code = extractDataBlockFromFetchRsp(NULL, pStart, NULL, &pStart);
2061 2062

    if (pRsp->completed == 1) {
D
dapan1121 已提交
2063
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, rowsOfSource:%" PRIu64
L
Liu Jicong 已提交
2064
             ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
2065
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRetrieveRsp->numOfRows,
2066 2067
             pDataInfo->totalRows, pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1,
             totalSources);
2068

2069
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
2070 2071
      pExchangeInfo->current += 1;
    } else {
D
dapan1121 已提交
2072
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " execId:%d numOfRows:%d, totalRows:%" PRIu64
L
Liu Jicong 已提交
2073
             ", totalBytes:%" PRIu64,
2074
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pSource->execId, pRetrieveRsp->numOfRows,
2075
             pLoadInfo->totalRows, pLoadInfo->totalSize);
2076 2077
    }

2078 2079 2080
    updateLoadRemoteInfo(pLoadInfo, pRetrieveRsp->numOfRows, pRetrieveRsp->compLen, startTs, pOperator);
    pDataInfo->totalRows += pRetrieveRsp->numOfRows;

2081
    taosMemoryFreeClear(pDataInfo->pRsp);
2082
    return TSDB_CODE_SUCCESS;
2083
  }
2084 2085
}

L
Liu Jicong 已提交
2086
static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
2087
  if (OPTR_IS_OPENED(pOperator)) {
H
Haojun Liao 已提交
2088 2089 2090
    return TSDB_CODE_SUCCESS;
  }

2091 2092
  int64_t st = taosGetTimestampUs();

L
Liu Jicong 已提交
2093
  SExchangeInfo* pExchangeInfo = pOperator->info;
2094
  if (!pExchangeInfo->seqLoadData) {
H
Haojun Liao 已提交
2095 2096 2097 2098 2099 2100
    int32_t code = prepareConcurrentlyLoad(pOperator);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
    }
  }

2101
  OPTR_SET_OPENED(pOperator);
2102
  pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
H
Haojun Liao 已提交
2103 2104 2105
  return TSDB_CODE_SUCCESS;
}

2106 2107 2108 2109 2110
static void freeBlock(void* pParam) {
  SSDataBlock* pBlock = *(SSDataBlock**)pParam;
  blockDataDestroy(pBlock);
}

2111
static SSDataBlock* doLoadRemoteDataImpl(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
2112 2113
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
2114

2115
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
2116
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
2117 2118
    return NULL;
  }
2119

2120
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
H
Haojun Liao 已提交
2121

2122
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
2123
  if (pOperator->status == OP_EXEC_DONE) {
L
Liu Jicong 已提交
2124 2125 2126
    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);
2127 2128 2129
    return NULL;
  }

2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142
  size_t size = taosArrayGetSize(pExchangeInfo->pResultBlockList);
  if (size == 0 || pExchangeInfo->rspBlockIndex >= size) {
    pExchangeInfo->rspBlockIndex = 0;
    taosArrayClearEx(pExchangeInfo->pResultBlockList, freeBlock);
    if (pExchangeInfo->seqLoadData) {
      seqLoadRemoteData(pOperator);
    } else {
      concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
    }

    if (taosArrayGetSize(pExchangeInfo->pResultBlockList) == 0) {
      return NULL;
    }
2143
  }
2144 2145 2146

  // we have buffered retrieved datablock, return it directly
  return taosArrayGetP(pExchangeInfo->pResultBlockList, pExchangeInfo->rspBlockIndex++);
H
Haojun Liao 已提交
2147
}
2148

2149 2150 2151 2152 2153 2154 2155 2156
static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;

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

L
Liu Jicong 已提交
2157
  while (1) {
2158 2159 2160 2161 2162 2163 2164
    SSDataBlock* pBlock = doLoadRemoteDataImpl(pOperator);
    if (pBlock == NULL) {
      return NULL;
    }

    SLimitInfo* pLimitInfo = &pExchangeInfo->limitInfo;
    if (hasLimitOffsetInfo(pLimitInfo)) {
2165
      int32_t status = handleLimitOffset(pOperator, pLimitInfo, pBlock, false);
2166 2167 2168
      if (status == PROJECT_RETRIEVE_CONTINUE) {
        continue;
      } else if (status == PROJECT_RETRIEVE_DONE) {
2169
        size_t rows = pBlock->info.rows;
2170 2171 2172 2173 2174 2175
        pExchangeInfo->limitInfo.numOfOutputRows += rows;

        if (rows == 0) {
          doSetOperatorCompleted(pOperator);
          return NULL;
        } else {
2176
          return pBlock;
2177 2178 2179
        }
      }
    } else {
2180
      return pBlock;
2181 2182 2183 2184
    }
  }
}

2185
static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo, const char* id) {
2186
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
H
Haojun Liao 已提交
2187 2188
  if (pInfo->pSourceDataInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
2189 2190
  }

L
Liu Jicong 已提交
2191
  for (int32_t i = 0; i < numOfSources; ++i) {
2192
    SSourceDataInfo dataInfo = {0};
H
Haojun Liao 已提交
2193
    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
2194
    dataInfo.taskId = id;
L
Liu Jicong 已提交
2195
    dataInfo.index = i;
X
Xiaoyu Wang 已提交
2196
    SSourceDataInfo* pDs = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
2197
    if (pDs == NULL) {
H
Haojun Liao 已提交
2198 2199 2200 2201 2202 2203 2204 2205
      taosArrayDestroy(pInfo->pSourceDataInfo);
      return TSDB_CODE_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
}

2206
static int32_t initExchangeOperator(SExchangePhysiNode* pExNode, SExchangeInfo* pInfo, const char* id) {
2207
  size_t numOfSources = LIST_LENGTH(pExNode->pSrcEndPoints);
H
Haojun Liao 已提交
2208

2209
  if (numOfSources == 0) {
X
Xiaoyu Wang 已提交
2210
    qError("%s invalid number: %d of sources in exchange operator", id, (int32_t)numOfSources);
2211 2212 2213
    return TSDB_CODE_INVALID_PARA;
  }

H
Haojun Liao 已提交
2214
  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
wmmhello's avatar
wmmhello 已提交
2215
  if (pInfo->pSources == NULL) {
2216
    return TSDB_CODE_OUT_OF_MEMORY;
H
Haojun Liao 已提交
2217 2218
  }

L
Liu Jicong 已提交
2219
  for (int32_t i = 0; i < numOfSources; ++i) {
D
dapan1121 已提交
2220
    SDownstreamSourceNode* pNode = (SDownstreamSourceNode*)nodesListGetNode((SNodeList*)pExNode->pSrcEndPoints, i);
H
Haojun Liao 已提交
2221 2222
    taosArrayPush(pInfo->pSources, pNode);
  }
2223

2224
  initLimitInfo(pExNode->node.pLimit, pExNode->node.pSlimit, &pInfo->limitInfo);
2225 2226
  pInfo->self = taosAddRef(exchangeObjRefPool, pInfo);

2227
  return initDataSource(numOfSources, pInfo, id);
2228 2229 2230 2231 2232 2233
}

SOperatorInfo* createExchangeOperatorInfo(void* pTransporter, SExchangePhysiNode* pExNode, SExecTaskInfo* pTaskInfo) {
  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
  if (pInfo == NULL || pOperator == NULL) {
H
Haojun Liao 已提交
2234
    goto _error;
2235
  }
H
Haojun Liao 已提交
2236

2237
  int32_t code = initExchangeOperator(pExNode, pInfo, GET_TASKID(pTaskInfo));
2238 2239 2240
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
2241 2242

  tsem_init(&pInfo->ready, 0, 0);
2243 2244
  pInfo->pDummyBlock = createResDataBlock(pExNode->node.pOutputDataBlockDesc);
  pInfo->pResultBlockList = taosArrayInit(1, POINTER_BYTES);
2245

2246
  pInfo->seqLoadData = false;
2247
  pInfo->pTransporter = pTransporter;
2248

2249
  pOperator->name = "ExchangeOperator";
X
Xiaoyu Wang 已提交
2250
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
X
Xiaoyu Wang 已提交
2251
  pOperator->blocking = false;
2252 2253
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
2254
  pOperator->exprSupp.numOfExprs = taosArrayGetSize(pInfo->pDummyBlock->pDataBlock);
X
Xiaoyu Wang 已提交
2255
  pOperator->pTaskInfo = pTaskInfo;
2256

L
Liu Jicong 已提交
2257 2258
  pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                         destroyExchangeOperatorInfo, NULL, NULL, NULL);
2259
  return pOperator;
H
Haojun Liao 已提交
2260

L
Liu Jicong 已提交
2261
_error:
H
Haojun Liao 已提交
2262
  if (pInfo != NULL) {
2263
    doDestroyExchangeOperatorInfo(pInfo);
H
Haojun Liao 已提交
2264 2265
  }

wafwerar's avatar
wafwerar 已提交
2266
  taosMemoryFreeClear(pOperator);
2267
  pTaskInfo->code = code;
H
Haojun Liao 已提交
2268
  return NULL;
2269 2270
}

dengyihao's avatar
dengyihao 已提交
2271 2272
static int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                                const char* pKey);
2273

L
Liu Jicong 已提交
2274
static bool needToMerge(SSDataBlock* pBlock, SArray* groupInfo, char** buf, int32_t rowIndex) {
2275 2276 2277 2278
  size_t size = taosArrayGetSize(groupInfo);
  if (size == 0) {
    return true;
  }
2279

2280 2281
  for (int32_t i = 0; i < size; ++i) {
    int32_t* index = taosArrayGet(groupInfo, i);
2282

2283
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
L
Liu Jicong 已提交
2284
    bool             isNull = colDataIsNull(pColInfo, rowIndex, pBlock->info.rows, NULL);
2285

2286 2287 2288
    if ((isNull && buf[i] != NULL) || (!isNull && buf[i] == NULL)) {
      return false;
    }
2289

2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302
    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;
      }
2303 2304 2305
    }
  }

2306
  return 0;
2307 2308
}

2309
static bool saveCurrentTuple(char** rowColData, SArray* pColumnList, SSDataBlock* pBlock, int32_t rowIndex) {
L
Liu Jicong 已提交
2310
  int32_t size = (int32_t)taosArrayGetSize(pColumnList);
2311

L
Liu Jicong 已提交
2312 2313
  for (int32_t i = 0; i < size; ++i) {
    int32_t*         index = taosArrayGet(pColumnList, i);
2314
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
H
Haojun Liao 已提交
2315

2316 2317 2318
    char* data = colDataGetData(pColInfo, rowIndex);
    memcpy(rowColData[i], data, colDataGetLength(pColInfo, rowIndex));
  }
2319

2320 2321
  return true;
}
2322

X
Xiaoyu Wang 已提交
2323
int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t* order, int32_t* scanFlag) {
2324
  // todo add more information about exchange operation
2325
  int32_t type = pOperator->operatorType;
X
Xiaoyu Wang 已提交
2326
  if (type == QUERY_NODE_PHYSICAL_PLAN_EXCHANGE || type == QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN ||
2327
      type == QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN || type == QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN ||
2328
      type == QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN || type == QUERY_NODE_PHYSICAL_PLAN_LAST_ROW_SCAN) {
2329 2330 2331
    *order = TSDB_ORDER_ASC;
    *scanFlag = MAIN_SCAN;
    return TSDB_CODE_SUCCESS;
2332
  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
2333 2334 2335 2336
    STableScanInfo* pTableScanInfo = pOperator->info;
    *order = pTableScanInfo->cond.order;
    *scanFlag = pTableScanInfo->scanFlag;
    return TSDB_CODE_SUCCESS;
2337 2338 2339 2340 2341
  } else if (type == QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN) {
    STableMergeScanInfo* pTableScanInfo = pOperator->info;
    *order = pTableScanInfo->cond.order;
    *scanFlag = pTableScanInfo->scanFlag;
    return TSDB_CODE_SUCCESS;
2342
  } else {
H
Haojun Liao 已提交
2343
    if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
2344
      return TSDB_CODE_INVALID_PARA;
H
Haojun Liao 已提交
2345
    } else {
2346
      return getTableScanInfo(pOperator->pDownstream[0], order, scanFlag);
2347 2348 2349
    }
  }
}
2350

2351
// this is a blocking operator
L
Liu Jicong 已提交
2352
static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
H
Haojun Liao 已提交
2353 2354
  if (OPTR_IS_OPENED(pOperator)) {
    return TSDB_CODE_SUCCESS;
2355 2356
  }

H
Haojun Liao 已提交
2357
  SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
2358
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
2359

2360 2361
  SExprSupp*     pSup = &pOperator->exprSupp;
  SOperatorInfo* downstream = pOperator->pDownstream[0];
2362

2363 2364
  int64_t st = taosGetTimestampUs();

2365 2366 2367
  int32_t order = TSDB_ORDER_ASC;
  int32_t scanFlag = MAIN_SCAN;

H
Haojun Liao 已提交
2368
  while (1) {
2369
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
2370 2371 2372 2373
    if (pBlock == NULL) {
      break;
    }

2374 2375
    int32_t code = getTableScanInfo(pOperator, &order, &scanFlag);
    if (code != TSDB_CODE_SUCCESS) {
2376
      T_LONG_JMP(pTaskInfo->env, code);
2377
    }
2378

2379
    // there is an scalar expression that needs to be calculated before apply the group aggregation.
2380 2381 2382
    if (pAggInfo->scalarExprSup.pExprInfo != NULL) {
      SExprSupp* pSup1 = &pAggInfo->scalarExprSup;
      code = projectApplyFunctions(pSup1->pExprInfo, pBlock, pBlock, pSup1->pCtx, pSup1->numOfExprs, NULL);
2383
      if (code != TSDB_CODE_SUCCESS) {
2384
        T_LONG_JMP(pTaskInfo->env, code);
2385
      }
2386 2387
    }

2388
    // the pDataBlock are always the same one, no need to call this again
2389
    setExecutionContext(pOperator, pOperator->exprSupp.numOfExprs, pBlock->info.groupId);
2390
    setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, true);
2391
    code = doAggregateImpl(pOperator, pSup->pCtx);
2392
    if (code != 0) {
2393
      T_LONG_JMP(pTaskInfo->env, code);
2394
    }
2395 2396
  }

2397
  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, 0);
H
Haojun Liao 已提交
2398
  OPTR_SET_OPENED(pOperator);
2399

2400
  pOperator->cost.openCost = (taosGetTimestampUs() - st) / 1000.0;
H
Haojun Liao 已提交
2401 2402 2403
  return TSDB_CODE_SUCCESS;
}

2404
static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
2405
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
2406 2407 2408 2409 2410 2411
  SOptrBasicInfo*   pInfo = &pAggInfo->binfo;

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

L
Liu Jicong 已提交
2412
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
2413
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
H
Haojun Liao 已提交
2414
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
2415
    doSetOperatorCompleted(pOperator);
H
Haojun Liao 已提交
2416 2417 2418
    return NULL;
  }

H
Haojun Liao 已提交
2419
  blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
S
slzhou 已提交
2420 2421
  while (1) {
    doBuildResultDatablock(pOperator, pInfo, &pAggInfo->groupResInfo, pAggInfo->aggSup.pResultBuf);
2422
    doFilter(pAggInfo->pCondition, pInfo->pRes, NULL);
S
slzhou 已提交
2423

2424
    if (!hasRemainResults(&pAggInfo->groupResInfo)) {
S
slzhou 已提交
2425 2426 2427
      doSetOperatorCompleted(pOperator);
      break;
    }
2428

S
slzhou 已提交
2429 2430 2431 2432
    if (pInfo->pRes->info.rows > 0) {
      break;
    }
  }
2433

2434
  size_t rows = blockDataGetNumOfRows(pInfo->pRes);
2435 2436
  pOperator->resultInfo.totalRows += rows;

2437
  return (rows == 0) ? NULL : pInfo->pRes;
2438 2439
}

wmmhello's avatar
wmmhello 已提交
2440
int32_t aggEncodeResultRow(SOperatorInfo* pOperator, char** result, int32_t* length) {
2441
  if (result == NULL || length == NULL) {
wmmhello's avatar
wmmhello 已提交
2442 2443 2444
    return TSDB_CODE_TSC_INVALID_INPUT;
  }
  SOptrBasicInfo* pInfo = (SOptrBasicInfo*)(pOperator->info);
2445
  SAggSupporter*  pSup = (SAggSupporter*)POINTER_SHIFT(pOperator->info, sizeof(SOptrBasicInfo));
2446
  int32_t         size = tSimpleHashGetSize(pSup->pResultRowHashTable);
2447 2448 2449
  size_t          keyLen = sizeof(uint64_t) * 2;  // estimate the key length
  int32_t         totalSize =
      sizeof(int32_t) + sizeof(int32_t) + size * (sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize);
wmmhello's avatar
wmmhello 已提交
2450

C
Cary Xu 已提交
2451 2452 2453 2454 2455 2456
  // no result
  if (getTotalBufSize(pSup->pResultBuf) == 0) {
    *result = NULL;
    *length = 0;
    return TSDB_CODE_SUCCESS;
  }
2457

wmmhello's avatar
wmmhello 已提交
2458
  *result = (char*)taosMemoryCalloc(1, totalSize);
L
Liu Jicong 已提交
2459
  if (*result == NULL) {
wmmhello's avatar
wmmhello 已提交
2460
    return TSDB_CODE_OUT_OF_MEMORY;
wmmhello's avatar
wmmhello 已提交
2461
  }
wmmhello's avatar
wmmhello 已提交
2462

wmmhello's avatar
wmmhello 已提交
2463
  int32_t offset = sizeof(int32_t);
wmmhello's avatar
wmmhello 已提交
2464 2465
  *(int32_t*)(*result + offset) = size;
  offset += sizeof(int32_t);
2466 2467

  // prepare memory
2468
  SResultRowPosition* pos = &pInfo->resultRowInfo.cur;
dengyihao's avatar
dengyihao 已提交
2469 2470
  void*               pPage = getBufPage(pSup->pResultBuf, pos->pageId);
  SResultRow*         pRow = (SResultRow*)((char*)pPage + pos->offset);
2471 2472
  setBufPageDirty(pPage, true);
  releaseBufPage(pSup->pResultBuf, pPage);
L
Liu Jicong 已提交
2473

2474 2475 2476 2477
  int32_t iter = 0;
  void*   pIter = NULL;
  while ((pIter = tSimpleHashIterate(pSup->pResultRowHashTable, pIter, &iter))) {
    void*               key = tSimpleHashGetKey(pIter, &keyLen);
2478
    SResultRowPosition* p1 = (SResultRowPosition*)pIter;
2479

dengyihao's avatar
dengyihao 已提交
2480
    pPage = (SFilePage*)getBufPage(pSup->pResultBuf, p1->pageId);
2481
    pRow = (SResultRow*)((char*)pPage + p1->offset);
2482 2483
    setBufPageDirty(pPage, true);
    releaseBufPage(pSup->pResultBuf, pPage);
wmmhello's avatar
wmmhello 已提交
2484 2485 2486

    // recalculate the result size
    int32_t realTotalSize = offset + sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize;
L
Liu Jicong 已提交
2487
    if (realTotalSize > totalSize) {
wmmhello's avatar
wmmhello 已提交
2488
      char* tmp = (char*)taosMemoryRealloc(*result, realTotalSize);
L
Liu Jicong 已提交
2489
      if (tmp == NULL) {
wafwerar's avatar
wafwerar 已提交
2490
        taosMemoryFree(*result);
wmmhello's avatar
wmmhello 已提交
2491
        *result = NULL;
wmmhello's avatar
wmmhello 已提交
2492
        return TSDB_CODE_OUT_OF_MEMORY;
L
Liu Jicong 已提交
2493
      } else {
wmmhello's avatar
wmmhello 已提交
2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505
        *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);
2506
    memcpy(*result + offset, pRow, pSup->resultRowSize);
wmmhello's avatar
wmmhello 已提交
2507 2508 2509
    offset += pSup->resultRowSize;
  }

wmmhello's avatar
wmmhello 已提交
2510 2511 2512 2513
  *(int32_t*)(*result) = offset;
  *length = offset;

  return TDB_CODE_SUCCESS;
wmmhello's avatar
wmmhello 已提交
2514 2515
}

2516
int32_t aggDecodeResultRow(SOperatorInfo* pOperator, char* result) {
2517
  if (result == NULL) {
wmmhello's avatar
wmmhello 已提交
2518
    return TSDB_CODE_TSC_INVALID_INPUT;
wmmhello's avatar
wmmhello 已提交
2519
  }
wmmhello's avatar
wmmhello 已提交
2520
  SOptrBasicInfo* pInfo = (SOptrBasicInfo*)(pOperator->info);
2521
  SAggSupporter*  pSup = (SAggSupporter*)POINTER_SHIFT(pOperator->info, sizeof(SOptrBasicInfo));
wmmhello's avatar
wmmhello 已提交
2522 2523

  //  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
2524
  int32_t length = *(int32_t*)(result);
wmmhello's avatar
wmmhello 已提交
2525
  int32_t offset = sizeof(int32_t);
2526 2527 2528 2529

  int32_t count = *(int32_t*)(result + offset);
  offset += sizeof(int32_t);

L
Liu Jicong 已提交
2530
  while (count-- > 0 && length > offset) {
wmmhello's avatar
wmmhello 已提交
2531 2532 2533
    int32_t keyLen = *(int32_t*)(result + offset);
    offset += sizeof(int32_t);

L
Liu Jicong 已提交
2534
    uint64_t    tableGroupId = *(uint64_t*)(result + offset);
2535
    SResultRow* resultRow = getNewResultRow(pSup->pResultBuf, &pSup->currentPageId, pSup->resultRowSize);
L
Liu Jicong 已提交
2536
    if (!resultRow) {
wmmhello's avatar
wmmhello 已提交
2537
      return TSDB_CODE_TSC_INVALID_INPUT;
wmmhello's avatar
wmmhello 已提交
2538
    }
2539

wmmhello's avatar
wmmhello 已提交
2540
    // add a new result set for a new group
2541
    SResultRowPosition pos = {.pageId = resultRow->pageId, .offset = resultRow->offset};
2542
    tSimpleHashPut(pSup->pResultRowHashTable, result + offset, keyLen, &pos, sizeof(SResultRowPosition));
wmmhello's avatar
wmmhello 已提交
2543 2544 2545

    offset += keyLen;
    int32_t valueLen = *(int32_t*)(result + offset);
L
Liu Jicong 已提交
2546
    if (valueLen != pSup->resultRowSize) {
wmmhello's avatar
wmmhello 已提交
2547
      return TSDB_CODE_TSC_INVALID_INPUT;
wmmhello's avatar
wmmhello 已提交
2548 2549 2550 2551 2552 2553 2554 2555 2556
    }
    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;

dengyihao's avatar
dengyihao 已提交
2557
    pInfo->resultRowInfo.cur = (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
H
Haojun Liao 已提交
2558
    // releaseBufPage(pSup->pResultBuf, getBufPage(pSup->pResultBuf, pageId));
wmmhello's avatar
wmmhello 已提交
2559 2560
  }

L
Liu Jicong 已提交
2561
  if (offset != length) {
wmmhello's avatar
wmmhello 已提交
2562
    return TSDB_CODE_TSC_INVALID_INPUT;
wmmhello's avatar
wmmhello 已提交
2563
  }
wmmhello's avatar
wmmhello 已提交
2564
  return TDB_CODE_SUCCESS;
wmmhello's avatar
wmmhello 已提交
2565 2566
}

2567 2568 2569 2570 2571
int32_t handleLimitOffset(SOperatorInfo* pOperator, SLimitInfo* pLimitInfo, SSDataBlock* pBlock, bool holdDataInBuf) {
  if (pLimitInfo->remainGroupOffset > 0) {
    if (pLimitInfo->currentGroupId == 0) {  // it is the first group
      pLimitInfo->currentGroupId = pBlock->info.groupId;
      blockDataCleanup(pBlock);
2572
      return PROJECT_RETRIEVE_CONTINUE;
2573 2574 2575
    } else if (pLimitInfo->currentGroupId != pBlock->info.groupId) {
      // now it is the data from a new group
      pLimitInfo->remainGroupOffset -= 1;
2576 2577

      // ignore data block in current group
2578 2579
      if (pLimitInfo->remainGroupOffset > 0) {
        blockDataCleanup(pBlock);
2580 2581 2582 2583 2584
        return PROJECT_RETRIEVE_CONTINUE;
      }
    }

    // set current group id of the project operator
2585
    pLimitInfo->currentGroupId = pBlock->info.groupId;
2586 2587
  }

2588
  // here check for a new group data, we need to handle the data of the previous group.
2589 2590 2591
  if (pLimitInfo->currentGroupId != 0 && pLimitInfo->currentGroupId != pBlock->info.groupId) {
    pLimitInfo->numOfOutputGroups += 1;
    if ((pLimitInfo->slimit.limit > 0) && (pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups)) {
2592
      pOperator->status = OP_EXEC_DONE;
2593
      blockDataCleanup(pBlock);
2594 2595 2596 2597 2598

      return PROJECT_RETRIEVE_DONE;
    }

    // reset the value for a new group data
2599 2600
    pLimitInfo->numOfOutputRows = 0;
    pLimitInfo->remainOffset = pLimitInfo->limit.offset;
2601 2602 2603 2604 2605

    // existing rows that belongs to previous group.
    if (pBlock->info.rows > 0) {
      return PROJECT_RETRIEVE_DONE;
    }
2606 2607 2608 2609 2610
  }

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

  // set current group id
2611
  pLimitInfo->currentGroupId = pBlock->info.groupId;
2612

2613 2614 2615
  if (pLimitInfo->remainOffset >= pBlock->info.rows) {
    pLimitInfo->remainOffset -= pBlock->info.rows;
    blockDataCleanup(pBlock);
2616
    return PROJECT_RETRIEVE_CONTINUE;
2617 2618 2619
  } else if (pLimitInfo->remainOffset < pBlock->info.rows && pLimitInfo->remainOffset > 0) {
    blockDataTrimFirstNRows(pBlock, pLimitInfo->remainOffset);
    pLimitInfo->remainOffset = 0;
2620 2621
  }

2622
  // check for the limitation in each group
2623 2624 2625 2626
  if (pLimitInfo->limit.limit >= 0 && pLimitInfo->numOfOutputRows + pBlock->info.rows >= pLimitInfo->limit.limit) {
    int32_t keepRows = (int32_t)(pLimitInfo->limit.limit - pLimitInfo->numOfOutputRows);
    blockDataKeepFirstNRows(pBlock, keepRows);
    if (pLimitInfo->slimit.limit > 0 && pLimitInfo->slimit.limit <= pLimitInfo->numOfOutputGroups) {
2627 2628 2629
      pOperator->status = OP_EXEC_DONE;
    }

2630
    return PROJECT_RETRIEVE_DONE;
2631
  }
2632

2633
  // todo optimize performance
2634 2635
  // 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.
2636 2637
  if ((!holdDataInBuf) || (pBlock->info.rows >= pOperator->resultInfo.threshold) || pLimitInfo->slimit.offset != -1 ||
      pLimitInfo->slimit.limit != -1) {
2638
    return PROJECT_RETRIEVE_DONE;
L
Liu Jicong 已提交
2639
  } else {  // not full enough, continue to accumulate the output data in the buffer.
2640 2641 2642 2643
    return PROJECT_RETRIEVE_CONTINUE;
  }
}

2644
static void doApplyScalarCalculation(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_t order, int32_t scanFlag);
L
Liu Jicong 已提交
2645 2646
static void doHandleRemainBlockForNewGroupImpl(SOperatorInfo* pOperator, SFillOperatorInfo* pInfo,
                                               SResultInfo* pResultInfo, SExecTaskInfo* pTaskInfo) {
2647
  pInfo->totalInputRows = pInfo->existNewGroupBlock->info.rows;
2648 2649 2650 2651 2652
  SSDataBlock* pResBlock = pInfo->pFinalRes;

  int32_t order = TSDB_ORDER_ASC;
  int32_t scanFlag = MAIN_SCAN;
  getTableScanInfo(pOperator, &order, &scanFlag);
H
Haojun Liao 已提交
2653

L
Liu Jicong 已提交
2654 2655
  int64_t ekey =
      Q_STATUS_EQUAL(pTaskInfo->status, TASK_COMPLETED) ? pInfo->win.ekey : pInfo->existNewGroupBlock->info.window.ekey;
2656 2657
  taosResetFillInfo(pInfo->pFillInfo, getFillInfoStart(pInfo->pFillInfo));

2658
  blockDataCleanup(pInfo->pRes);
2659 2660 2661 2662
  doApplyScalarCalculation(pOperator, pInfo->existNewGroupBlock, order, scanFlag);

  taosFillSetStartInfo(pInfo->pFillInfo, pInfo->pRes->info.rows, ekey);
  taosFillSetInputDataBlock(pInfo->pFillInfo, pInfo->pRes);
2663

2664 2665
  int32_t numOfResultRows = pResultInfo->capacity - pResBlock->info.rows;
  taosFillResultDataBlock(pInfo->pFillInfo, pResBlock, numOfResultRows);
H
Haojun Liao 已提交
2666

2667
  pInfo->curGroupId = pInfo->existNewGroupBlock->info.groupId;
2668 2669 2670
  pInfo->existNewGroupBlock = NULL;
}

L
Liu Jicong 已提交
2671 2672
static void doHandleRemainBlockFromNewGroup(SOperatorInfo* pOperator, SFillOperatorInfo* pInfo,
                                            SResultInfo* pResultInfo, SExecTaskInfo* pTaskInfo) {
2673
  if (taosFillHasMoreResults(pInfo->pFillInfo)) {
H
Haojun Liao 已提交
2674 2675
    int32_t numOfResultRows = pResultInfo->capacity - pInfo->pFinalRes->info.rows;
    taosFillResultDataBlock(pInfo->pFillInfo, pInfo->pFinalRes, numOfResultRows);
2676 2677
    pInfo->pRes->info.groupId = pInfo->curGroupId;
    return;
2678 2679 2680 2681
  }

  // handle the cached new group data block
  if (pInfo->existNewGroupBlock) {
2682 2683 2684 2685 2686 2687
    doHandleRemainBlockForNewGroupImpl(pOperator, pInfo, pResultInfo, pTaskInfo);
  }
}

static void doApplyScalarCalculation(SOperatorInfo* pOperator, SSDataBlock* pBlock, int32_t order, int32_t scanFlag) {
  SFillOperatorInfo* pInfo = pOperator->info;
L
Liu Jicong 已提交
2688 2689
  SExprSupp*         pSup = &pOperator->exprSupp;
  SSDataBlock*       pResBlock = pInfo->pFinalRes;
2690 2691 2692 2693 2694 2695 2696 2697 2698

  setInputDataBlock(pOperator, pSup->pCtx, pBlock, order, scanFlag, false);
  projectApplyFunctions(pSup->pExprInfo, pInfo->pRes, pBlock, pSup->pCtx, pSup->numOfExprs, NULL);
  pInfo->pRes->info.groupId = pBlock->info.groupId;

  SColumnInfoData* pDst = taosArrayGet(pInfo->pRes->pDataBlock, pInfo->primaryTsCol);
  SColumnInfoData* pSrc = taosArrayGet(pBlock->pDataBlock, pInfo->primarySrcSlotId);
  colDataAssign(pDst, pSrc, pInfo->pRes->info.rows, &pResBlock->info);

L
Liu Jicong 已提交
2699
  for (int32_t i = 0; i < pInfo->numOfNotFillExpr; ++i) {
2700 2701 2702 2703
    SFillColInfo* pCol = &pInfo->pFillInfo->pFillCol[i + pInfo->numOfExpr];
    ASSERT(pCol->notFillCol);

    SExprInfo* pExpr = pCol->pExpr;
L
Liu Jicong 已提交
2704 2705
    int32_t    srcSlotId = pExpr->base.pParam[0].pCol->slotId;
    int32_t    dstSlotId = pExpr->base.resSchema.slotId;
2706 2707 2708 2709

    SColumnInfoData* pDst1 = taosArrayGet(pInfo->pRes->pDataBlock, dstSlotId);
    SColumnInfoData* pSrc1 = taosArrayGet(pBlock->pDataBlock, srcSlotId);
    colDataAssign(pDst1, pSrc1, pInfo->pRes->info.rows, &pResBlock->info);
2710 2711 2712
  }
}

S
slzhou 已提交
2713
static SSDataBlock* doFillImpl(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
2714 2715
  SFillOperatorInfo* pInfo = pOperator->info;
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
2716

H
Haojun Liao 已提交
2717
  SResultInfo* pResultInfo = &pOperator->resultInfo;
H
Haojun Liao 已提交
2718
  SSDataBlock* pResBlock = pInfo->pFinalRes;
2719 2720

  blockDataCleanup(pResBlock);
2721

H
Haojun Liao 已提交
2722 2723
  int32_t order = TSDB_ORDER_ASC;
  int32_t scanFlag = MAIN_SCAN;
2724
  getTableScanInfo(pOperator, &order, &scanFlag);
2725

2726
  doHandleRemainBlockFromNewGroup(pOperator, pInfo, pResultInfo, pTaskInfo);
2727
  if (pResBlock->info.rows > 0) {
2728
    pResBlock->info.groupId = pInfo->curGroupId;
2729
    return pResBlock;
H
Haojun Liao 已提交
2730
  }
2731

H
Haojun Liao 已提交
2732
  SOperatorInfo* pDownstream = pOperator->pDownstream[0];
L
Liu Jicong 已提交
2733
  while (1) {
2734
    SSDataBlock* pBlock = pDownstream->fpSet.getNextFn(pDownstream);
2735 2736
    if (pBlock == NULL) {
      if (pInfo->totalInputRows == 0) {
2737
        doSetOperatorCompleted(pOperator);
2738 2739
        return NULL;
      }
2740

2741
      taosFillSetStartInfo(pInfo->pFillInfo, 0, pInfo->win.ekey);
2742
    } else {
2743
      blockDataUpdateTsWindow(pBlock, pInfo->primarySrcSlotId);
2744 2745

      blockDataCleanup(pInfo->pRes);
2746
      doApplyScalarCalculation(pOperator, pBlock, order, scanFlag);
2747

H
Haojun Liao 已提交
2748 2749 2750
      if (pInfo->curGroupId == 0 || pInfo->curGroupId == pInfo->pRes->info.groupId) {
        pInfo->curGroupId = pInfo->pRes->info.groupId;  // the first data block
        pInfo->totalInputRows += pInfo->pRes->info.rows;
2751

H
Haojun Liao 已提交
2752 2753 2754 2755 2756
        if (order == pInfo->pFillInfo->order) {
          taosFillSetStartInfo(pInfo->pFillInfo, pInfo->pRes->info.rows, pBlock->info.window.ekey);
        } else {
          taosFillSetStartInfo(pInfo->pFillInfo, pInfo->pRes->info.rows, pBlock->info.window.skey);
        }
H
Haojun Liao 已提交
2757
        taosFillSetInputDataBlock(pInfo->pFillInfo, pInfo->pRes);
L
Liu Jicong 已提交
2758
      } else if (pInfo->curGroupId != pBlock->info.groupId) {  // the new group data block
2759 2760 2761 2762 2763
        pInfo->existNewGroupBlock = pBlock;

        // Fill the previous group data block, before handle the data block of new group.
        // Close the fill operation for previous group data block
        taosFillSetStartInfo(pInfo->pFillInfo, 0, pInfo->win.ekey);
2764 2765 2766
      }
    }

2767 2768
    int32_t numOfResultRows = pOperator->resultInfo.capacity - pResBlock->info.rows;
    taosFillResultDataBlock(pInfo->pFillInfo, pResBlock, numOfResultRows);
2769 2770

    // current group has no more result to return
2771
    if (pResBlock->info.rows > 0) {
2772 2773
      // 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
2774
      if (pResBlock->info.rows > pResultInfo->threshold || pBlock == NULL || pInfo->existNewGroupBlock != NULL) {
2775
        pResBlock->info.groupId = pInfo->curGroupId;
2776
        return pResBlock;
2777 2778
      }

2779
      doHandleRemainBlockFromNewGroup(pOperator, pInfo, pResultInfo, pTaskInfo);
2780
      if (pResBlock->info.rows >= pOperator->resultInfo.threshold || pBlock == NULL) {
2781
        pResBlock->info.groupId = pInfo->curGroupId;
2782
        return pResBlock;
2783 2784 2785
      }
    } else if (pInfo->existNewGroupBlock) {  // try next group
      assert(pBlock != NULL);
2786 2787 2788 2789

      blockDataCleanup(pResBlock);

      doHandleRemainBlockForNewGroupImpl(pOperator, pInfo, pResultInfo, pTaskInfo);
2790
      if (pResBlock->info.rows > pResultInfo->threshold) {
2791
        pResBlock->info.groupId = pInfo->curGroupId;
2792
        return pResBlock;
2793 2794 2795 2796 2797 2798 2799
      }
    } else {
      return NULL;
    }
  }
}

S
slzhou 已提交
2800 2801 2802 2803 2804 2805 2806 2807
static SSDataBlock* doFill(SOperatorInfo* pOperator) {
  SFillOperatorInfo* pInfo = pOperator->info;
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;

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

S
slzhou 已提交
2808
  SSDataBlock* fillResult = NULL;
S
slzhou 已提交
2809
  while (true) {
S
slzhou 已提交
2810
    fillResult = doFillImpl(pOperator);
S
slzhou 已提交
2811 2812 2813 2814 2815
    if (fillResult == NULL) {
      doSetOperatorCompleted(pOperator);
      break;
    }

2816
    doFilter(pInfo->pCondition, fillResult, pInfo->pColMatchColInfo);
S
slzhou 已提交
2817 2818 2819 2820 2821
    if (fillResult->info.rows > 0) {
      break;
    }
  }

S
slzhou 已提交
2822
  if (fillResult != NULL) {
2823
    pOperator->resultInfo.totalRows += fillResult->info.rows;
S
slzhou 已提交
2824
  }
S
slzhou 已提交
2825

S
slzhou 已提交
2826
  return fillResult;
S
slzhou 已提交
2827 2828
}

2829
void destroyExprInfo(SExprInfo* pExpr, int32_t numOfExprs) {
C
Cary Xu 已提交
2830 2831 2832 2833 2834
  for (int32_t i = 0; i < numOfExprs; ++i) {
    SExprInfo* pExprInfo = &pExpr[i];
    for (int32_t j = 0; j < pExprInfo->base.numOfParams; ++j) {
      if (pExprInfo->base.pParam[j].type == FUNC_PARAM_TYPE_COLUMN) {
        taosMemoryFreeClear(pExprInfo->base.pParam[j].pCol);
H
Haojun Liao 已提交
2835
      }
2836
    }
C
Cary Xu 已提交
2837 2838 2839

    taosMemoryFree(pExprInfo->base.pParam);
    taosMemoryFree(pExprInfo->pExpr);
H
Haojun Liao 已提交
2840 2841 2842
  }
}

2843 2844 2845 2846 2847
static void destroyOperatorInfo(SOperatorInfo* pOperator) {
  if (pOperator == NULL) {
    return;
  }

2848
  if (pOperator->fpSet.closeFn != NULL) {
2849
    pOperator->fpSet.closeFn(pOperator->info);
2850 2851
  }

H
Haojun Liao 已提交
2852
  if (pOperator->pDownstream != NULL) {
L
Liu Jicong 已提交
2853
    for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
H
Haojun Liao 已提交
2854
      destroyOperatorInfo(pOperator->pDownstream[i]);
2855 2856
    }

wafwerar's avatar
wafwerar 已提交
2857
    taosMemoryFreeClear(pOperator->pDownstream);
H
Haojun Liao 已提交
2858
    pOperator->numOfDownstream = 0;
2859 2860
  }

2861
  cleanupExprSupp(&pOperator->exprSupp);
wafwerar's avatar
wafwerar 已提交
2862
  taosMemoryFreeClear(pOperator);
2863 2864
}

2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879
int32_t getBufferPgSize(int32_t rowSize, uint32_t* defaultPgsz, uint32_t* defaultBufsz) {
  *defaultPgsz = 4096;
  while (*defaultPgsz < rowSize * 4) {
    *defaultPgsz <<= 1u;
  }

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

  return 0;
}

dengyihao's avatar
dengyihao 已提交
2880 2881
int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                         const char* pKey) {
L
Liu Jicong 已提交
2882
  int32_t    code = 0;
2883 2884
  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);

2885
  pAggSup->currentPageId = -1;
dengyihao's avatar
dengyihao 已提交
2886 2887
  pAggSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
  pAggSup->keyBuf = taosMemoryCalloc(1, keyBufSize + POINTER_BYTES + sizeof(int64_t));
2888
  pAggSup->pResultRowHashTable = tSimpleHashInit(10, hashFn);
2889

H
Haojun Liao 已提交
2890
  if (pAggSup->keyBuf == NULL || pAggSup->pResultRowHashTable == NULL) {
2891 2892 2893
    return TSDB_CODE_OUT_OF_MEMORY;
  }

dengyihao's avatar
dengyihao 已提交
2894
  uint32_t defaultPgsz = 0;
2895 2896
  uint32_t defaultBufsz = 0;
  getBufferPgSize(pAggSup->resultRowSize, &defaultPgsz, &defaultBufsz);
H
Haojun Liao 已提交
2897

wafwerar's avatar
wafwerar 已提交
2898
  if (!osTempSpaceAvailable()) {
H
Haojun Liao 已提交
2899 2900 2901
    code = TSDB_CODE_NO_AVAIL_DISK;
    qError("Init stream agg supporter failed since %s, %s", terrstr(code), pKey);
    return code;
wafwerar's avatar
wafwerar 已提交
2902
  }
2903

H
Haojun Liao 已提交
2904
  code = createDiskbasedBuf(&pAggSup->pResultBuf, defaultPgsz, defaultBufsz, pKey, tsTempDir);
H
Haojun Liao 已提交
2905
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
2906
    qError("Create agg result buf failed since %s, %s", tstrerror(code), pKey);
H
Haojun Liao 已提交
2907 2908 2909
    return code;
  }

H
Haojun Liao 已提交
2910
  return code;
2911 2912
}

2913
void cleanupAggSup(SAggSupporter* pAggSup) {
wafwerar's avatar
wafwerar 已提交
2914
  taosMemoryFreeClear(pAggSup->keyBuf);
2915
  tSimpleHashCleanup(pAggSup->pResultRowHashTable);
H
Haojun Liao 已提交
2916
  destroyDiskbasedBuf(pAggSup->pResultBuf);
2917 2918
}

L
Liu Jicong 已提交
2919 2920
int32_t initAggInfo(SExprSupp* pSup, SAggSupporter* pAggSup, SExprInfo* pExprInfo, int32_t numOfCols, size_t keyBufSize,
                    const char* pkey) {
2921 2922 2923 2924 2925
  int32_t code = initExprSupp(pSup, pExprInfo, numOfCols);
  if (code != TSDB_CODE_SUCCESS) {
    return code;
  }

2926 2927 2928 2929 2930
  code = doInitAggInfoSup(pAggSup, pSup->pCtx, numOfCols, keyBufSize, pkey);
  if (code != TSDB_CODE_SUCCESS) {
    return code;
  }

L
Liu Jicong 已提交
2931
  for (int32_t i = 0; i < numOfCols; ++i) {
2932
    pSup->pCtx[i].saveHandle.pBuf = pAggSup->pResultBuf;
2933 2934
  }

2935
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
2936 2937
}

L
Liu Jicong 已提交
2938
void initResultSizeInfo(SResultInfo* pResultInfo, int32_t numOfRows) {
wmmhello's avatar
wmmhello 已提交
2939
  ASSERT(numOfRows != 0);
2940 2941
  pResultInfo->capacity = numOfRows;
  pResultInfo->threshold = numOfRows * 0.75;
2942

2943 2944
  if (pResultInfo->threshold == 0) {
    pResultInfo->threshold = numOfRows;
2945 2946 2947
  }
}

2948 2949 2950 2951 2952
void initBasicInfo(SOptrBasicInfo* pInfo, SSDataBlock* pBlock) {
  pInfo->pRes = pBlock;
  initResultRowInfo(&pInfo->resultRowInfo);
}

5
54liuyao 已提交
2953
void* destroySqlFunctionCtx(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
2954 2955 2956 2957 2958 2959 2960 2961 2962 2963
  if (pCtx == NULL) {
    return NULL;
  }

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

    taosMemoryFreeClear(pCtx[i].subsidiaries.pCtx);
2964
    taosMemoryFreeClear(pCtx[i].subsidiaries.buf);
2965 2966 2967 2968 2969 2970 2971 2972
    taosMemoryFree(pCtx[i].input.pData);
    taosMemoryFree(pCtx[i].input.pColumnDataAgg);
  }

  taosMemoryFreeClear(pCtx);
  return NULL;
}

2973
int32_t initExprSupp(SExprSupp* pSup, SExprInfo* pExprInfo, int32_t numOfExpr) {
2974 2975 2976 2977
  pSup->pExprInfo = pExprInfo;
  pSup->numOfExprs = numOfExpr;
  if (pSup->pExprInfo != NULL) {
    pSup->pCtx = createSqlFunctionCtx(pExprInfo, numOfExpr, &pSup->rowEntryInfoOffset);
2978 2979 2980
    if (pSup->pCtx == NULL) {
      return TSDB_CODE_OUT_OF_MEMORY;
    }
2981
  }
2982 2983

  return TSDB_CODE_SUCCESS;
2984 2985
}

2986 2987 2988 2989
void cleanupExprSupp(SExprSupp* pSupp) {
  destroySqlFunctionCtx(pSupp->pCtx, pSupp->numOfExprs);
  if (pSupp->pExprInfo != NULL) {
    destroyExprInfo(pSupp->pExprInfo, pSupp->numOfExprs);
C
Cary Xu 已提交
2990
    taosMemoryFreeClear(pSupp->pExprInfo);
2991 2992 2993 2994
  }
  taosMemoryFree(pSupp->rowEntryInfoOffset);
}

L
Liu Jicong 已提交
2995
SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
S
slzhou 已提交
2996
                                           SSDataBlock* pResultBlock, SNode* pCondition, SExprInfo* pScalarExprInfo,
2997
                                           int32_t numOfScalarExpr, bool mergeResult, SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
2998
  SAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SAggOperatorInfo));
L
Liu Jicong 已提交
2999
  SOperatorInfo*    pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
3000 3001 3002
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
3003

dengyihao's avatar
dengyihao 已提交
3004
  size_t keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
3005

3006
  initResultSizeInfo(&pOperator->resultInfo, 4096);
3007
  int32_t code = initAggInfo(&pOperator->exprSupp, &pInfo->aggSup, pExprInfo, numOfCols, keyBufSize, pTaskInfo->id.str);
L
Liu Jicong 已提交
3008
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3009 3010
    goto _error;
  }
H
Haojun Liao 已提交
3011

3012
  initBasicInfo(&pInfo->binfo, pResultBlock);
3013 3014 3015 3016
  code = initExprSupp(&pInfo->scalarExprSup, pScalarExprInfo, numOfScalarExpr);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
3017

3018
  pInfo->binfo.mergeResultBlock = mergeResult;
3019
  pInfo->groupId = UINT64_MAX;
S
slzhou 已提交
3020
  pInfo->pCondition = pCondition;
dengyihao's avatar
dengyihao 已提交
3021
  pOperator->name = "TableAggregate";
3022
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_HASH_AGG;
3023
  pOperator->blocking = true;
dengyihao's avatar
dengyihao 已提交
3024 3025 3026
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
3027

3028 3029
  pOperator->fpSet = createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, NULL, destroyAggOperatorInfo,
                                         aggEncodeResultRow, aggDecodeResultRow, NULL);
H
Haojun Liao 已提交
3030

3031 3032 3033 3034 3035 3036
  if (downstream->operatorType == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
    STableScanInfo* pTableScanInfo = downstream->info;
    pTableScanInfo->pdInfo.pExprSup = &pOperator->exprSupp;
    pTableScanInfo->pdInfo.pAggSup = &pInfo->aggSup;
  }

H
Haojun Liao 已提交
3037 3038 3039 3040
  code = appendDownstream(pOperator, &downstream, 1);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
3041 3042

  return pOperator;
L
Liu Jicong 已提交
3043
_error:
3044
  destroyAggOperatorInfo(pInfo);
wafwerar's avatar
wafwerar 已提交
3045
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3046 3047
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
3048 3049
}

3050
void cleanupBasicInfo(SOptrBasicInfo* pInfo) {
3051
  assert(pInfo != NULL);
H
Haojun Liao 已提交
3052
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
3053 3054
}

H
Haojun Liao 已提交
3055 3056 3057 3058 3059 3060 3061
static void freeItem(void* pItem) {
  void** p = pItem;
  if (*p != NULL) {
    taosMemoryFreeClear(*p);
  }
}

3062
void destroyAggOperatorInfo(void* param) {
L
Liu Jicong 已提交
3063
  SAggOperatorInfo* pInfo = (SAggOperatorInfo*)param;
L
Liu Jicong 已提交
3064 3065
  cleanupBasicInfo(&pInfo->binfo);

H
Haojun Liao 已提交
3066
  cleanupAggSup(&pInfo->aggSup);
S
shenglian zhou 已提交
3067
  cleanupExprSupp(&pInfo->scalarExprSup);
H
Haojun Liao 已提交
3068
  cleanupGroupResInfo(&pInfo->groupResInfo);
D
dapan1121 已提交
3069
  taosMemoryFreeClear(param);
3070
}
3071

3072
void destroyFillOperatorInfo(void* param) {
L
Liu Jicong 已提交
3073
  SFillOperatorInfo* pInfo = (SFillOperatorInfo*)param;
3074
  pInfo->pFillInfo = taosDestroyFillInfo(pInfo->pFillInfo);
H
Haojun Liao 已提交
3075
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
H
Haojun Liao 已提交
3076 3077 3078 3079 3080 3081 3082
  pInfo->pFinalRes = blockDataDestroy(pInfo->pFinalRes);

  if (pInfo->pNotFillExprInfo != NULL) {
    destroyExprInfo(pInfo->pNotFillExprInfo, pInfo->numOfNotFillExpr);
    taosMemoryFree(pInfo->pNotFillExprInfo);
  }

wafwerar's avatar
wafwerar 已提交
3083
  taosMemoryFreeClear(pInfo->p);
3084
  taosArrayDestroy(pInfo->pColMatchColInfo);
D
dapan1121 已提交
3085
  taosMemoryFreeClear(param);
3086 3087
}

3088
void destroyExchangeOperatorInfo(void* param) {
L
Liu Jicong 已提交
3089
  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
3090 3091 3092
  taosRemoveRef(exchangeObjRefPool, pExInfo->self);
}

L
Liu Jicong 已提交
3093
void freeSourceDataInfo(void* p) {
3094 3095 3096 3097
  SSourceDataInfo* pInfo = (SSourceDataInfo*)p;
  taosMemoryFreeClear(pInfo->pRsp);
}

3098
void doDestroyExchangeOperatorInfo(void* param) {
X
Xiaoyu Wang 已提交
3099
  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
3100

H
Haojun Liao 已提交
3101
  taosArrayDestroy(pExInfo->pSources);
3102
  taosArrayDestroyEx(pExInfo->pSourceDataInfo, freeSourceDataInfo);
3103 3104 3105 3106

  if (pExInfo->pResultBlockList != NULL) {
    taosArrayDestroyEx(pExInfo->pResultBlockList, freeBlock);
    pExInfo->pResultBlockList = NULL;
H
Haojun Liao 已提交
3107 3108
  }

3109
  blockDataDestroy(pExInfo->pDummyBlock);
L
Liu Jicong 已提交
3110

3111
  tsem_destroy(&pExInfo->ready);
D
dapan1121 已提交
3112
  taosMemoryFreeClear(param);
H
Haojun Liao 已提交
3113 3114
}

H
Haojun Liao 已提交
3115 3116 3117 3118
static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SExprInfo* pNotFillExpr,
                            int32_t numOfNotFillCols, SNodeListNode* pValNode, STimeWindow win, int32_t capacity,
                            const char* id, SInterval* pInterval, int32_t fillType, int32_t order) {
  SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pNotFillExpr, numOfNotFillCols, pValNode);
H
Haojun Liao 已提交
3119

H
Haojun Liao 已提交
3120 3121 3122
  int64_t     startKey = (order == TSDB_ORDER_ASC) ? win.skey : win.ekey;
  STimeWindow w = getAlignQueryTimeWindow(pInterval, pInterval->precision, startKey);
  w = getFirstQualifiedTimeWindow(startKey, &w, pInterval, order);
H
Haojun Liao 已提交
3123

L
Liu Jicong 已提交
3124 3125
  pInfo->pFillInfo = taosCreateFillInfo(w.skey, numOfCols, numOfNotFillCols, capacity, pInterval, fillType, pColInfo,
                                        pInfo->primaryTsCol, order, id);
H
Haojun Liao 已提交
3126

H
Haojun Liao 已提交
3127 3128 3129 3130 3131 3132 3133
  if (order == TSDB_ORDER_ASC) {
    pInfo->win.skey = win.skey;
    pInfo->win.ekey = win.ekey;
  } else {
    pInfo->win.skey = win.ekey;
    pInfo->win.ekey = win.skey;
  }
L
Liu Jicong 已提交
3134
  pInfo->p = taosMemoryCalloc(numOfCols, POINTER_BYTES);
3135

H
Haojun Liao 已提交
3136
  if (pInfo->pFillInfo == NULL || pInfo->p == NULL) {
H
Haojun Liao 已提交
3137 3138
    taosMemoryFree(pInfo->pFillInfo);
    taosMemoryFree(pInfo->p);
H
Haojun Liao 已提交
3139 3140 3141 3142 3143 3144
    return TSDB_CODE_OUT_OF_MEMORY;
  } else {
    return TSDB_CODE_SUCCESS;
  }
}

3145 3146 3147 3148 3149 3150
static bool isWstartColumnExist(SFillOperatorInfo* pInfo) {
  if (pInfo->numOfNotFillExpr == 0) {
    return false;
  }
  for (int32_t i = 0; i < pInfo->numOfNotFillExpr; ++i) {
    SExprInfo* exprInfo = pInfo->pNotFillExprInfo + i;
3151
    if (exprInfo->pExpr->nodeType == QUERY_NODE_COLUMN && exprInfo->base.numOfParams == 1 &&
3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166
        exprInfo->base.pParam[0].pCol->colType == COLUMN_TYPE_WINDOW_START) {
      return true;
    }
  }
  return false;
}

static int32_t createWStartTsAsNotFillExpr(SFillOperatorInfo* pInfo, SFillPhysiNode* pPhyFillNode) {
  bool wstartExist = isWstartColumnExist(pInfo);
  if (wstartExist == false) {
    if (pPhyFillNode->pWStartTs->type != QUERY_NODE_TARGET) {
      qError("pWStartTs of fill physical node is not a target node");
      return TSDB_CODE_QRY_SYS_ERROR;
    }

3167 3168
    SExprInfo* notFillExprs =
        taosMemoryRealloc(pInfo->pNotFillExprInfo, (pInfo->numOfNotFillExpr + 1) * sizeof(SExprInfo));
3169 3170 3171 3172 3173 3174 3175 3176 3177 3178
    if (notFillExprs == NULL) {
      return TSDB_CODE_OUT_OF_MEMORY;
    }

    createExprFromTargetNode(notFillExprs + pInfo->numOfNotFillExpr, (STargetNode*)pPhyFillNode->pWStartTs);

    ++pInfo->numOfNotFillExpr;
    pInfo->pNotFillExprInfo = notFillExprs;
    return TSDB_CODE_SUCCESS;
  }
3179

3180 3181 3182
  return TSDB_CODE_SUCCESS;
}

L
Liu Jicong 已提交
3183 3184
SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SFillPhysiNode* pPhyFillNode,
                                      SExecTaskInfo* pTaskInfo) {
3185 3186 3187 3188 3189 3190
  SFillOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SFillOperatorInfo));
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }

L
Liu Jicong 已提交
3191
  SSDataBlock* pResBlock = createResDataBlock(pPhyFillNode->node.pOutputDataBlockDesc);
3192
  SExprInfo*   pExprInfo = createExprInfo(pPhyFillNode->pFillExprs, NULL, &pInfo->numOfExpr);
H
Haojun Liao 已提交
3193
  pInfo->pNotFillExprInfo = createExprInfo(pPhyFillNode->pNotFillExprs, NULL, &pInfo->numOfNotFillExpr);
3194 3195 3196 3197
  int32_t code = createWStartTsAsNotFillExpr(pInfo, pPhyFillNode);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
L
Liu Jicong 已提交
3198
  SInterval* pInterval =
3199
      QUERY_NODE_PHYSICAL_PLAN_MERGE_ALIGNED_INTERVAL == downstream->operatorType
L
Liu Jicong 已提交
3200 3201
          ? &((SMergeAlignedIntervalAggOperatorInfo*)downstream->info)->intervalAggOperatorInfo->interval
          : &((SIntervalAggOperatorInfo*)downstream->info)->interval;
3202

3203
  int32_t order = (pPhyFillNode->inputTsOrder == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
3204
  int32_t type = convertFillType(pPhyFillNode->mode);
3205

H
Haojun Liao 已提交
3206
  SResultInfo* pResultInfo = &pOperator->resultInfo;
3207
  initResultSizeInfo(&pOperator->resultInfo, 4096);
H
Haojun Liao 已提交
3208
  blockDataEnsureCapacity(pResBlock, pOperator->resultInfo.capacity);
3209
  initExprSupp(&pOperator->exprSupp, pExprInfo, pInfo->numOfExpr);
H
Haojun Liao 已提交
3210

H
Haojun Liao 已提交
3211 3212
  pInfo->primaryTsCol = ((STargetNode*)pPhyFillNode->pWStartTs)->slotId;
  pInfo->primarySrcSlotId = ((SColumnNode*)((STargetNode*)pPhyFillNode->pWStartTs)->pExpr)->slotId;
3213

3214
  int32_t numOfOutputCols = 0;
3215
  SArray* pColMatchColInfo = extractColMatchInfo(pPhyFillNode->pFillExprs, pPhyFillNode->node.pOutputDataBlockDesc,
3216
                                                 &numOfOutputCols, COL_MATCH_FROM_SLOT_ID);
3217

3218
  code = initFillInfo(pInfo, pExprInfo, pInfo->numOfExpr, pInfo->pNotFillExprInfo, pInfo->numOfNotFillExpr,
3219 3220
                      (SNodeListNode*)pPhyFillNode->pValues, pPhyFillNode->timeRange, pResultInfo->capacity,
                      pTaskInfo->id.str, pInterval, type, order);
3221 3222 3223
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
3224

3225
  pInfo->pRes = pResBlock;
H
Haojun Liao 已提交
3226 3227 3228
  pInfo->pFinalRes = createOneDataBlock(pResBlock, false);
  blockDataEnsureCapacity(pInfo->pFinalRes, pOperator->resultInfo.capacity);

3229 3230 3231 3232 3233 3234 3235
  pInfo->pCondition = pPhyFillNode->node.pConditions;
  pInfo->pColMatchColInfo = pColMatchColInfo;
  pOperator->name = "FillOperator";
  pOperator->blocking = false;
  pOperator->status = OP_NOT_OPENED;
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_FILL;
  pOperator->exprSupp.pExprInfo = pExprInfo;
3236
  pOperator->exprSupp.numOfExprs = pInfo->numOfExpr;
3237 3238
  pOperator->info = pInfo;
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
3239

L
Liu Jicong 已提交
3240
  pOperator->fpSet =
3241
      createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, NULL, destroyFillOperatorInfo, NULL, NULL, NULL);
3242

3243
  code = appendDownstream(pOperator, &downstream, 1);
3244
  return pOperator;
H
Haojun Liao 已提交
3245

L
Liu Jicong 已提交
3246
_error:
wafwerar's avatar
wafwerar 已提交
3247 3248
  taosMemoryFreeClear(pOperator);
  taosMemoryFreeClear(pInfo);
H
Haojun Liao 已提交
3249
  return NULL;
3250 3251
}

D
dapan1121 已提交
3252
static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model, char* dbFName) {
wafwerar's avatar
wafwerar 已提交
3253
  SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
3254
  setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
3255

3256
  pTaskInfo->schemaInfo.dbname = strdup(dbFName);
3257
  pTaskInfo->cost.created = taosGetTimestampMs();
H
Haojun Liao 已提交
3258
  pTaskInfo->id.queryId = queryId;
dengyihao's avatar
dengyihao 已提交
3259
  pTaskInfo->execModel = model;
H
Haojun Liao 已提交
3260

wafwerar's avatar
wafwerar 已提交
3261
  char* p = taosMemoryCalloc(1, 128);
L
Liu Jicong 已提交
3262
  snprintf(p, 128, "TID:0x%" PRIx64 " QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
3263
  pTaskInfo->id.str = p;
H
Haojun Liao 已提交
3264

3265 3266
  return pTaskInfo;
}
H
Haojun Liao 已提交
3267

H
Haojun Liao 已提交
3268
static SArray* extractColumnInfo(SNodeList* pNodeList);
3269

H
Haojun Liao 已提交
3270 3271
SSchemaWrapper* extractQueriedColumnSchema(SScanPhysiNode* pScanNode);

3272
int32_t extractTableSchemaInfo(SReadHandle* pHandle, SScanPhysiNode* pScanNode, SExecTaskInfo* pTaskInfo) {
3273 3274
  SMetaReader mr = {0};
  metaReaderInit(&mr, pHandle->meta, 0);
3275
  int32_t code = metaGetTableEntryByUid(&mr, pScanNode->uid);
3276
  if (code != TSDB_CODE_SUCCESS) {
L
Liu Jicong 已提交
3277 3278
    qError("failed to get the table meta, uid:0x%" PRIx64 ", suid:0x%" PRIx64 ", %s", pScanNode->uid, pScanNode->suid,
           GET_TASKID(pTaskInfo));
H
Haojun Liao 已提交
3279

D
dapan1121 已提交
3280
    metaReaderClear(&mr);
3281
    return terrno;
D
dapan1121 已提交
3282
  }
3283

3284 3285
  SSchemaInfo* pSchemaInfo = &pTaskInfo->schemaInfo;
  pSchemaInfo->tablename = strdup(mr.me.name);
3286 3287

  if (mr.me.type == TSDB_SUPER_TABLE) {
3288 3289
    pSchemaInfo->sw = tCloneSSchemaWrapper(&mr.me.stbEntry.schemaRow);
    pSchemaInfo->tversion = mr.me.stbEntry.schemaTag.version;
3290
  } else if (mr.me.type == TSDB_CHILD_TABLE) {
3291 3292
    tDecoderClear(&mr.coder);

3293 3294
    tb_uid_t suid = mr.me.ctbEntry.suid;
    metaGetTableEntryByUid(&mr, suid);
3295 3296
    pSchemaInfo->sw = tCloneSSchemaWrapper(&mr.me.stbEntry.schemaRow);
    pSchemaInfo->tversion = mr.me.stbEntry.schemaTag.version;
3297
  } else {
3298
    pSchemaInfo->sw = tCloneSSchemaWrapper(&mr.me.ntbEntry.schemaRow);
3299
  }
3300 3301

  metaReaderClear(&mr);
3302

H
Haojun Liao 已提交
3303 3304 3305 3306 3307
  pSchemaInfo->qsw = extractQueriedColumnSchema(pScanNode);
  return TSDB_CODE_SUCCESS;
}

SSchemaWrapper* extractQueriedColumnSchema(SScanPhysiNode* pScanNode) {
3308 3309 3310
  int32_t numOfCols = LIST_LENGTH(pScanNode->pScanCols);
  int32_t numOfTags = LIST_LENGTH(pScanNode->pScanPseudoCols);

3311
  SSchemaWrapper* pqSw = taosMemoryCalloc(1, sizeof(SSchemaWrapper));
3312
  pqSw->pSchema = taosMemoryCalloc(numOfCols + numOfTags, sizeof(SSchema));
3313

L
Liu Jicong 已提交
3314
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
3315
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pScanNode->pScanCols, i);
3316 3317
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

H
Haojun Liao 已提交
3318 3319 3320 3321 3322
    SSchema* pSchema = &pqSw->pSchema[pqSw->nCols++];
    pSchema->colId = pColNode->colId;
    pSchema->type = pColNode->node.resType.type;
    pSchema->type = pColNode->node.resType.bytes;
    strncpy(pSchema->name, pColNode->colName, tListLen(pSchema->name));
3323 3324
  }

3325
  // this the tags and pseudo function columns, we only keep the tag columns
3326
  for (int32_t i = 0; i < numOfTags; ++i) {
3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pScanNode->pScanPseudoCols, i);

    int32_t type = nodeType(pNode->pExpr);
    if (type == QUERY_NODE_COLUMN) {
      SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

      SSchema* pSchema = &pqSw->pSchema[pqSw->nCols++];
      pSchema->colId = pColNode->colId;
      pSchema->type = pColNode->node.resType.type;
      pSchema->type = pColNode->node.resType.bytes;
      strncpy(pSchema->name, pColNode->colName, tListLen(pSchema->name));
    }
  }

H
Haojun Liao 已提交
3341
  return pqSw;
3342 3343
}

3344 3345
static void cleanupTableSchemaInfo(SSchemaInfo* pSchemaInfo) {
  taosMemoryFreeClear(pSchemaInfo->dbname);
3346
  taosMemoryFreeClear(pSchemaInfo->tablename);
3347 3348
  tDeleteSSchemaWrapper(pSchemaInfo->sw);
  tDeleteSSchemaWrapper(pSchemaInfo->qsw);
3349 3350
}

3351 3352 3353 3354
static void cleanupStreamInfo(SStreamTaskInfo* pStreamInfo) {
  tDeleteSSchemaWrapper(pStreamInfo->schema);
}

H
Haojun Liao 已提交
3355
static int32_t sortTableGroup(STableListInfo* pTableListInfo) {
wmmhello's avatar
wmmhello 已提交
3356
  taosArrayClear(pTableListInfo->pGroupList);
H
Haojun Liao 已提交
3357
  SArray* sortSupport = taosArrayInit(16, sizeof(uint64_t));
3358
  if (sortSupport == NULL) return TSDB_CODE_OUT_OF_MEMORY;
wmmhello's avatar
wmmhello 已提交
3359 3360
  for (int32_t i = 0; i < taosArrayGetSize(pTableListInfo->pTableList); i++) {
    STableKeyInfo* info = taosArrayGet(pTableListInfo->pTableList, i);
3361
    uint64_t*      groupId = taosHashGet(pTableListInfo->map, &info->uid, sizeof(uint64_t));
wmmhello's avatar
wmmhello 已提交
3362 3363

    int32_t index = taosArraySearchIdx(sortSupport, groupId, compareUint64Val, TD_EQ);
3364 3365 3366 3367
    if (index == -1) {
      void*   p = taosArraySearch(sortSupport, groupId, compareUint64Val, TD_GT);
      SArray* tGroup = taosArrayInit(8, sizeof(STableKeyInfo));
      if (tGroup == NULL) {
wmmhello's avatar
wmmhello 已提交
3368 3369 3370
        taosArrayDestroy(sortSupport);
        return TSDB_CODE_OUT_OF_MEMORY;
      }
3371
      if (taosArrayPush(tGroup, info) == NULL) {
wmmhello's avatar
wmmhello 已提交
3372 3373 3374 3375
        qError("taos push info array error");
        taosArrayDestroy(sortSupport);
        return TSDB_CODE_QRY_APP_ERROR;
      }
3376
      if (p == NULL) {
wmmhello's avatar
wmmhello 已提交
3377
        if (taosArrayPush(sortSupport, groupId) == NULL) {
wmmhello's avatar
wmmhello 已提交
3378 3379 3380 3381
          qError("taos push support array error");
          taosArrayDestroy(sortSupport);
          return TSDB_CODE_QRY_APP_ERROR;
        }
wmmhello's avatar
wmmhello 已提交
3382
        if (taosArrayPush(pTableListInfo->pGroupList, &tGroup) == NULL) {
wmmhello's avatar
wmmhello 已提交
3383 3384 3385 3386
          qError("taos push group array error");
          taosArrayDestroy(sortSupport);
          return TSDB_CODE_QRY_APP_ERROR;
        }
3387
      } else {
wmmhello's avatar
wmmhello 已提交
3388
        int32_t pos = TARRAY_ELEM_IDX(sortSupport, p);
3389
        if (taosArrayInsert(sortSupport, pos, groupId) == NULL) {
wmmhello's avatar
wmmhello 已提交
3390 3391 3392 3393
          qError("taos insert support array error");
          taosArrayDestroy(sortSupport);
          return TSDB_CODE_QRY_APP_ERROR;
        }
3394
        if (taosArrayInsert(pTableListInfo->pGroupList, pos, &tGroup) == NULL) {
wmmhello's avatar
wmmhello 已提交
3395 3396 3397 3398 3399
          qError("taos insert group array error");
          taosArrayDestroy(sortSupport);
          return TSDB_CODE_QRY_APP_ERROR;
        }
      }
3400
    } else {
wmmhello's avatar
wmmhello 已提交
3401
      SArray* tGroup = (SArray*)taosArrayGetP(pTableListInfo->pGroupList, index);
3402
      if (taosArrayPush(tGroup, info) == NULL) {
wmmhello's avatar
wmmhello 已提交
3403 3404 3405 3406 3407 3408 3409 3410 3411 3412
        qError("taos push uid array error");
        taosArrayDestroy(sortSupport);
        return TSDB_CODE_QRY_APP_ERROR;
      }
    }
  }
  taosArrayDestroy(sortSupport);
  return TDB_CODE_SUCCESS;
}

3413
bool groupbyTbname(SNodeList* pGroupList) {
3414
  bool bytbname = false;
H
Haojun Liao 已提交
3415
  if (LIST_LENGTH(pGroupList) > 0) {
3416 3417 3418 3419 3420 3421 3422 3423 3424 3425
    SNode* p = nodesListGetNode(pGroupList, 0);
    if (p->type == QUERY_NODE_FUNCTION) {
      // partition by tbname/group by tbname
      bytbname = (strcmp(((struct SFunctionNode*)p)->functionName, "tbname") == 0);
    }
  }

  return bytbname;
}

wmmhello's avatar
wmmhello 已提交
3426 3427
int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle, SNodeList* group) {
  if (group == NULL) {
wmmhello's avatar
wmmhello 已提交
3428 3429 3430 3431 3432 3433 3434 3435
    return TDB_CODE_SUCCESS;
  }

  pTableListInfo->map = taosHashInit(32, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
  if (pTableListInfo->map == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

3436 3437
  bool assignUid = groupbyTbname(group);

H
Haojun Liao 已提交
3438
  size_t numOfTables = taosArrayGetSize(pTableListInfo->pTableList);
3439

H
Haojun Liao 已提交
3440 3441 3442
  if (assignUid) {
    for (int32_t i = 0; i < numOfTables; i++) {
      STableKeyInfo* info = taosArrayGet(pTableListInfo->pTableList, i);
3443
      info->groupId = info->uid;
H
Haojun Liao 已提交
3444 3445 3446 3447 3448 3449
      taosHashPut(pTableListInfo->map, &(info->uid), sizeof(uint64_t), &info->groupId, sizeof(uint64_t));
    }
  } else {
    int32_t code = getColInfoResultForGroupby(pHandle->meta, group, pTableListInfo);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
wmmhello's avatar
wmmhello 已提交
3450 3451
    }
  }
3452

3453
  if (pTableListInfo->needSortTableByGroupId) {
H
Haojun Liao 已提交
3454
    return sortTableGroup(pTableListInfo);
3455 3456
  }

wmmhello's avatar
wmmhello 已提交
3457 3458 3459
  return TDB_CODE_SUCCESS;
}

3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476
static int32_t initTableblockDistQueryCond(uint64_t uid, SQueryTableDataCond* pCond) {
  memset(pCond, 0, sizeof(SQueryTableDataCond));

  pCond->order = TSDB_ORDER_ASC;
  pCond->numOfCols = 1;
  pCond->colList = taosMemoryCalloc(1, sizeof(SColumnInfo));
  if (pCond->colList == NULL) {
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return terrno;
  }

  pCond->colList->colId = 1;
  pCond->colList->type = TSDB_DATA_TYPE_TIMESTAMP;
  pCond->colList->bytes = sizeof(TSKEY);

  pCond->twindows = (STimeWindow){.skey = INT64_MIN, .ekey = INT64_MAX};
  pCond->suid = uid;
3477
  pCond->type = TIMEWINDOW_RANGE_CONTAINED;
3478
  pCond->startVersion = -1;
L
Liu Jicong 已提交
3479
  pCond->endVersion = -1;
3480 3481 3482 3483

  return TSDB_CODE_SUCCESS;
}

H
Haojun Liao 已提交
3484
SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
L
Liu Jicong 已提交
3485 3486
                                  STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond,
                                  const char* pUser) {
3487 3488
  int32_t type = nodeType(pPhyNode);

X
Xiaoyu Wang 已提交
3489
  if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
3490
    SOperatorInfo* pOperator = NULL;
H
Haojun Liao 已提交
3491
    if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == type) {
dengyihao's avatar
dengyihao 已提交
3492
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
3493

L
Liu Jicong 已提交
3494 3495 3496
      int32_t code =
          createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort, pHandle,
                                  pTableListInfo, pTagCond, pTagIndexCond, GET_TASKID(pTaskInfo));
3497
      if (code) {
wmmhello's avatar
wmmhello 已提交
3498
        pTaskInfo->code = code;
3499
        qError("failed to createScanTableListInfo, code:%s, %s", tstrerror(code), GET_TASKID(pTaskInfo));
D
dapan1121 已提交
3500 3501
        return NULL;
      }
wmmhello's avatar
wmmhello 已提交
3502

3503
      code = extractTableSchemaInfo(pHandle, &pTableScanNode->scan, pTaskInfo);
S
slzhou 已提交
3504
      if (code) {
3505
        pTaskInfo->code = terrno;
wmmhello's avatar
wmmhello 已提交
3506 3507 3508
        return NULL;
      }

3509
      pOperator = createTableScanOperatorInfo(pTableScanNode, pHandle, pTaskInfo);
3510 3511
      STableScanInfo* pScanInfo = pOperator->info;
      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
S
slzhou 已提交
3512 3513
    } else if (QUERY_NODE_PHYSICAL_PLAN_TABLE_MERGE_SCAN == type) {
      STableMergeScanPhysiNode* pTableScanNode = (STableMergeScanPhysiNode*)pPhyNode;
L
Liu Jicong 已提交
3514 3515 3516
      int32_t                   code =
          createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort, pHandle,
                                  pTableListInfo, pTagCond, pTagIndexCond, GET_TASKID(pTaskInfo));
L
Liu Jicong 已提交
3517
      if (code) {
wmmhello's avatar
wmmhello 已提交
3518
        pTaskInfo->code = code;
H
Haojun Liao 已提交
3519
        qError("failed to createScanTableListInfo, code: %s", tstrerror(code));
wmmhello's avatar
wmmhello 已提交
3520 3521
        return NULL;
      }
3522

3523
      code = extractTableSchemaInfo(pHandle, &pTableScanNode->scan, pTaskInfo);
wmmhello's avatar
wmmhello 已提交
3524 3525 3526 3527
      if (code) {
        pTaskInfo->code = terrno;
        return NULL;
      }
wmmhello's avatar
wmmhello 已提交
3528

3529
      pOperator = createTableMergeScanOperatorInfo(pTableScanNode, pTableListInfo, pHandle, pTaskInfo);
wmmhello's avatar
wmmhello 已提交
3530

3531 3532
      STableScanInfo* pScanInfo = pOperator->info;
      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
H
Haojun Liao 已提交
3533
    } else if (QUERY_NODE_PHYSICAL_PLAN_EXCHANGE == type) {
3534
      pOperator = createExchangeOperatorInfo(pHandle->pMsgCb->clientRpc, (SExchangePhysiNode*)pPhyNode, pTaskInfo);
H
Haojun Liao 已提交
3535
    } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN == type) {
5
54liuyao 已提交
3536
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
5
54liuyao 已提交
3537
      if (pHandle->vnode) {
L
Liu Jicong 已提交
3538 3539 3540
        int32_t code =
            createScanTableListInfo(&pTableScanNode->scan, pTableScanNode->pGroupTags, pTableScanNode->groupSort,
                                    pHandle, pTableListInfo, pTagCond, pTagIndexCond, GET_TASKID(pTaskInfo));
L
Liu Jicong 已提交
3541
        if (code) {
wmmhello's avatar
wmmhello 已提交
3542
          pTaskInfo->code = code;
H
Haojun Liao 已提交
3543
          qError("failed to createScanTableListInfo, code: %s", tstrerror(code));
wmmhello's avatar
wmmhello 已提交
3544 3545
          return NULL;
        }
L
Liu Jicong 已提交
3546 3547 3548 3549 3550

#ifndef NDEBUG
        int32_t sz = taosArrayGetSize(pTableListInfo->pTableList);
        for (int32_t i = 0; i < sz; i++) {
          STableKeyInfo* pKeyInfo = taosArrayGet(pTableListInfo->pTableList, i);
S
Shengliang Guan 已提交
3551
          qDebug("creating stream task: add table %" PRId64, pKeyInfo->uid);
L
Liu Jicong 已提交
3552 3553
        }
#endif
3554
      }
3555

H
Haojun Liao 已提交
3556
      pTaskInfo->schemaInfo.qsw = extractQueriedColumnSchema(&pTableScanNode->scan);
3557
      pOperator = createStreamScanOperatorInfo(pHandle, pTableScanNode, pTagCond, pTaskInfo);
H
Haojun Liao 已提交
3558
    } else if (QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN == type) {
L
Liu Jicong 已提交
3559
      SSystemTableScanPhysiNode* pSysScanPhyNode = (SSystemTableScanPhysiNode*)pPhyNode;
3560
      pOperator = createSysTableScanOperatorInfo(pHandle, pSysScanPhyNode, pUser, pTaskInfo);
3561
    } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
X
Xiaoyu Wang 已提交
3562
      STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
3563
      int32_t code = getTableList(pHandle->meta, pHandle->vnode, pScanPhyNode, pTagCond, pTagIndexCond, pTableListInfo);
3564
      if (code != TSDB_CODE_SUCCESS) {
3565
        pTaskInfo->code = code;
H
Haojun Liao 已提交
3566
        qError("failed to getTableList, code: %s", tstrerror(code));
3567 3568 3569
        return NULL;
      }

3570
      pOperator = createTagScanOperatorInfo(pHandle, pScanPhyNode, pTableListInfo, pTaskInfo);
3571
    } else if (QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN == type) {
3572
      SBlockDistScanPhysiNode* pBlockNode = (SBlockDistScanPhysiNode*)pPhyNode;
3573 3574 3575
      pTableListInfo->pTableList = taosArrayInit(4, sizeof(STableKeyInfo));

      if (pBlockNode->tableType == TSDB_SUPER_TABLE) {
3576
        int32_t code = vnodeGetAllTableList(pHandle->vnode, pBlockNode->uid, pTableListInfo->pTableList);
3577 3578 3579 3580 3581
        if (code != TSDB_CODE_SUCCESS) {
          pTaskInfo->code = terrno;
          return NULL;
        }
      } else {  // Create one table group.
3582
        STableKeyInfo info = {.uid = pBlockNode->uid, .groupId = 0};
3583 3584 3585 3586
        taosArrayPush(pTableListInfo->pTableList, &info);
      }

      SQueryTableDataCond cond = {0};
L
Liu Jicong 已提交
3587
      int32_t             code = initTableblockDistQueryCond(pBlockNode->suid, &cond);
3588 3589
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
3590
      }
H
Haojun Liao 已提交
3591 3592 3593

      STsdbReader* pReader = NULL;
      tsdbReaderOpen(pHandle->vnode, &cond, pTableListInfo->pTableList, &pReader, "");
3594 3595
      cleanupQueryTableDataCond(&cond);

3596
      pOperator = createDataBlockInfoScanOperator(pReader, pHandle, cond.suid, pBlockNode, pTaskInfo);
H
Haojun Liao 已提交
3597 3598 3599
    } else if (QUERY_NODE_PHYSICAL_PLAN_LAST_ROW_SCAN == type) {
      SLastRowScanPhysiNode* pScanNode = (SLastRowScanPhysiNode*)pPhyNode;

L
Liu Jicong 已提交
3600 3601
      int32_t code = createScanTableListInfo(&pScanNode->scan, pScanNode->pGroupTags, true, pHandle, pTableListInfo,
                                             pTagCond, pTagIndexCond, GET_TASKID(pTaskInfo));
3602 3603 3604 3605
      if (code != TSDB_CODE_SUCCESS) {
        pTaskInfo->code = code;
        return NULL;
      }
3606

3607
      code = extractTableSchemaInfo(pHandle, &pScanNode->scan, pTaskInfo);
3608 3609 3610
      if (code != TSDB_CODE_SUCCESS) {
        pTaskInfo->code = code;
        return NULL;
H
Haojun Liao 已提交
3611 3612
      }

3613
      pOperator = createCacherowsScanOperator(pScanNode, pHandle, pTaskInfo);
3614
    } else if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
3615
      pOperator = createProjectOperatorInfo(NULL, (SProjectPhysiNode*)pPhyNode, pTaskInfo);
H
Haojun Liao 已提交
3616 3617
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
3618
    }
3619 3620 3621 3622 3623

    if (pOperator != NULL) {
      pOperator->resultDataBlockId = pPhyNode->pOutputDataBlockDesc->dataBlockId;
    }

3624
    return pOperator;
H
Haojun Liao 已提交
3625 3626
  }

3627 3628
  int32_t num = 0;
  size_t  size = LIST_LENGTH(pPhyNode->pChildren);
H
Haojun Liao 已提交
3629

3630
  SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
dengyihao's avatar
dengyihao 已提交
3631
  for (int32_t i = 0; i < size; ++i) {
3632
    SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
3633
    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, pTableListInfo, pTagCond, pTagIndexCond, pUser);
3634
    if (ops[i] == NULL) {
H
Haojun Liao 已提交
3635
      taosMemoryFree(ops);
3636 3637
      return NULL;
    }
3638 3639

    ops[i]->resultDataBlockId = pChildNode->pOutputDataBlockDesc->dataBlockId;
3640
  }
H
Haojun Liao 已提交
3641

3642
  SOperatorInfo* pOptr = NULL;
H
Haojun Liao 已提交
3643
  if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
3644
    pOptr = createProjectOperatorInfo(ops[0], (SProjectPhysiNode*)pPhyNode, pTaskInfo);
3645
  } else if (QUERY_NODE_PHYSICAL_PLAN_HASH_AGG == type) {
H
Haojun Liao 已提交
3646 3647
    SAggPhysiNode* pAggNode = (SAggPhysiNode*)pPhyNode;
    SExprInfo*     pExprInfo = createExprInfo(pAggNode->pAggFuncs, pAggNode->pGroupKeys, &num);
3648
    SSDataBlock*   pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
3649

dengyihao's avatar
dengyihao 已提交
3650
    int32_t    numOfScalarExpr = 0;
3651 3652 3653 3654 3655
    SExprInfo* pScalarExprInfo = NULL;
    if (pAggNode->pExprs != NULL) {
      pScalarExprInfo = createExprInfo(pAggNode->pExprs, NULL, &numOfScalarExpr);
    }

H
Haojun Liao 已提交
3656 3657
    if (pAggNode->pGroupKeys != NULL) {
      SArray* pColList = extractColumnInfo(pAggNode->pGroupKeys);
dengyihao's avatar
dengyihao 已提交
3658
      pOptr = createGroupOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pAggNode->node.pConditions,
wmmhello's avatar
wmmhello 已提交
3659
                                      pScalarExprInfo, numOfScalarExpr, pTaskInfo);
H
Haojun Liao 已提交
3660
    } else {
L
Liu Jicong 已提交
3661
      pOptr = createAggregateOperatorInfo(ops[0], pExprInfo, num, pResBlock, pAggNode->node.pConditions,
3662
                                          pScalarExprInfo, numOfScalarExpr, pAggNode->mergeDataBlock, pTaskInfo);
H
Haojun Liao 已提交
3663
    }
3664
  } else if (QUERY_NODE_PHYSICAL_PLAN_HASH_INTERVAL == type) {
H
Haojun Liao 已提交
3665
    SIntervalPhysiNode* pIntervalPhyNode = (SIntervalPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
3666

H
Haojun Liao 已提交
3667
    SExprInfo*   pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
3668
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
3669

dengyihao's avatar
dengyihao 已提交
3670 3671 3672 3673 3674 3675
    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
3676

X
Xiaoyu Wang 已提交
3677 3678 3679 3680 3681
    STimeWindowAggSupp as = {
        .waterMark = pIntervalPhyNode->window.watermark,
        .calTrigger = pIntervalPhyNode->window.triggerType,
        .maxTs = INT64_MIN,
    };
3682
    ASSERT(as.calTrigger != STREAM_TRIGGER_MAX_DELAY);
3683

3684
    int32_t tsSlotId = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
L
Liu Jicong 已提交
3685
    bool    isStream = (QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL == type);
3686 3687
    pOptr = createIntervalOperatorInfo(ops[0], pExprInfo, num, pResBlock, &interval, tsSlotId, &as, pIntervalPhyNode,
                                       pTaskInfo, isStream);
3688

3689
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL == type) {
3690
    pOptr = createStreamIntervalOperatorInfo(ops[0], pPhyNode, pTaskInfo);
3691 3692
  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_ALIGNED_INTERVAL == type) {
    SMergeAlignedIntervalPhysiNode* pIntervalPhyNode = (SMergeAlignedIntervalPhysiNode*)pPhyNode;
3693
    pOptr = createMergeAlignedIntervalOperatorInfo(ops[0], pIntervalPhyNode, pTaskInfo);
S
shenglian zhou 已提交
3694
  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_INTERVAL == type) {
X
Xiaoyu Wang 已提交
3695
    SMergeIntervalPhysiNode* pIntervalPhyNode = (SMergeIntervalPhysiNode*)pPhyNode;
3696
    pOptr = createMergeIntervalOperatorInfo(ops[0], pIntervalPhyNode, pTaskInfo);
5
54liuyao 已提交
3697
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_INTERVAL == type) {
3698
    int32_t children = 0;
5
54liuyao 已提交
3699 3700
    pOptr = createStreamFinalIntervalOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_INTERVAL == type) {
5
54liuyao 已提交
3701
    int32_t children = pHandle->numOfVgroups;
5
54liuyao 已提交
3702
    pOptr = createStreamFinalIntervalOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
H
Haojun Liao 已提交
3703
  } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {
3704
    pOptr = createSortOperatorInfo(ops[0], (SSortPhysiNode*)pPhyNode, pTaskInfo);
S
shenglian zhou 已提交
3705 3706
  } else if (QUERY_NODE_PHYSICAL_PLAN_GROUP_SORT == type) {
    pOptr = createGroupSortOperatorInfo(ops[0], (SGroupSortPhysiNode*)pPhyNode, pTaskInfo);
X
Xiaoyu Wang 已提交
3707
  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE == type) {
3708
    SMergePhysiNode* pMergePhyNode = (SMergePhysiNode*)pPhyNode;
3709
    pOptr = createMultiwayMergeOperatorInfo(ops, size, pMergePhyNode, pTaskInfo);
3710
  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_SESSION == type) {
H
Haojun Liao 已提交
3711
    SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
3712
    pOptr = createSessionAggOperatorInfo(ops[0], pSessionNode, pTaskInfo);
3713
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SESSION == type) {
3714 3715 3716 3717 3718
    pOptr = createStreamSessionAggOperatorInfo(ops[0], pPhyNode, pTaskInfo);
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SEMI_SESSION == type) {
    int32_t children = 0;
    pOptr = createStreamFinalSessionAggOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_FINAL_SESSION == type) {
3719
    int32_t children = pHandle->numOfVgroups;
3720
    pOptr = createStreamFinalSessionAggOperatorInfo(ops[0], pPhyNode, pTaskInfo, children);
H
Haojun Liao 已提交
3721
  } else if (QUERY_NODE_PHYSICAL_PLAN_PARTITION == type) {
3722
    pOptr = createPartitionOperatorInfo(ops[0], (SPartitionPhysiNode*)pPhyNode, pTaskInfo);
3723 3724
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_PARTITION == type) {
    pOptr = createStreamPartitionOperatorInfo(ops[0], (SPartitionPhysiNode*)pPhyNode, pTaskInfo);
3725
  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_STATE == type) {
dengyihao's avatar
dengyihao 已提交
3726
    SStateWinodwPhysiNode* pStateNode = (SStateWinodwPhysiNode*)pPhyNode;
3727
    pOptr = createStatewindowOperatorInfo(ops[0], pStateNode, pTaskInfo);
3728
  } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_STATE == type) {
5
54liuyao 已提交
3729
    pOptr = createStreamStateAggOperatorInfo(ops[0], pPhyNode, pTaskInfo);
3730
  } else if (QUERY_NODE_PHYSICAL_PLAN_MERGE_JOIN == type) {
3731
    pOptr = createMergeJoinOperatorInfo(ops, size, (SSortMergeJoinPhysiNode*)pPhyNode, pTaskInfo);
3732
  } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
H
Haojun Liao 已提交
3733
    pOptr = createFillOperatorInfo(ops[0], (SFillPhysiNode*)pPhyNode, pTaskInfo);
H
Haojun Liao 已提交
3734 3735
  } else if (QUERY_NODE_PHYSICAL_PLAN_INDEF_ROWS_FUNC == type) {
    pOptr = createIndefinitOutputOperatorInfo(ops[0], pPhyNode, pTaskInfo);
3736 3737
  } else if (QUERY_NODE_PHYSICAL_PLAN_INTERP_FUNC == type) {
    pOptr = createTimeSliceOperatorInfo(ops[0], pPhyNode, pTaskInfo);
H
Haojun Liao 已提交
3738 3739
  } else {
    ASSERT(0);
H
Haojun Liao 已提交
3740
  }
3741

3742
  taosMemoryFree(ops);
3743 3744 3745 3746
  if (pOptr) {
    pOptr->resultDataBlockId = pPhyNode->pOutputDataBlockDesc->dataBlockId;
  }

3747
  return pOptr;
3748
}
H
Haojun Liao 已提交
3749

H
Haojun Liao 已提交
3750
SArray* extractColumnInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
3751
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
3752 3753 3754 3755 3756 3757
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
3758 3759
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
H
Haojun Liao 已提交
3760

3761 3762 3763
    if (nodeType(pNode->pExpr) == QUERY_NODE_COLUMN) {
      SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

3764
      SColumn c = extractColumnFromColumnNode(pColNode);
3765 3766
      taosArrayPush(pList, &c);
    } else if (nodeType(pNode->pExpr) == QUERY_NODE_VALUE) {
L
Liu Jicong 已提交
3767 3768
      SValueNode* pValNode = (SValueNode*)pNode->pExpr;
      SColumn     c = {0};
3769
      c.slotId = pNode->slotId;
L
Liu Jicong 已提交
3770 3771 3772 3773
      c.colId = pNode->slotId;
      c.type = pValNode->node.type;
      c.bytes = pValNode->node.resType.bytes;
      c.scale = pValNode->node.resType.scale;
3774 3775 3776 3777
      c.precision = pValNode->node.resType.precision;

      taosArrayPush(pList, &c);
    }
H
Haojun Liao 已提交
3778 3779 3780 3781 3782
  }

  return pList;
}

L
Liu Jicong 已提交
3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795
static int32_t extractTbscanInStreamOpTree(SOperatorInfo* pOperator, STableScanInfo** ppInfo) {
  if (pOperator->operatorType != QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN) {
    if (pOperator->numOfDownstream == 0) {
      qError("failed to find stream scan operator");
      return TSDB_CODE_QRY_APP_ERROR;
    }

    if (pOperator->numOfDownstream > 1) {
      qError("join not supported for stream block scan");
      return TSDB_CODE_QRY_APP_ERROR;
    }
    return extractTbscanInStreamOpTree(pOperator->pDownstream[0], ppInfo);
  } else {
3796 3797 3798
    SStreamScanInfo* pInfo = pOperator->info;
    ASSERT(pInfo->pTableScanOp->operatorType == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN);
    *ppInfo = pInfo->pTableScanOp->info;
L
Liu Jicong 已提交
3799 3800 3801 3802
    return 0;
  }
}

3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824
int32_t extractTableScanNode(SPhysiNode* pNode, STableScanPhysiNode** ppNode) {
  if (pNode->pChildren == NULL || LIST_LENGTH(pNode->pChildren) == 0) {
    if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == pNode->type) {
      *ppNode = (STableScanPhysiNode*)pNode;
      return 0;
    } else {
      ASSERT(0);
      terrno = TSDB_CODE_QRY_APP_ERROR;
      return -1;
    }
  } else {
    if (LIST_LENGTH(pNode->pChildren) != 1) {
      ASSERT(0);
      terrno = TSDB_CODE_QRY_APP_ERROR;
      return -1;
    }
    SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pNode->pChildren, 0);
    return extractTableScanNode(pChildNode, ppNode);
  }
  return -1;
}

3825
#if 0
L
Liu Jicong 已提交
3826 3827 3828 3829 3830
int32_t rebuildReader(SOperatorInfo* pOperator, SSubplan* plan, SReadHandle* pHandle, int64_t uid, int64_t ts) {
  STableScanInfo* pTableScanInfo = NULL;
  if (extractTbscanInStreamOpTree(pOperator, &pTableScanInfo) < 0) {
    return -1;
  }
3831

L
Liu Jicong 已提交
3832 3833 3834 3835
  STableScanPhysiNode* pNode = NULL;
  if (extractTableScanNode(plan->pNode, &pNode) < 0) {
    ASSERT(0);
  }
3836

H
Haojun Liao 已提交
3837
  tsdbReaderClose(pTableScanInfo->dataReader);
3838

L
Liu Jicong 已提交
3839
  STableListInfo info = {0};
H
Haojun Liao 已提交
3840
  pTableScanInfo->dataReader = doCreateDataReader(pNode, pHandle, &info, NULL);
L
Liu Jicong 已提交
3841 3842 3843 3844
  if (pTableScanInfo->dataReader == NULL) {
    ASSERT(0);
    qError("failed to create data reader");
    return TSDB_CODE_QRY_APP_ERROR;
3845
  }
L
Liu Jicong 已提交
3846
  // TODO: set uid and ts to data reader
3847 3848
  return 0;
}
3849
#endif
3850

C
Cary Xu 已提交
3851
int32_t encodeOperator(SOperatorInfo* ops, char** result, int32_t* length, int32_t* nOptrWithVal) {
wmmhello's avatar
wmmhello 已提交
3852
  int32_t code = TDB_CODE_SUCCESS;
3853
  char*   pCurrent = NULL;
wmmhello's avatar
wmmhello 已提交
3854
  int32_t currLength = 0;
3855
  if (ops->fpSet.encodeResultRow) {
C
Cary Xu 已提交
3856
    if (result == NULL || length == NULL || nOptrWithVal == NULL) {
wmmhello's avatar
wmmhello 已提交
3857 3858 3859
      return TSDB_CODE_TSC_INVALID_INPUT;
    }
    code = ops->fpSet.encodeResultRow(ops, &pCurrent, &currLength);
wmmhello's avatar
wmmhello 已提交
3860

3861 3862
    if (code != TDB_CODE_SUCCESS) {
      if (*result != NULL) {
wmmhello's avatar
wmmhello 已提交
3863 3864 3865 3866
        taosMemoryFree(*result);
        *result = NULL;
      }
      return code;
C
Cary Xu 已提交
3867 3868 3869
    } else if (currLength == 0) {
      ASSERT(!pCurrent);
      goto _downstream;
wmmhello's avatar
wmmhello 已提交
3870
    }
wmmhello's avatar
wmmhello 已提交
3871

C
Cary Xu 已提交
3872 3873
    ++(*nOptrWithVal);

C
Cary Xu 已提交
3874
    ASSERT(currLength >= 0);
wmmhello's avatar
wmmhello 已提交
3875

3876
    if (*result == NULL) {
wmmhello's avatar
wmmhello 已提交
3877
      *result = (char*)taosMemoryCalloc(1, currLength + sizeof(int32_t));
wmmhello's avatar
wmmhello 已提交
3878 3879 3880 3881 3882 3883
      if (*result == NULL) {
        taosMemoryFree(pCurrent);
        return TSDB_CODE_OUT_OF_MEMORY;
      }
      memcpy(*result + sizeof(int32_t), pCurrent, currLength);
      *(int32_t*)(*result) = currLength + sizeof(int32_t);
3884
    } else {
wmmhello's avatar
wmmhello 已提交
3885
      int32_t sizePre = *(int32_t*)(*result);
3886
      char*   tmp = (char*)taosMemoryRealloc(*result, sizePre + currLength);
wmmhello's avatar
wmmhello 已提交
3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898
      if (tmp == NULL) {
        taosMemoryFree(pCurrent);
        taosMemoryFree(*result);
        *result = NULL;
        return TSDB_CODE_OUT_OF_MEMORY;
      }
      *result = tmp;
      memcpy(*result + sizePre, pCurrent, currLength);
      *(int32_t*)(*result) += currLength;
    }
    taosMemoryFree(pCurrent);
    *length = *(int32_t*)(*result);
wmmhello's avatar
wmmhello 已提交
3899 3900
  }

C
Cary Xu 已提交
3901
_downstream:
wmmhello's avatar
wmmhello 已提交
3902
  for (int32_t i = 0; i < ops->numOfDownstream; ++i) {
C
Cary Xu 已提交
3903
    code = encodeOperator(ops->pDownstream[i], result, length, nOptrWithVal);
3904
    if (code != TDB_CODE_SUCCESS) {
wmmhello's avatar
wmmhello 已提交
3905
      return code;
wmmhello's avatar
wmmhello 已提交
3906 3907
    }
  }
wmmhello's avatar
wmmhello 已提交
3908
  return TDB_CODE_SUCCESS;
wmmhello's avatar
wmmhello 已提交
3909 3910
}

H
Haojun Liao 已提交
3911
int32_t decodeOperator(SOperatorInfo* ops, const char* result, int32_t length) {
wmmhello's avatar
wmmhello 已提交
3912
  int32_t code = TDB_CODE_SUCCESS;
3913 3914
  if (ops->fpSet.decodeResultRow) {
    if (result == NULL) {
wmmhello's avatar
wmmhello 已提交
3915 3916
      return TSDB_CODE_TSC_INVALID_INPUT;
    }
H
Haojun Liao 已提交
3917

3918
    ASSERT(length == *(int32_t*)result);
H
Haojun Liao 已提交
3919 3920

    const char* data = result + sizeof(int32_t);
L
Liu Jicong 已提交
3921
    code = ops->fpSet.decodeResultRow(ops, (char*)data);
3922
    if (code != TDB_CODE_SUCCESS) {
wmmhello's avatar
wmmhello 已提交
3923 3924
      return code;
    }
wmmhello's avatar
wmmhello 已提交
3925

wmmhello's avatar
wmmhello 已提交
3926
    int32_t totalLength = *(int32_t*)result;
3927 3928
    int32_t dataLength = *(int32_t*)data;

3929
    if (totalLength == dataLength + sizeof(int32_t)) {  // the last data
wmmhello's avatar
wmmhello 已提交
3930 3931
      result = NULL;
      length = 0;
3932
    } else {
wmmhello's avatar
wmmhello 已提交
3933 3934 3935 3936
      result += dataLength;
      *(int32_t*)(result) = totalLength - dataLength;
      length = totalLength - dataLength;
    }
wmmhello's avatar
wmmhello 已提交
3937 3938
  }

wmmhello's avatar
wmmhello 已提交
3939 3940
  for (int32_t i = 0; i < ops->numOfDownstream; ++i) {
    code = decodeOperator(ops->pDownstream[i], result, length);
3941
    if (code != TDB_CODE_SUCCESS) {
wmmhello's avatar
wmmhello 已提交
3942
      return code;
wmmhello's avatar
wmmhello 已提交
3943 3944
    }
  }
wmmhello's avatar
wmmhello 已提交
3945
  return TDB_CODE_SUCCESS;
wmmhello's avatar
wmmhello 已提交
3946 3947
}

D
dapan1121 已提交
3948
int32_t createDataSinkParam(SDataSinkNode* pNode, void** pParam, qTaskInfo_t* pTaskInfo, SReadHandle* readHandle) {
D
dapan1121 已提交
3949
  SExecTaskInfo* pTask = *(SExecTaskInfo**)pTaskInfo;
3950

D
dapan1121 已提交
3951
  switch (pNode->type) {
D
dapan1121 已提交
3952 3953 3954 3955 3956 3957
    case QUERY_NODE_PHYSICAL_PLAN_QUERY_INSERT: {
      SInserterParam* pInserterParam = taosMemoryCalloc(1, sizeof(SInserterParam));
      if (NULL == pInserterParam) {
        return TSDB_CODE_OUT_OF_MEMORY;
      }
      pInserterParam->readHandle = readHandle;
L
Liu Jicong 已提交
3958

D
dapan1121 已提交
3959 3960 3961
      *pParam = pInserterParam;
      break;
    }
D
dapan1121 已提交
3962
    case QUERY_NODE_PHYSICAL_PLAN_DELETE: {
3963
      SDeleterParam* pDeleterParam = taosMemoryCalloc(1, sizeof(SDeleterParam));
D
dapan1121 已提交
3964 3965 3966 3967
      if (NULL == pDeleterParam) {
        return TSDB_CODE_OUT_OF_MEMORY;
      }
      int32_t tbNum = taosArrayGetSize(pTask->tableqinfoList.pTableList);
D
dapan1121 已提交
3968
      pDeleterParam->suid = pTask->tableqinfoList.suid;
D
dapan1121 已提交
3969 3970 3971 3972 3973 3974
      pDeleterParam->pUidList = taosArrayInit(tbNum, sizeof(uint64_t));
      if (NULL == pDeleterParam->pUidList) {
        taosMemoryFree(pDeleterParam);
        return TSDB_CODE_OUT_OF_MEMORY;
      }
      for (int32_t i = 0; i < tbNum; ++i) {
3975
        STableKeyInfo* pTable = taosArrayGet(pTask->tableqinfoList.pTableList, i);
D
dapan1121 已提交
3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988
        taosArrayPush(pDeleterParam->pUidList, &pTable->uid);
      }

      *pParam = pDeleterParam;
      break;
    }
    default:
      break;
  }

  return TSDB_CODE_SUCCESS;
}

dengyihao's avatar
dengyihao 已提交
3989
int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
D
dapan1121 已提交
3990
                               char* sql, EOPTR_EXEC_MODEL model) {
H
Haojun Liao 已提交
3991 3992
  uint64_t queryId = pPlan->id.queryId;

H
Haojun Liao 已提交
3993
  int32_t code = TSDB_CODE_SUCCESS;
D
dapan1121 已提交
3994
  *pTaskInfo = createExecTaskInfo(queryId, taskId, model, pPlan->dbFName);
H
Haojun Liao 已提交
3995 3996 3997 3998
  if (*pTaskInfo == NULL) {
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    goto _complete;
  }
H
Haojun Liao 已提交
3999

H
Haojun Liao 已提交
4000 4001 4002 4003
  if (pHandle && pHandle->pStateBackend) {
    (*pTaskInfo)->streamInfo.pState = pHandle->pStateBackend;
  }

4004
  (*pTaskInfo)->sql = sql;
D
dapan1121 已提交
4005
  sql = NULL;
4006
  (*pTaskInfo)->pSubplan = pPlan;
L
Liu Jicong 已提交
4007 4008
  (*pTaskInfo)->pRoot = createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, &(*pTaskInfo)->tableqinfoList,
                                           pPlan->pTagCond, pPlan->pTagIndexCond, pPlan->user);
L
Liu Jicong 已提交
4009

D
dapan1121 已提交
4010
  if (NULL == (*pTaskInfo)->pRoot) {
4011
    code = (*pTaskInfo)->code;
D
dapan1121 已提交
4012
    goto _complete;
4013 4014
  }

H
Haojun Liao 已提交
4015 4016
  return code;

H
Haojun Liao 已提交
4017
_complete:
D
dapan1121 已提交
4018
  taosMemoryFree(sql);
H
Haojun Liao 已提交
4019
  doDestroyTask(*pTaskInfo);
H
Haojun Liao 已提交
4020 4021
  terrno = code;
  return code;
H
Haojun Liao 已提交
4022 4023
}

4024
void doDestroyTableList(STableListInfo* pTableqinfoList) {
wmmhello's avatar
wmmhello 已提交
4025 4026
  taosArrayDestroy(pTableqinfoList->pTableList);
  taosHashCleanup(pTableqinfoList->map);
4027 4028
  if (pTableqinfoList->needSortTableByGroupId) {
    for (int32_t i = 0; i < taosArrayGetSize(pTableqinfoList->pGroupList); i++) {
wmmhello's avatar
wmmhello 已提交
4029
      SArray* tmp = taosArrayGetP(pTableqinfoList->pGroupList, i);
4030 4031 4032
      if (tmp == pTableqinfoList->pTableList) {
        continue;
      }
wmmhello's avatar
wmmhello 已提交
4033 4034 4035 4036
      taosArrayDestroy(tmp);
    }
  }
  taosArrayDestroy(pTableqinfoList->pGroupList);
4037

wmmhello's avatar
wmmhello 已提交
4038 4039
  pTableqinfoList->pTableList = NULL;
  pTableqinfoList->map = NULL;
4040 4041
}

L
Liu Jicong 已提交
4042
void doDestroyTask(SExecTaskInfo* pTaskInfo) {
H
Haojun Liao 已提交
4043 4044
  qDebug("%s execTask is freed", GET_TASKID(pTaskInfo));

wmmhello's avatar
wmmhello 已提交
4045
  doDestroyTableList(&pTaskInfo->tableqinfoList);
H
Haojun Liao 已提交
4046
  destroyOperatorInfo(pTaskInfo->pRoot);
4047
  cleanupTableSchemaInfo(&pTaskInfo->schemaInfo);
4048
  cleanupStreamInfo(&pTaskInfo->streamInfo);
4049 4050

  nodesDestroyNode((SNode*)pTaskInfo->pSubplan);
4051

wafwerar's avatar
wafwerar 已提交
4052 4053 4054
  taosMemoryFreeClear(pTaskInfo->sql);
  taosMemoryFreeClear(pTaskInfo->id.str);
  taosMemoryFreeClear(pTaskInfo);
4055 4056 4057 4058
}

static int64_t getQuerySupportBufSize(size_t numOfTables) {
  size_t s1 = sizeof(STableQueryInfo);
L
Liu Jicong 已提交
4059 4060
  //  size_t s3 = sizeof(STableCheckInfo);  buffer consumption in tsdb
  return (int64_t)(s1 * 1.5 * numOfTables);
4061 4062 4063 4064 4065 4066 4067
}

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
4068
    while (1) {
4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094
      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
4095

H
Haojun Liao 已提交
4096
int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SArray* pExecInfoList) {
4097
  SExplainExecInfo  execInfo = {0};
H
Haojun Liao 已提交
4098
  SExplainExecInfo* pExplainInfo = taosArrayPush(pExecInfoList, &execInfo);
4099

H
Haojun Liao 已提交
4100 4101 4102 4103 4104
  pExplainInfo->numOfRows = operatorInfo->resultInfo.totalRows;
  pExplainInfo->startupCost = operatorInfo->cost.openCost;
  pExplainInfo->totalCost = operatorInfo->cost.totalCost;
  pExplainInfo->verboseLen = 0;
  pExplainInfo->verboseInfo = NULL;
D
dapan1121 已提交
4105

4106
  if (operatorInfo->fpSet.getExplainFn) {
4107 4108
    int32_t code =
        operatorInfo->fpSet.getExplainFn(operatorInfo, &pExplainInfo->verboseInfo, &pExplainInfo->verboseLen);
D
dapan1121 已提交
4109
    if (code) {
4110
      qError("%s operator getExplainFn failed, code:%s", GET_TASKID(operatorInfo->pTaskInfo), tstrerror(code));
D
dapan1121 已提交
4111 4112 4113
      return code;
    }
  }
dengyihao's avatar
dengyihao 已提交
4114

D
dapan1121 已提交
4115
  int32_t code = 0;
D
dapan1121 已提交
4116
  for (int32_t i = 0; i < operatorInfo->numOfDownstream; ++i) {
H
Haojun Liao 已提交
4117 4118
    code = getOperatorExplainExecInfo(operatorInfo->pDownstream[i], pExecInfoList);
    if (code != TSDB_CODE_SUCCESS) {
4119
      //      taosMemoryFreeClear(*pRes);
D
dapan1121 已提交
4120 4121 4122 4123 4124
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
D
dapan1121 已提交
4125
}
5
54liuyao 已提交
4126

L
Liu Jicong 已提交
4127
int32_t initStreamAggSupporter(SStreamAggSupporter* pSup, const char* pKey, SqlFunctionCtx* pCtx, int32_t numOfOutput,
4128
                               int32_t size) {
4129
  pSup->currentPageId = -1;
4130
  pSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
5
54liuyao 已提交
4131 4132
  pSup->keySize = sizeof(int64_t) + sizeof(TSKEY);
  pSup->pKeyBuf = taosMemoryCalloc(1, pSup->keySize);
4133 4134
  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
  pSup->pResultRows = taosHashInit(1024, hashFn, false, HASH_NO_LOCK);
5
54liuyao 已提交
4135 4136 4137
  if (pSup->pKeyBuf == NULL || pSup->pResultRows == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }
4138
  pSup->valueSize = size;
5
54liuyao 已提交
4139

4140
  pSup->pScanBlock = createSpecialDataBlock(STREAM_CLEAR);
5
54liuyao 已提交
4141 4142 4143 4144 4145 4146 4147 4148 4149
  int32_t pageSize = 4096;
  while (pageSize < pSup->resultRowSize * 4) {
    pageSize <<= 1u;
  }
  // at least four pages need to be in buffer
  int32_t bufSize = 4096 * 256;
  if (bufSize <= pageSize) {
    bufSize = pageSize * 4;
  }
wafwerar's avatar
wafwerar 已提交
4150 4151 4152 4153 4154 4155
  if (!osTempSpaceAvailable()) {
    terrno = TSDB_CODE_NO_AVAIL_DISK;
    qError("Init stream agg supporter failed since %s", terrstr(terrno));
    return terrno;
  }
  int32_t code = createDiskbasedBuf(&pSup->pResultBuf, pageSize, bufSize, pKey, tsTempDir);
L
Liu Jicong 已提交
4156
  for (int32_t i = 0; i < numOfOutput; ++i) {
4157
    pCtx[i].saveHandle.pBuf = pSup->pResultBuf;
4158
  }
4159

4160
  return code;
5
54liuyao 已提交
4161
}
4162 4163 4164 4165 4166 4167 4168 4169 4170 4171

int32_t setOutputBuf(STimeWindow* win, SResultRow** pResult, int64_t tableGroupId, SqlFunctionCtx* pCtx,
                     int32_t numOfOutput, int32_t* rowEntryInfoOffset, SAggSupporter* pAggSup,
                     SExecTaskInfo* pTaskInfo) {
  SWinKey key = {
      .ts = win->skey,
      .groupId = tableGroupId,
  };
  char*   value = NULL;
  int32_t size = pAggSup->resultRowSize;
5
54liuyao 已提交
4172 4173

  tSimpleHashPut(pAggSup->pResultRowHashTable, &key, sizeof(SWinKey), NULL, 0);
4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186
  if (streamStateAddIfNotExist(pTaskInfo->streamInfo.pState, &key, (void**)&value, &size) < 0) {
    return TSDB_CODE_QRY_OUT_OF_MEMORY;
  }
  *pResult = (SResultRow*)value;
  ASSERT(*pResult);
  // set time window for current result
  (*pResult)->win = (*win);
  setResultRowInitCtx(*pResult, pCtx, numOfOutput, rowEntryInfoOffset);
  return TSDB_CODE_SUCCESS;
}

int32_t releaseOutputBuf(SExecTaskInfo* pTaskInfo, SWinKey* pKey, SResultRow* pResult) {
  streamStateReleaseBuf(pTaskInfo->streamInfo.pState, pKey, pResult);
4187
  /*taosMemoryFree((*(void**)pResult));*/
4188 4189 4190
  return TSDB_CODE_SUCCESS;
}

5
54liuyao 已提交
4191
int32_t saveOutputBuf(SExecTaskInfo* pTaskInfo, SWinKey* pKey, SResultRow* pResult, int32_t resSize) {
4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246
  streamStatePut(pTaskInfo->streamInfo.pState, pKey, pResult, resSize);
  return TSDB_CODE_SUCCESS;
}

int32_t buildDataBlockFromGroupRes(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock, SExprSupp* pSup,
                                   SGroupResInfo* pGroupResInfo) {
  SExprInfo*      pExprInfo = pSup->pExprInfo;
  int32_t         numOfExprs = pSup->numOfExprs;
  int32_t*        rowEntryOffset = pSup->rowEntryInfoOffset;
  SqlFunctionCtx* pCtx = pSup->pCtx;

  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);

  for (int32_t i = pGroupResInfo->index; i < numOfRows; i += 1) {
    SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
    int32_t     size = 0;
    void*       pVal = NULL;
    SWinKey     key = {
            .ts = *(TSKEY*)pPos->key,
            .groupId = pPos->groupId,
    };
    int32_t code = streamStateGet(pTaskInfo->streamInfo.pState, &key, &pVal, &size);
    ASSERT(code == 0);
    SResultRow* pRow = (SResultRow*)pVal;
    doUpdateNumOfRows(pCtx, pRow, numOfExprs, rowEntryOffset);
    // no results, continue to check the next one
    if (pRow->numOfRows == 0) {
      pGroupResInfo->index += 1;
      releaseOutputBuf(pTaskInfo, &key, pRow);
      continue;
    }

    if (pBlock->info.groupId == 0) {
      pBlock->info.groupId = pPos->groupId;
    } else {
      // current value belongs to different group, it can't be packed into one datablock
      if (pBlock->info.groupId != pPos->groupId) {
        releaseOutputBuf(pTaskInfo, &key, pRow);
        break;
      }
    }

    if (pBlock->info.rows + pRow->numOfRows > pBlock->info.capacity) {
      ASSERT(pBlock->info.rows > 0);
      releaseOutputBuf(pTaskInfo, &key, pRow);
      break;
    }

    pGroupResInfo->index += 1;

    for (int32_t j = 0; j < numOfExprs; ++j) {
      int32_t slotId = pExprInfo[j].base.resSchema.slotId;

      pCtx[j].resultInfo = getResultEntryInfo(pRow, j, rowEntryOffset);
      if (pCtx[j].fpSet.finalize) {
4247 4248 4249 4250
        int32_t code1 = pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
        if (TAOS_FAILED(code1)) {
          qError("%s build result data block error, code %s", GET_TASKID(pTaskInfo), tstrerror(code1));
          T_LONG_JMP(pTaskInfo->env, code1);
4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263
        }
      } else if (strcmp(pCtx[j].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
        // do nothing, todo refactor
      } else {
        // expand the result into multiple rows. E.g., _wstart, top(k, 20)
        // the _wstart 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);
        }
      }
    }
5
54liuyao 已提交
4264

4265
    pBlock->info.rows += pRow->numOfRows;
5
54liuyao 已提交
4266
    releaseOutputBuf(pTaskInfo, &key, pRow);
4267 4268 4269 4270
  }
  blockDataUpdateTsWindow(pBlock, 0);
  return TSDB_CODE_SUCCESS;
}