executorimpl.c 200.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
 *
 * This program is free software: you can use, redistribute, and/or modify
 * it under the terms of the GNU Affero General Public License, version 3
 * or later ("AGPL"), as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.
 *
 * You should have received a copy of the GNU Affero General Public License
 * along with this program. If not, see <http://www.gnu.org/licenses/>.
 */
15

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

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

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

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

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

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

enum {
L
Liu Jicong 已提交
51 52
  TS_JOIN_TS_EQUAL = 0,
  TS_JOIN_TS_NOT_EQUALS = 1,
53 54 55 56 57
  TS_JOIN_TAG_NOT_EQUALS = 2,
};

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

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

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

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

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

#define CLEAR_QUERY_STATUS(q, st)   ((q)->status &= (~(st)))
#define GET_NUM_OF_TABLEGROUP(q)    taosArrayGetSize((q)->tableqinfoGroupInfo.pGroupList)
#define QUERY_IS_INTERVAL_QUERY(_q) ((_q)->interval.interval > 0)

L
Liu Jicong 已提交
94 95 96
int32_t getMaximumIdleDurationSec() { return tsShellActivityTimer * 2; }

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

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

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

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

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

L
Liu Jicong 已提交
110
static void    releaseQueryBuf(size_t numOfTables);
111

H
Haojun Liao 已提交
112
static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr);
113 114 115 116 117 118

static void destroySFillOperatorInfo(void* param, int32_t numOfOutput);
static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput);
static void destroyTagScanOperatorInfo(void* param, int32_t numOfOutput);
static void destroyOrderOperatorInfo(void* param, int32_t numOfOutput);
static void destroyAggOperatorInfo(void* param, int32_t numOfOutput);
X
Xiaoyu Wang 已提交
119

H
Haojun Liao 已提交
120
static void destroyIntervalOperatorInfo(void* param, int32_t numOfOutput);
H
Haojun Liao 已提交
121 122
static void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput);

123
static void destroyOperatorInfo(SOperatorInfo* pOperator);
124
static void destroySysTableScannerOperatorInfo(void* param, int32_t numOfOutput);
125

126
void doSetOperatorCompleted(SOperatorInfo* pOperator) {
127
  pOperator->status = OP_EXEC_DONE;
H
Haojun Liao 已提交
128
  if (pOperator->pTaskInfo != NULL) {
129
    setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
130 131
  }
}
132

H
Haojun Liao 已提交
133
int32_t operatorDummyOpenFn(SOperatorInfo* pOperator) {
134
  OPTR_SET_OPENED(pOperator);
135
  pOperator->cost.openCost = 0;
H
Haojun Liao 已提交
136
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
137 138
}

139
SOperatorFpSet createOperatorFpSet(__optr_open_fn_t openFn, __optr_fn_t nextFn, __optr_fn_t streamFn,
L
Liu Jicong 已提交
140 141
                                   __optr_fn_t cleanup, __optr_close_fn_t closeFn, __optr_encode_fn_t encode,
                                   __optr_decode_fn_t decode, __optr_get_explain_fn_t explain) {
142 143 144 145 146 147 148 149 150 151 152 153 154 155
  SOperatorFpSet fpSet = {
      ._openFn = openFn,
      .getNextFn = nextFn,
      .getStreamResFn = streamFn,
      .cleanupFn = cleanup,
      .closeFn = closeFn,
      .encodeResultRow = encode,
      .decodeResultRow = decode,
      .getExplainFn = explain,
  };

  return fpSet;
}

H
Haojun Liao 已提交
156
void operatorDummyCloseFn(void* param, int32_t numOfCols) {}
H
Haojun Liao 已提交
157

158 159
static int32_t doCopyToSDataBlock(SExecTaskInfo* taskInfo, SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf, SGroupResInfo* pGroupResInfo,
                                  int32_t orderType, int32_t* rowCellOffset, SqlFunctionCtx* pCtx, int32_t numOfExprs);
H
Haojun Liao 已提交
160

161
static void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size);
H
Haojun Liao 已提交
162
static void setResultBufSize(STaskAttr* pQueryAttr, SResultInfo* pResultInfo);
dengyihao's avatar
dengyihao 已提交
163 164
static void doSetTableGroupOutputBuf(SAggOperatorInfo* pAggInfo, int32_t numOfOutput, uint64_t groupId,
                                     SExecTaskInfo* pTaskInfo);
165

H
Haojun Liao 已提交
166
SArray* getOrderCheckColumns(STaskAttr* pQuery);
167 168

typedef struct SRowCompSupporter {
L
Liu Jicong 已提交
169 170 171
  STaskRuntimeEnv* pRuntimeEnv;
  int16_t          dataOffset;
  __compar_fn_t    comFunc;
172 173
} SRowCompSupporter;

L
Liu Jicong 已提交
174 175 176
static int compareRowData(const void* a, const void* b, const void* userData) {
  const SResultRow* pRow1 = (const SResultRow*)a;
  const SResultRow* pRow2 = (const SResultRow*)b;
177

L
Liu Jicong 已提交
178 179
  SRowCompSupporter* supporter = (SRowCompSupporter*)userData;
  STaskRuntimeEnv*   pRuntimeEnv = supporter->pRuntimeEnv;
180

L
Liu Jicong 已提交
181 182
  SFilePage* page1 = getBufPage(pRuntimeEnv->pResultBuf, pRow1->pageId);
  SFilePage* page2 = getBufPage(pRuntimeEnv->pResultBuf, pRow2->pageId);
183 184

  int16_t offset = supporter->dataOffset;
L
Liu Jicong 已提交
185 186
  char*   in1 = getPosInResultPage(pRuntimeEnv->pQueryAttr, page1, pRow1->offset, offset);
  char*   in2 = getPosInResultPage(pRuntimeEnv->pQueryAttr, page2, pRow2->offset, offset);
187 188 189 190

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

L
Liu Jicong 已提交
191
// setup the output buffer for each operator
192
SSDataBlock* createResDataBlock(SDataBlockDescNode* pNode) {
H
Haojun Liao 已提交
193
  int32_t numOfCols = LIST_LENGTH(pNode->pSlots);
H
Haojun Liao 已提交
194

wafwerar's avatar
wafwerar 已提交
195
  SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
H
Haojun Liao 已提交
196 197
  pBlock->pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData));

H
Haojun Liao 已提交
198
  pBlock->info.blockId = pNode->dataBlockId;
dengyihao's avatar
dengyihao 已提交
199
  pBlock->info.rowSize = pNode->totalRowSize;  // todo ??
5
54liuyao 已提交
200
  pBlock->info.type = STREAM_INVALID;
H
Haojun Liao 已提交
201

L
Liu Jicong 已提交
202
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
203
    SColumnInfoData idata = {{0}};
L
Liu Jicong 已提交
204
    SSlotDescNode*  pDescNode = nodesListGetNode(pNode->pSlots, i);
X
Xiaoyu Wang 已提交
205 206 207
    //    if (!pDescNode->output) {  // todo disable it temporarily
    //      continue;
    //    }
H
Haojun Liao 已提交
208

dengyihao's avatar
dengyihao 已提交
209
    idata.info.type = pDescNode->dataType.type;
L
Liu Jicong 已提交
210 211
    idata.info.bytes = pDescNode->dataType.bytes;
    idata.info.scale = pDescNode->dataType.scale;
H
Haojun Liao 已提交
212
    idata.info.slotId = pDescNode->slotId;
H
Haojun Liao 已提交
213 214
    idata.info.precision = pDescNode->dataType.precision;

H
Hongze Cheng 已提交
215 216 217 218
    if (IS_VAR_DATA_TYPE(idata.info.type)) {
      pBlock->info.hasVarCol = true;
    }

H
Haojun Liao 已提交
219
    taosArrayPush(pBlock->pDataBlock, &idata);
H
Haojun Liao 已提交
220
  }
H
Haojun Liao 已提交
221

222
  pBlock->info.numOfCols = taosArrayGetSize(pBlock->pDataBlock);
H
Haojun Liao 已提交
223
  return pBlock;
H
Haojun Liao 已提交
224 225
}

L
Liu Jicong 已提交
226
static bool hasNull(SColumn* pColumn, SColumnDataAgg* pStatis) {
dengyihao's avatar
dengyihao 已提交
227 228
  if (TSDB_COL_IS_TAG(pColumn->flag) || TSDB_COL_IS_UD_COL(pColumn->flag) ||
      pColumn->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
229 230 231 232 233 234 235 236 237 238
    return false;
  }

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

  return true;
}

H
Haojun Liao 已提交
239
static void prepareResultListBuffer(SResultRowInfo* pResultRowInfo, jmp_buf env) {
240 241
  int64_t newCapacity = 0;

242 243 244 245 246 247 248 249 250 251 252
  // more than the capacity, reallocate the resources
  if (pResultRowInfo->size < pResultRowInfo->capacity) {
    return;
  }

  if (pResultRowInfo->capacity > 10000) {
    newCapacity = (int64_t)(pResultRowInfo->capacity * 1.25);
  } else {
    newCapacity = (int64_t)(pResultRowInfo->capacity * 1.5);
  }

253
  if (newCapacity <= pResultRowInfo->capacity) {
H
Haojun Liao 已提交
254 255 256
    newCapacity += 4;
  }

257 258 259 260 261 262
  char* p = taosMemoryRealloc(pResultRowInfo->pPosition, newCapacity * sizeof(SResultRowPosition));
  if (p == NULL) {
    longjmp(env, TSDB_CODE_OUT_OF_MEMORY);
  }

  pResultRowInfo->pPosition = (SResultRowPosition*)p;
263 264

  int32_t inc = (int32_t)newCapacity - pResultRowInfo->capacity;
wmmhello's avatar
wmmhello 已提交
265
  memset(&pResultRowInfo->pPosition[pResultRowInfo->capacity], 0, sizeof(SResultRowPosition) * inc);
266 267 268
  pResultRowInfo->capacity = (int32_t)newCapacity;
}

L
Liu Jicong 已提交
269 270
static bool chkResultRowFromKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, char* pData,
                                int16_t bytes, bool masterscan, uint64_t uid) {
271 272 273
  bool existed = false;
  SET_RES_WINDOW_KEY(pRuntimeEnv->keyBuf, pData, bytes, uid);

L
Liu Jicong 已提交
274 275
  SResultRow** p1 =
      (SResultRow**)taosHashGet(pRuntimeEnv->pResultRowHashTable, pRuntimeEnv->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
276 277 278 279 280 281 282 283 284 285 286

  // 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
287
        //        existed = (pResultRowInfo->pResult[0] == (*p1));
288 289
      } 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
290 291
        int64_t* index =
            taosHashGet(pRuntimeEnv->pResultRowListSet, pRuntimeEnv->keyBuf, GET_RES_EXT_WINDOW_KEY_LEN(bytes));
292 293 294 295 296 297 298 299 300 301 302 303 304 305
        if (index != NULL) {
          existed = true;
        } else {
          existed = false;
        }
      }
    }

    return existed;
  }

  return p1 != NULL;
}

306
SResultRow* getNewResultRow_rv(SDiskbasedBuf* pResultBuf, int64_t tableGroupId, int32_t interBufSize) {
L
Liu Jicong 已提交
307
  SFilePage* pData = NULL;
308 309 310 311 312 313 314 315 316 317 318 319 320

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

  if (taosArrayGetSize(list) == 0) {
    pData = getNewBufPage(pResultBuf, tableGroupId, &pageId);
    pData->num = sizeof(SFilePage);
  } else {
    SPageInfo* pi = getLastPageInfo(list);
    pData = getBufPage(pResultBuf, getPageId(pi));
    pageId = getPageId(pi);

wmmhello's avatar
wmmhello 已提交
321
    if (pData->num + interBufSize > getBufPageSize(pResultBuf)) {
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
      // release current page first, and prepare the next one
      releaseBufPageInfo(pResultBuf, pi);

      pData = getNewBufPage(pResultBuf, tableGroupId, &pageId);
      if (pData != NULL) {
        pData->num = sizeof(SFilePage);
      }
    }
  }

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

  // set the number of rows in current disk page
  SResultRow* pResultRow = (SResultRow*)((char*)pData + pData->num);
  pResultRow->pageId = pageId;
  pResultRow->offset = (int32_t)pData->num;

wmmhello's avatar
wmmhello 已提交
341
  pData->num += interBufSize;
342 343 344 345

  return pResultRow;
}

5
54liuyao 已提交
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
void doClearWindow(SIntervalAggOperatorInfo* pInfo, char* pData, int16_t bytes,
    uint64_t groupId, int32_t numOfOutput) {
  SAggSupporter* pSup = &pInfo->aggSup;
  SET_RES_WINDOW_KEY(pSup->keyBuf, pData, bytes, groupId);
  SResultRowPosition* p1 =
      (SResultRowPosition*)taosHashGet(pSup->pResultRowHashTable, pSup->keyBuf,
          GET_RES_WINDOW_KEY_LEN(bytes));
  SResultRow* pResult = getResultRowByPos(pSup->pResultBuf, p1);
  SqlFunctionCtx* pCtx = pInfo->binfo.pCtx;
  for (int32_t i = 0; i < numOfOutput; ++i) {
    pCtx[i].resultInfo = getResultCell(pResult, i, pInfo->binfo.rowCellInfoOffset);
    struct SResultRowEntryInfo* pResInfo = pCtx[i].resultInfo;
    if (fmIsWindowPseudoColumnFunc(pCtx[i].functionId)) {
      continue;
    }
    pResInfo->initialized = false;
    if (pCtx[i].functionId != -1) {
      pCtx[i].fpSet.init(&pCtx[i], pResInfo);
    }
  }
}

368 369 370 371 372 373 374
/**
 * the struct of key in hash table
 * +----------+---------------+
 * | group id |   key data    |
 * | 8 bytes  | actual length |
 * +----------+---------------+
 */
375 376 377
SResultRow* doSetResultOutBufByKey(SDiskbasedBuf* pResultBuf, SResultRowInfo* pResultRowInfo, char* pData,
                                   int16_t bytes, bool masterscan, uint64_t groupId, SExecTaskInfo* pTaskInfo,
                                   bool isIntervalQuery, SAggSupporter* pSup) {
378
  SET_RES_WINDOW_KEY(pSup->keyBuf, pData, bytes, groupId);
H
Haojun Liao 已提交
379

dengyihao's avatar
dengyihao 已提交
380 381
  SResultRowPosition* p1 =
      (SResultRowPosition*)taosHashGet(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes));
H
Haojun Liao 已提交
382

383 384
  SResultRow* pResult = NULL;

H
Haojun Liao 已提交
385 386
  // in case of repeat scan/reverse scan, no new time window added.
  if (isIntervalQuery) {
387 388
    if (masterscan && p1 != NULL) {  // the *p1 may be NULL in case of sliding+offset exists.
      pResult = getResultRowByPos(pResultBuf, p1);
H
Haojun Liao 已提交
389 390
    }
  } else {
dengyihao's avatar
dengyihao 已提交
391 392
    // In case of group by column query, the required SResultRow object must be existInCurrentResusltRowInfo in the
    // pResultRowInfo object.
H
Haojun Liao 已提交
393
    if (p1 != NULL) {
394
      pResult = getResultRowByPos(pResultBuf, p1);
H
Haojun Liao 已提交
395 396 397
    }
  }

L
Liu Jicong 已提交
398
  // 1. close current opened time window
399
  if (pResultRowInfo->cur.pageId != -1 && ((pResult == NULL) || (pResult->pageId != pResultRowInfo->cur.pageId &&
L
Liu Jicong 已提交
400
                                                                 pResult->offset != pResultRowInfo->cur.offset))) {
401 402 403 404 405 406 407 408 409 410 411
    // todo extract function
    SResultRowPosition pos = pResultRowInfo->cur;
    SFilePage*         pPage = getBufPage(pResultBuf, pos.pageId);
    SResultRow*        pRow = (SResultRow*)((char*)pPage + pos.offset);
    closeResultRow(pRow);
    releaseBufPage(pResultBuf, pPage);
  }

  // allocate a new buffer page
  prepareResultListBuffer(pResultRowInfo, pTaskInfo->env);
  if (pResult == NULL) {
H
Haojun Liao 已提交
412
    ASSERT(pSup->resultRowSize > 0);
413 414
    pResult = getNewResultRow_rv(pResultBuf, groupId, pSup->resultRowSize);
    initResultRow(pResult);
H
Haojun Liao 已提交
415

416 417
    // add a new result set for a new group
    SResultRowPosition pos = {.pageId = pResult->pageId, .offset = pResult->offset};
L
Liu Jicong 已提交
418 419
    taosHashPut(pSup->pResultRowHashTable, pSup->keyBuf, GET_RES_WINDOW_KEY_LEN(bytes), &pos,
                sizeof(SResultRowPosition));
H
Haojun Liao 已提交
420 421
  }

422 423 424 425 426
  // 2. set the new time window to be the new active time window
  pResultRowInfo->pPosition[pResultRowInfo->size++] =
      (SResultRowPosition){.pageId = pResult->pageId, .offset = pResult->offset};
  pResultRowInfo->cur = (SResultRowPosition){.pageId = pResult->pageId, .offset = pResult->offset};

H
Haojun Liao 已提交
427 428 429 430 431
  // too many time window in query
  if (pResultRowInfo->size > MAX_INTERVAL_TIME_WINDOW) {
    longjmp(pTaskInfo->env, TSDB_CODE_QRY_TOO_MANY_TIMEWINDOW);
  }

H
Haojun Liao 已提交
432
  return pResult;
H
Haojun Liao 已提交
433 434
}

435
// get the correct time window according to the handled timestamp
L
Liu Jicong 已提交
436
static STimeWindow getCurrentActiveTimeWindow(SResultRowInfo* pResultRowInfo, int64_t ts, STaskAttr* pQueryAttr) {
437
  STimeWindow w = {0};
H
Haojun Liao 已提交
438
#if 0
L
Liu Jicong 已提交
439 440
  if (pResultRowInfo->curPos == -1) {  // the first window, from the previous stored value
                                       //    getInitialStartTimeWindow(pQueryAttr, ts, &w);
441 442

    if (pQueryAttr->interval.intervalUnit == 'n' || pQueryAttr->interval.intervalUnit == 'y') {
L
Liu Jicong 已提交
443 444 445
      w.ekey =
          taosTimeAdd(w.skey, pQueryAttr->interval.interval, pQueryAttr->interval.intervalUnit, pQueryAttr->precision) -
          1;
446 447 448 449
    } else {
      w.ekey = w.skey + pQueryAttr->interval.interval - 1;
    }
  } else {
H
Haojun Liao 已提交
450
    w = pRow->win;
451 452 453 454 455 456 457 458 459
  }

  /*
   * query border check, skey should not be bounded by the query time range, since the value skey will
   * be used as the time window index value. So we only change ekey of time window accordingly.
   */
  if (w.ekey > pQueryAttr->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) {
    w.ekey = pQueryAttr->window.ekey;
  }
H
Haojun Liao 已提交
460
#endif
461 462 463 464 465

  return w;
}

// a new buffer page for each table. Needs to opt this design
L
Liu Jicong 已提交
466
static int32_t addNewWindowResultBuf(SResultRow* pWindowRes, SDiskbasedBuf* pResultBuf, int32_t tid, uint32_t size) {
467 468 469 470
  if (pWindowRes->pageId != -1) {
    return 0;
  }

L
Liu Jicong 已提交
471
  SFilePage* pData = NULL;
472 473 474 475 476 477

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

  if (taosArrayGetSize(list) == 0) {
H
Haojun Liao 已提交
478
    pData = getNewBufPage(pResultBuf, tid, &pageId);
479
    pData->num = sizeof(SFilePage);
480 481
  } else {
    SPageInfo* pi = getLastPageInfo(list);
482
    pData = getBufPage(pResultBuf, getPageId(pi));
483
    pageId = getPageId(pi);
484

485
    if (pData->num + size > getBufPageSize(pResultBuf)) {
486
      // release current page first, and prepare the next one
487
      releaseBufPageInfo(pResultBuf, pi);
488

H
Haojun Liao 已提交
489
      pData = getNewBufPage(pResultBuf, tid, &pageId);
490
      if (pData != NULL) {
491
        pData->num = sizeof(SFilePage);
492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511
      }
    }
  }

  if (pData == NULL) {
    return -1;
  }

  // set the number of rows in current disk page
  if (pWindowRes->pageId == -1) {  // not allocated yet, allocate new buffer
    pWindowRes->pageId = pageId;
    pWindowRes->offset = (int32_t)pData->num;

    pData->num += size;
    assert(pWindowRes->pageId >= 0);
  }

  return 0;
}

L
Liu Jicong 已提交
512 513 514
static bool chkWindowOutputBufByKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, STimeWindow* win,
                                    bool masterscan, SResultRow** pResult, int64_t groupId, SqlFunctionCtx* pCtx,
                                    int32_t numOfOutput, int32_t* rowCellInfoOffset) {
515
  assert(win->skey <= win->ekey);
L
Liu Jicong 已提交
516
  return chkResultRowFromKey(pRuntimeEnv, pResultRowInfo, (char*)&win->skey, TSDB_KEYSIZE, masterscan, groupId);
517 518
}

dengyihao's avatar
dengyihao 已提交
519 520
static void doUpdateResultRowIndex(SResultRowInfo* pResultRowInfo, TSKEY lastKey, bool ascQuery,
                                   bool timeWindowInterpo) {
521
  int64_t skey = TSKEY_INITIAL_VAL;
H
Haojun Liao 已提交
522
#if 0
523 524
  int32_t i = 0;
  for (i = pResultRowInfo->size - 1; i >= 0; --i) {
L
Liu Jicong 已提交
525
    SResultRow* pResult = pResultRowInfo->pResult[i];
526 527 528 529 530 531
    if (pResult->closed) {
      break;
    }

    // new closed result rows
    if (timeWindowInterpo) {
L
Liu Jicong 已提交
532 533 534
      if (pResult->endInterp &&
          ((pResult->win.skey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery))) {
        if (i > 0) {  // the first time window, the startInterp is false.
535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553
          assert(pResult->startInterp);
        }

        closeResultRow(pResultRowInfo, i);
      } else {
        skey = pResult->win.skey;
      }
    } else {
      if ((pResult->win.ekey <= lastKey && ascQuery) || (pResult->win.skey >= lastKey && !ascQuery)) {
        closeResultRow(pResultRowInfo, i);
      } else {
        skey = pResult->win.skey;
      }
    }
  }

  // all result rows are closed, set the last one to be the skey
  if (skey == TSKEY_INITIAL_VAL) {
    if (pResultRowInfo->size == 0) {
L
Liu Jicong 已提交
554
      //      assert(pResultRowInfo->current == NULL);
555 556 557 558 559 560 561
      assert(pResultRowInfo->curPos == -1);
      pResultRowInfo->curPos = -1;
    } else {
      pResultRowInfo->curPos = pResultRowInfo->size - 1;
    }
  } else {
    for (i = pResultRowInfo->size - 1; i >= 0; --i) {
L
Liu Jicong 已提交
562
      SResultRow* pResult = pResultRowInfo->pResult[i];
563 564 565 566 567 568 569 570 571 572 573
      if (pResult->closed) {
        break;
      }
    }

    if (i == pResultRowInfo->size - 1) {
      pResultRowInfo->curPos = i;
    } else {
      pResultRowInfo->curPos = i + 1;  // current not closed result object
    }
  }
H
Haojun Liao 已提交
574
#endif
575
}
576
//
dengyihao's avatar
dengyihao 已提交
577
// static void updateResultRowInfoActiveIndex(SResultRowInfo* pResultRowInfo, const STimeWindow* pWin, TSKEY lastKey,
578 579 580 581 582 583 584 585 586
//                                           bool ascQuery, bool interp) {
//  if ((lastKey > pWin->ekey && ascQuery) || (lastKey < pWin->ekey && (!ascQuery))) {
//    closeAllResultRows(pResultRowInfo);
//    pResultRowInfo->curPos = pResultRowInfo->size - 1;
//  } else {
//    int32_t step = ascQuery ? 1 : -1;
//    doUpdateResultRowIndex(pResultRowInfo, lastKey - step, ascQuery, interp);
//  }
//}
587

588
//  query_range_start, query_range_end, window_duration, window_start, window_end
589
void initExecTimeWindowInfo(SColumnInfoData* pColData, STimeWindow* pQueryWindow) {
590 591 592
  pColData->info.type = TSDB_DATA_TYPE_TIMESTAMP;
  pColData->info.bytes = sizeof(int64_t);

593
  colInfoDataEnsureCapacity(pColData, 0, 5);
594 595 596 597 598 599 600 601 602
  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);
}

603
void doApplyFunctions(SExecTaskInfo* taskInfo, SqlFunctionCtx* pCtx, STimeWindow* pWin, SColumnInfoData* pTimeWindowData, int32_t offset,
dengyihao's avatar
dengyihao 已提交
604
                      int32_t forwardStep, TSKEY* tsCol, int32_t numOfTotal, int32_t numOfOutput, int32_t order) {
605 606 607
  for (int32_t k = 0; k < numOfOutput; ++k) {
    pCtx[k].startTs = pWin->skey;

H
Haojun Liao 已提交
608
    // keep it temporarily
dengyihao's avatar
dengyihao 已提交
609 610
    bool    hasAgg = pCtx[k].input.colDataAggIsSet;
    int32_t numOfRows = pCtx[k].input.numOfRows;
H
Haojun Liao 已提交
611
    int32_t startOffset = pCtx[k].input.startRowIndex;
612

H
Haojun Liao 已提交
613
    int32_t pos = (order == TSDB_ORDER_ASC) ? offset : offset - (forwardStep - 1);
614 615
    pCtx[k].input.startRowIndex = pos;
    pCtx[k].input.numOfRows = forwardStep;
616 617

    if (tsCol != NULL) {
H
Haojun Liao 已提交
618
      pCtx[k].ptsList = tsCol;
619 620 621 622 623 624 625 626
    }

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

627 628
    if (fmIsWindowPseudoColumnFunc(pCtx[k].functionId)) {
      SResultRowEntryInfo* pEntryInfo = GET_RES_INFO(&pCtx[k]);
dengyihao's avatar
dengyihao 已提交
629
      char*                p = GET_ROWCELL_INTERBUF(pEntryInfo);
630

631
      SColumnInfoData idata = {0};
dengyihao's avatar
dengyihao 已提交
632
      idata.info.type = TSDB_DATA_TYPE_BIGINT;
633
      idata.info.bytes = tDataTypes[TSDB_DATA_TYPE_BIGINT].bytes;
dengyihao's avatar
dengyihao 已提交
634
      idata.pData = p;
635 636 637 638

      SScalarParam out = {.columnData = &idata};
      SScalarParam tw = {.numOfRows = 5, .columnData = pTimeWindowData};
      pCtx[k].sfp.process(&tw, 1, &out);
639 640 641
      pEntryInfo->numOfRes = 1;
      continue;
    }
642
    int32_t code = TSDB_CODE_SUCCESS;
643
    if (functionNeedToExecute(&pCtx[k]) && pCtx[k].fpSet.process != NULL) {
644 645 646 647 648 649
      code = pCtx[k].fpSet.process(&pCtx[k]);
      if (code != TSDB_CODE_SUCCESS) {
        qError("%s apply functions error, code: %s", GET_TASKID(taskInfo), tstrerror(code));
        taskInfo->code = code;
        longjmp(taskInfo->env, code);
      }
650 651 652
    }

    // restore it
653 654 655
    pCtx[k].input.colDataAggIsSet = hasAgg;
    pCtx[k].input.startRowIndex = startOffset;
    pCtx[k].input.numOfRows = numOfRows;
656 657 658
  }
}

L
Liu Jicong 已提交
659
static FORCE_INLINE TSKEY reviseWindowEkey(STaskAttr* pQueryAttr, STimeWindow* pWindow) {
dengyihao's avatar
dengyihao 已提交
660
  TSKEY   ekey = -1;
661 662
  int32_t order = TSDB_ORDER_ASC;
  if (order == TSDB_ORDER_ASC) {
663 664 665 666 667 668 669 670 671 672 673 674 675 676
    ekey = pWindow->ekey;
    if (ekey > pQueryAttr->window.ekey) {
      ekey = pQueryAttr->window.ekey;
    }
  } else {
    ekey = pWindow->skey;
    if (ekey < pQueryAttr->window.ekey) {
      ekey = pQueryAttr->window.ekey;
    }
  }

  return ekey;
}

dengyihao's avatar
dengyihao 已提交
677
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
678
                                   int32_t scanFlag, bool createDummyCol);
679

dengyihao's avatar
dengyihao 已提交
680 681
static void doSetInputDataBlockInfo(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock,
                                    int32_t order) {
682
  for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
683
    pCtx[i].order = order;
L
Liu Jicong 已提交
684
    pCtx[i].size = pBlock->info.rows;
685
    setBlockStatisInfo(&pCtx[i], &pOperator->pExpr[i], pBlock);
686 687 688
  }
}

689
void setInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order, int32_t scanFlag,
dengyihao's avatar
dengyihao 已提交
690
                       bool createDummyCol) {
691
  if (pBlock->pBlockAgg != NULL) {
H
Haojun Liao 已提交
692
    doSetInputDataBlockInfo(pOperator, pCtx, pBlock, order);
693
  } else {
694
    doSetInputDataBlock(pOperator, pCtx, pBlock, order, scanFlag, createDummyCol);
H
Haojun Liao 已提交
695
  }
696 697
}

L
Liu Jicong 已提交
698 699
static int32_t doCreateConstantValColumnInfo(SInputColumnInfoData* pInput, SFunctParam* pFuncParam, int32_t paramIndex,
                                             int32_t numOfRows) {
700 701 702 703 704 705 706 707
  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)
708 709
    pColInfo->info.type = pFuncParam->param.nType;
    pColInfo->info.bytes = pFuncParam->param.nLen;
710 711

    pInput->pData[paramIndex] = pColInfo;
712 713
  } else {
    pColInfo = pInput->pData[paramIndex];
714 715
  }

716
  colInfoDataEnsureCapacity(pColInfo, 0, numOfRows);
717

718
  int8_t type = pFuncParam->param.nType;
719 720
  if (type == TSDB_DATA_TYPE_BIGINT || type == TSDB_DATA_TYPE_UBIGINT) {
    int64_t v = pFuncParam->param.i;
dengyihao's avatar
dengyihao 已提交
721
    for (int32_t i = 0; i < numOfRows; ++i) {
722 723 724 725
      colDataAppendInt64(pColInfo, i, &v);
    }
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
dengyihao's avatar
dengyihao 已提交
726
    for (int32_t i = 0; i < numOfRows; ++i) {
727 728
      colDataAppendDouble(pColInfo, i, &v);
    }
729
  } else if (type == TSDB_DATA_TYPE_VARCHAR) {
L
Liu Jicong 已提交
730
    char* tmp = taosMemoryMalloc(pFuncParam->param.nLen + VARSTR_HEADER_SIZE);
731
    STR_WITH_SIZE_TO_VARSTR(tmp, pFuncParam->param.pz, pFuncParam->param.nLen);
L
Liu Jicong 已提交
732
    for (int32_t i = 0; i < numOfRows; ++i) {
733 734
      colDataAppend(pColInfo, i, tmp, false);
    }
735 736 737 738 739
  }

  return TSDB_CODE_SUCCESS;
}

dengyihao's avatar
dengyihao 已提交
740
static int32_t doSetInputDataBlock(SOperatorInfo* pOperator, SqlFunctionCtx* pCtx, SSDataBlock* pBlock, int32_t order,
741
    int32_t scanFlag, bool createDummyCol) {
742 743
  int32_t code = TSDB_CODE_SUCCESS;

744
  for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
L
Liu Jicong 已提交
745 746 747
    pCtx[i].order = order;
    pCtx[i].size = pBlock->info.rows;
    pCtx[i].pSrcBlock = pBlock;
748
    pCtx[i].scanFlag  = scanFlag;
H
Haojun Liao 已提交
749

750
    SInputColumnInfoData* pInput = &pCtx[i].input;
751
    pInput->uid = pBlock->info.uid;
C
Cary Xu 已提交
752
    pInput->colDataAggIsSet = false;
753

754 755
    SExprInfo* pOneExpr = &pOperator->pExpr[i];
    for (int32_t j = 0; j < pOneExpr->base.numOfParams; ++j) {
dengyihao's avatar
dengyihao 已提交
756
      SFunctParam* pFuncParam = &pOneExpr->base.pParam[j];
G
Ganlin Zhao 已提交
757 758
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
dengyihao's avatar
dengyihao 已提交
759
        pInput->pData[j] = taosArrayGet(pBlock->pDataBlock, slotId);
760 761 762
        pInput->totalRows = pBlock->info.rows;
        pInput->numOfRows = pBlock->info.rows;
        pInput->startRowIndex = 0;
763

764
        // NOTE: the last parameter is the primary timestamp column
765 766 767
        if (fmIsTimelineFunc(pCtx[i].functionId) && (j == pOneExpr->base.numOfParams - 1)) {
          pInput->pPTS = pInput->pData[j];
        }
768 769
        ASSERT(pInput->pData[j] != NULL);
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
770 771 772
        // 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) {
773 774 775 776
          pInput->totalRows = pBlock->info.rows;
          pInput->numOfRows = pBlock->info.rows;
          pInput->startRowIndex = 0;

777
          code = doCreateConstantValColumnInfo(pInput, pFuncParam, j, pBlock->info.rows);
778 779 780
          if (code != TSDB_CODE_SUCCESS) {
            return code;
          }
781
        }
G
Ganlin Zhao 已提交
782 783
      }
    }
H
Haojun Liao 已提交
784

H
Haojun Liao 已提交
785
    //    setBlockStatisInfo(&pCtx[i], pBlock, pOperator->pExpr[i].base.pColumns);
H
Haojun Liao 已提交
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823
    //      uint32_t flag = pOperator->pExpr[i].base.pParam[0].pCol->flag;
    //      if (TSDB_COL_IS_NORMAL_COL(flag) /*|| (pCtx[i].functionId == FUNCTION_BLKINFO) ||
    //          (TSDB_COL_IS_TAG(flag) && pOperator->pRuntimeEnv->scanFlag == MERGE_STAGE)*/) {

    //        SColumn* pCol = pOperator->pExpr[i].base.pParam[0].pCol;
    //        if (pCtx[i].columnIndex == -1) {
    //          for(int32_t j = 0; j < pBlock->info.numOfCols; ++j) {
    //            SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, j);
    //            if (pColData->info.colId == pCol->colId) {
    //              pCtx[i].columnIndex = j;
    //              break;
    //            }
    //          }
    //        }

    //        uint32_t status = aAggs[pCtx[i].functionId].status;
    //        if ((status & (FUNCSTATE_SELECTIVITY | FUNCSTATE_NEED_TS)) != 0) {
    //          SColumnInfoData* tsInfo = taosArrayGet(pBlock->pDataBlock, 0);
    // In case of the top/bottom query again the nest query result, which has no timestamp column
    // don't set the ptsList attribute.
    //          if (tsInfo->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
    //            pCtx[i].ptsList = (int64_t*) tsInfo->pData;
    //          } else {
    //            pCtx[i].ptsList = NULL;
    //          }
    //        }
    //      } else if (TSDB_COL_IS_UD_COL(pCol->flag) && (pOperator->pRuntimeEnv->scanFlag == MERGE_STAGE)) {
    //        SColIndex*       pColIndex = &pOperator->pExpr[i].base.colInfo;
    //        SColumnInfoData* p = taosArrayGet(pBlock->pDataBlock, pColIndex->colIndex);
    //
    //        pCtx[i].pInput = p->pData;
    //        assert(p->info.colId == pColIndex->info.colId && pCtx[i].inputType == p->info.type);
    //        for(int32_t j = 0; j < pBlock->info.rows; ++j) {
    //          char* dst = p->pData + j * p->info.bytes;
    //          taosVariantDump(&pOperator->pExpr[i].base.param[1], dst, p->info.type, true);
    //        }
    //      }
  }
824 825

  return code;
H
Haojun Liao 已提交
826 827
}

828
static int32_t doAggregateImpl(SOperatorInfo* pOperator, TSKEY startTs, SqlFunctionCtx* pCtx) {
829
  for (int32_t k = 0; k < pOperator->numOfExprs; ++k) {
H
Haojun Liao 已提交
830
    if (functionNeedToExecute(&pCtx[k])) {
831 832 833
      pCtx[k].startTs = startTs;
      // todo add a dummy funtion to avoid process check
      if (pCtx[k].fpSet.process != NULL) {
834 835
        int32_t code = pCtx[k].fpSet.process(&pCtx[k]);
        if (code != TSDB_CODE_SUCCESS) {
836 837
          qError("%s aggregate function error happens, code: %s", GET_TASKID(pOperator->pTaskInfo), tstrerror(code));
          return code;
838
        }
839
      }
840 841
    }
  }
842 843

  return TSDB_CODE_SUCCESS;
844 845
}

H
Haojun Liao 已提交
846
static void setPseudoOutputColInfo(SSDataBlock* pResult, SqlFunctionCtx* pCtx, SArray* pPseudoList) {
dengyihao's avatar
dengyihao 已提交
847
  size_t num = (pPseudoList != NULL) ? taosArrayGetSize(pPseudoList) : 0;
H
Haojun Liao 已提交
848 849 850 851 852
  for (int32_t i = 0; i < num; ++i) {
    pCtx[i].pOutput = taosArrayGet(pResult->pDataBlock, i);
  }
}

853
int32_t projectApplyFunctions(SExprInfo* pExpr, SSDataBlock* pResult, SSDataBlock* pSrcBlock, SqlFunctionCtx* pCtx,
X
Xiaoyu Wang 已提交
854
                              int32_t numOfOutput, SArray* pPseudoList) {
H
Haojun Liao 已提交
855
  setPseudoOutputColInfo(pResult, pCtx, pPseudoList);
H
Haojun Liao 已提交
856
  pResult->info.groupId = pSrcBlock->info.groupId;
H
Haojun Liao 已提交
857

dengyihao's avatar
dengyihao 已提交
858 859
  // if the source equals to the destination, it is to create a new column as the result of scalar function or some
  // operators.
860 861
  bool createNewColModel = (pResult == pSrcBlock);

862 863
  int32_t numOfRows = 0;

864
  for (int32_t k = 0; k < numOfOutput; ++k) {
dengyihao's avatar
dengyihao 已提交
865
    int32_t         outputSlotId = pExpr[k].base.resSchema.slotId;
866 867
    SqlFunctionCtx* pfCtx = &pCtx[k];

L
Liu Jicong 已提交
868
    if (pExpr[k].pExpr->nodeType == QUERY_NODE_COLUMN) {  // it is a project query
869
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
870
      if (pResult->info.rows > 0 && !createNewColModel) {
X
Xiaoyu Wang 已提交
871 872
        colDataMergeCol(pColInfoData, pResult->info.rows, &pResult->info.capacity, pfCtx->input.pData[0],
                        pfCtx->input.numOfRows);
873 874 875
      } else {
        colDataAssign(pColInfoData, pfCtx->input.pData[0], pfCtx->input.numOfRows);
      }
876

877
      numOfRows = pfCtx->input.numOfRows;
878
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_VALUE) {
879
      SColumnInfoData* pColInfoData = taosArrayGet(pResult->pDataBlock, outputSlotId);
880

dengyihao's avatar
dengyihao 已提交
881
      int32_t offset = createNewColModel ? 0 : pResult->info.rows;
882
      for (int32_t i = 0; i < pSrcBlock->info.rows; ++i) {
dengyihao's avatar
dengyihao 已提交
883 884 885
        colDataAppend(pColInfoData, i + offset,
                      taosVariantGet(&pExpr[k].base.pParam[0].param, pExpr[k].base.pParam[0].param.nType),
                      TSDB_DATA_TYPE_NULL == pExpr[k].base.pParam[0].param.nType);
886
      }
887 888

      numOfRows = pSrcBlock->info.rows;
H
Haojun Liao 已提交
889
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_OPERATOR) {
890 891 892
      SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
      taosArrayPush(pBlockList, &pSrcBlock);

893
      SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
894
      SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
895

896
      SScalarParam dest = {.columnData = &idata};
897 898
      scalarCalculate(pExpr[k].pExpr->_optrRoot.pRootNode, pBlockList, &dest);

dengyihao's avatar
dengyihao 已提交
899
      int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
900
      colInfoDataEnsureCapacity(pResColData, startOffset, pResult->info.capacity);
901
      colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
902 903

      numOfRows = dest.numOfRows;
904 905
      taosArrayDestroy(pBlockList);
    } else if (pExpr[k].pExpr->nodeType == QUERY_NODE_FUNCTION) {
906
      ASSERT(!fmIsAggFunc(pfCtx->functionId));
907

908 909
      // _rowts/_c0, not tbname column
      if (fmIsPseudoColumnFunc(pfCtx->functionId) && (!fmIsScanPseudoColumnFunc(pfCtx->functionId))) {
H
Haojun Liao 已提交
910
        // do nothing
911
      } else if (fmIsNonstandardSQLFunc(pfCtx->functionId)) {
dengyihao's avatar
dengyihao 已提交
912
        SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[k]);
913
        pfCtx->fpSet.init(&pCtx[k], pResInfo);
914

915
        pfCtx->pOutput = taosArrayGet(pResult->pDataBlock, outputSlotId);
dengyihao's avatar
dengyihao 已提交
916
        pfCtx->offset = createNewColModel ? 0 : pResult->info.rows;  // set the start offset
H
Haojun Liao 已提交
917

918
        // set the timestamp(_rowts) output buffer
919 920
        if (taosArrayGetSize(pPseudoList) > 0) {
          int32_t* outputColIndex = taosArrayGet(pPseudoList, 0);
921
          pfCtx->pTsOutput = (SColumnInfoData*)pCtx[*outputColIndex].pOutput;
922
        }
H
Haojun Liao 已提交
923

924
        numOfRows = pfCtx->fpSet.process(pfCtx);
H
Haojun Liao 已提交
925 926 927
      } else {
        SArray* pBlockList = taosArrayInit(4, POINTER_BYTES);
        taosArrayPush(pBlockList, &pSrcBlock);
G
Ganlin Zhao 已提交
928

929
        SColumnInfoData* pResColData = taosArrayGet(pResult->pDataBlock, outputSlotId);
930
        SColumnInfoData  idata = {.info = pResColData->info, .hasNull = true};
H
Haojun Liao 已提交
931

932
        SScalarParam dest = {.columnData = &idata};
X
Xiaoyu Wang 已提交
933
        int32_t      code = scalarCalculate((SNode*)pExpr[k].pExpr->_function.pFunctNode, pBlockList, &dest);
934 935 936 937
        if (code != TSDB_CODE_SUCCESS) {
          taosArrayDestroy(pBlockList);
          return code;
        }
938

dengyihao's avatar
dengyihao 已提交
939
        int32_t startOffset = createNewColModel ? 0 : pResult->info.rows;
940
        colInfoDataEnsureCapacity(pResColData, startOffset, pResult->info.capacity);
941
        colDataMergeCol(pResColData, startOffset, &pResult->info.capacity, &idata, dest.numOfRows);
942 943

        numOfRows = dest.numOfRows;
H
Haojun Liao 已提交
944 945
        taosArrayDestroy(pBlockList);
      }
946
    } else {
947
      ASSERT(0);
948 949
    }
  }
950

951 952 953
  if (!createNewColModel) {
    pResult->info.rows += numOfRows;
  }
954 955

  return TSDB_CODE_SUCCESS;
956 957
}

958 959 960
static void setResultRowKey(SResultRow* pResultRow, char* pData, int16_t type) {
  if (IS_VAR_DATA_TYPE(type)) {
    // todo disable this
961

962 963 964 965 966 967 968 969 970
    //    if (pResultRow->key == NULL) {
    //      pResultRow->key = taosMemoryMalloc(varDataTLen(pData));
    //      varDataCopy(pResultRow->key, pData);
    //    } else {
    //      ASSERT(memcmp(pResultRow->key, pData, varDataTLen(pData)) == 0);
    //    }
  } else {
    int64_t v = -1;
    GET_TYPED_DATA(v, int64_t, type, pData);
971

972 973
    pResultRow->win.skey = v;
    pResultRow->win.ekey = v;
974 975 976
  }
}

977 978 979 980 981
int32_t setGroupResultOutputBuf(SOptrBasicInfo* binfo, int32_t numOfCols, char* pData, int16_t type, int16_t bytes,
                                int32_t groupId, SDiskbasedBuf* pBuf, SExecTaskInfo* pTaskInfo,
                                SAggSupporter* pAggSup) {
  SResultRowInfo* pResultRowInfo = &binfo->resultRowInfo;
  SqlFunctionCtx* pCtx = binfo->pCtx;
982

983 984 985
  SResultRow* pResultRow =
      doSetResultOutBufByKey(pBuf, pResultRowInfo, (char*)pData, bytes, true, groupId, pTaskInfo, false, pAggSup);
  assert(pResultRow != NULL);
986

987
  setResultRowKey(pResultRow, pData, type);
988
  setResultRowInitCtx(pResultRow, pCtx, numOfCols, binfo->rowCellInfoOffset);
989
  return TSDB_CODE_SUCCESS;
990 991
}

992 993
static bool functionNeedToExecute(SqlFunctionCtx* pCtx) {
  struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(pCtx);
994

995 996 997 998 999
  // in case of timestamp column, always generated results.
  int32_t functionId = pCtx->functionId;
  if (functionId == -1) {
    return false;
  }
1000

1001 1002
  if (pCtx->scanFlag == REPEAT_SCAN) {
    return fmIsRepeatScanFunc(pCtx->functionId);
1003 1004
  }

1005 1006
  if (isRowEntryCompleted(pResInfo)) {
    return false;
1007 1008
  }

1009 1010 1011 1012 1013 1014 1015 1016
//  if (functionId == FUNCTION_FIRST_DST || functionId == FUNCTION_FIRST) {
//    //    return QUERY_IS_ASC_QUERY(pQueryAttr);
//  }
//
//  // denote the order type
//  if ((functionId == FUNCTION_LAST_DST || functionId == FUNCTION_LAST)) {
//    //    return pCtx->param[0].i == pQueryAttr->order.order;
//  }
1017

1018 1019 1020 1021 1022
  // in the reverse table scan, only the following functions need to be executed
  //  if (IS_REVERSE_SCAN(pRuntimeEnv) ||
  //      (pRuntimeEnv->scanFlag == REPEAT_SCAN && functionId != FUNCTION_STDDEV && functionId != FUNCTION_PERCT)) {
  //    return false;
  //  }
1023 1024 1025 1026

  return true;
}

1027 1028 1029 1030 1031 1032 1033
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;
    }
1034

1035 1036 1037
    // Set the correct column info (data type and bytes)
    pInput->pData[paramIndex]->info.type = type;
    pInput->pData[paramIndex]->info.bytes = tDataTypes[type].bytes;
1038
  }
H
Haojun Liao 已提交
1039

1040 1041 1042 1043 1044 1045
  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;
1046 1047
    }
  } else {
1048
    da = pInput->pColumnDataAgg[paramIndex];
1049 1050
  }

1051
  ASSERT(!IS_VAR_DATA_TYPE(type));
1052

1053 1054 1055 1056 1057 1058
  if (type == TSDB_DATA_TYPE_BIGINT) {
    int64_t v = pFuncParam->param.i;
    *da = (SColumnDataAgg){.numOfNull = 0, .min = v, .max = v, .maxIndex = 0, .minIndex = 0, .sum = v * numOfRows};
  } else if (type == TSDB_DATA_TYPE_DOUBLE) {
    double v = pFuncParam->param.d;
    *da = (SColumnDataAgg){.numOfNull = 0, .maxIndex = 0, .minIndex = 0};
1059

1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071
    *(double*)&da->min = v;
    *(double*)&da->max = v;
    *(double*)&da->sum = v * numOfRows;
  } else if (type == TSDB_DATA_TYPE_BOOL) {  // todo validate this data type
    bool v = pFuncParam->param.i;

    *da = (SColumnDataAgg){.numOfNull = 0, .maxIndex = 0, .minIndex = 0};
    *(bool*)&da->min = 0;
    *(bool*)&da->max = v;
    *(bool*)&da->sum = v * numOfRows;
  } else if (type == TSDB_DATA_TYPE_TIMESTAMP) {
    // do nothing
1072
  } else {
1073
    ASSERT(0);
1074 1075
  }

1076 1077
  return TSDB_CODE_SUCCESS;
}
1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088

void setBlockStatisInfo(SqlFunctionCtx* pCtx, SExprInfo* pExprInfo, SSDataBlock* pBlock) {
  int32_t numOfRows = pBlock->info.rows;

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

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

1089 1090
    for (int32_t j = 0; j < pExprInfo->base.numOfParams; ++j) {
      SFunctParam* pFuncParam = &pExprInfo->base.pParam[j];
1091

1092 1093
      if (pFuncParam->type == FUNC_PARAM_TYPE_COLUMN) {
        int32_t slotId = pFuncParam->pCol->slotId;
1094 1095 1096 1097
        pInput->pColumnDataAgg[j] = pBlock->pBlockAgg[slotId];
        if (pInput->pColumnDataAgg[j] == NULL) {
          pInput->colDataAggIsSet = false;
        }
1098 1099 1100 1101

        // 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);
1102 1103
      } else if (pFuncParam->type == FUNC_PARAM_TYPE_VALUE) {
        doCreateConstantValColumnAggInfo(pInput, pFuncParam, pFuncParam->param.nType, j, pBlock->info.rows);
1104 1105
      }
    }
1106
  } else {
1107
    pInput->colDataAggIsSet = false;
1108 1109 1110
  }

  // set the statistics data for primary time stamp column
1111 1112 1113 1114 1115
  //  if (pCtx->functionId == FUNCTION_SPREAD && pColumn->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
  //    pCtx->isAggSet = true;
  //    pCtx->agg.min = pBlock->info.window.skey;
  //    pCtx->agg.max = pBlock->info.window.ekey;
  //  }
1116 1117 1118
}

// set the output buffer for the selectivity + tag query
1119
static int32_t setSelectValueColumnInfo(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1120 1121
  int32_t num = 0;

H
Haojun Liao 已提交
1122
  SqlFunctionCtx*  p = NULL;
1123 1124
  SqlFunctionCtx** pValCtx = taosMemoryCalloc(numOfOutput, POINTER_BYTES);
  if (pValCtx == NULL) {
1125 1126 1127 1128
    return TSDB_CODE_QRY_OUT_OF_MEMORY;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
1129 1130
    if (strcmp(pCtx[i].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
      pValCtx[num++] = &pCtx[i];
1131
    } else if (fmIsAggFunc(pCtx[i].functionId)) {
1132
      p = &pCtx[i];
1133
    }
1134 1135 1136 1137 1138 1139 1140 1141 1142 1143
//    if (functionId == FUNCTION_TAG_DUMMY || functionId == FUNCTION_TS_DUMMY) {
//      tagLen += pCtx[i].resDataInfo.bytes;
//      pTagCtx[num++] = &pCtx[i];
//    } else if (functionId == FUNCTION_TS || functionId == FUNCTION_TAG) {
//      // tag function may be the group by tag column
//      // ts may be the required primary timestamp column
//      continue;
//    } else {
//      // the column may be the normal column, group by normal_column, the functionId is FUNCTION_PRJ
//    }
1144
  }
1145

1146
  if (p != NULL) {
1147
    p->subsidiaries.pCtx = pValCtx;
1148
    p->subsidiaries.num = num;
1149
  } else {
1150
    taosMemoryFreeClear(pValCtx);
1151 1152 1153 1154 1155
  }

  return TSDB_CODE_SUCCESS;
}

1156
SqlFunctionCtx* createSqlFunctionCtx(SExprInfo* pExprInfo, int32_t numOfOutput, int32_t** rowCellInfoOffset) {
L
Liu Jicong 已提交
1157
  SqlFunctionCtx* pFuncCtx = (SqlFunctionCtx*)taosMemoryCalloc(numOfOutput, sizeof(SqlFunctionCtx));
H
Haojun Liao 已提交
1158 1159 1160 1161
  if (pFuncCtx == NULL) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
1162
  *rowCellInfoOffset = taosMemoryCalloc(numOfOutput, sizeof(int32_t));
H
Haojun Liao 已提交
1163
  if (*rowCellInfoOffset == 0) {
wafwerar's avatar
wafwerar 已提交
1164
    taosMemoryFreeClear(pFuncCtx);
H
Haojun Liao 已提交
1165 1166 1167 1168
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
1169
    SExprInfo* pExpr = &pExprInfo[i];
H
Haojun Liao 已提交
1170

L
Liu Jicong 已提交
1171
    SExprBasicInfo* pFunct = &pExpr->base;
H
Haojun Liao 已提交
1172
    SqlFunctionCtx* pCtx = &pFuncCtx[i];
H
Haojun Liao 已提交
1173

1174
    pCtx->functionId = -1;
1175
    pCtx->curBufPage = -1;
L
Liu Jicong 已提交
1176
    pCtx->pExpr = pExpr;
1177

H
Haojun Liao 已提交
1178
    if (pExpr->pExpr->nodeType == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
1179
      SFuncExecEnv env = {0};
H
Haojun Liao 已提交
1180 1181
      pCtx->functionId = pExpr->pExpr->_function.pFunctNode->funcId;

H
Haojun Liao 已提交
1182
      if (fmIsAggFunc(pCtx->functionId) || fmIsNonstandardSQLFunc(pCtx->functionId)) {
1183 1184 1185 1186
        bool isUdaf = fmIsUserDefinedFunc(pCtx->functionId);
        if (!isUdaf) {
          fmGetFuncExecFuncs(pCtx->functionId, &pCtx->fpSet);
        } else {
1187
          char* udfName = pExpr->pExpr->_function.pFunctNode->functionName;
1188 1189 1190
          strncpy(pCtx->udfName, udfName, strlen(udfName));
          fmGetUdafExecFuncs(pCtx->functionId, &pCtx->fpSet);
        }
1191 1192 1193
        pCtx->fpSet.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
      } else {
        fmGetScalarFuncExecFuncs(pCtx->functionId, &pCtx->sfp);
1194 1195 1196
        if (pCtx->sfp.getEnv != NULL) {
          pCtx->sfp.getEnv(pExpr->pExpr->_function.pFunctNode, &env);
        }
1197
      }
H
Haojun Liao 已提交
1198
      pCtx->resDataInfo.interBufSize = env.calcMemSize;
1199 1200
    } else if (pExpr->pExpr->nodeType == QUERY_NODE_COLUMN || pExpr->pExpr->nodeType == QUERY_NODE_OPERATOR ||
               pExpr->pExpr->nodeType == QUERY_NODE_VALUE) {
H
Haojun Liao 已提交
1201
      // for simple column, the result buffer needs to hold at least one element.
1202
      pCtx->resDataInfo.interBufSize = pFunct->resSchema.bytes;
H
Haojun Liao 已提交
1203
    }
H
Haojun Liao 已提交
1204

H
Haojun Liao 已提交
1205
    pCtx->input.numOfInputCols = pFunct->numOfParams;
wafwerar's avatar
wafwerar 已提交
1206 1207
    pCtx->input.pData = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
    pCtx->input.pColumnDataAgg = taosMemoryCalloc(pFunct->numOfParams, POINTER_BYTES);
H
Haojun Liao 已提交
1208

1209
    pCtx->pTsOutput = NULL;
L
Liu Jicong 已提交
1210
    pCtx->resDataInfo.bytes = pFunct->resSchema.bytes;
1211
    pCtx->resDataInfo.type = pFunct->resSchema.type;
L
Liu Jicong 已提交
1212
    pCtx->order = TSDB_ORDER_ASC;
1213
    pCtx->start.key = INT64_MIN;
L
Liu Jicong 已提交
1214
    pCtx->end.key = INT64_MIN;
1215
    pCtx->numOfParams = pExpr->base.numOfParams;
H
Haojun Liao 已提交
1216

1217
    pCtx->param = pFunct->pParam;
dengyihao's avatar
dengyihao 已提交
1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
    //    for (int32_t j = 0; j < pCtx->numOfParams; ++j) {
    //      // set the order information for top/bottom query
    //      int32_t functionId = pCtx->functionId;
    //      if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF) {
    //        int32_t f = getExprFunctionId(&pExpr[0]);
    //        assert(f == FUNCTION_TS || f == FUNCTION_TS_DUMMY);
    //
    //        //      pCtx->param[2].i = pQueryAttr->order.order;
    //        //      pCtx->param[2].nType = TSDB_DATA_TYPE_BIGINT;
    //        //      pCtx->param[3].i = functionId;
    //        //      pCtx->param[3].nType = TSDB_DATA_TYPE_BIGINT;
    //
    //        //      pCtx->param[1].i = pQueryAttr->order.col.info.colId;
    //      } else if (functionId == FUNCTION_INTERP) {
    //        //      pCtx->param[2].i = (int8_t)pQueryAttr->fillType;
    //        //      if (pQueryAttr->fillVal != NULL) {
    //        //        if (isNull((const char *)&pQueryAttr->fillVal[i], pCtx->inputType)) {
    //        //          pCtx->param[1].nType = TSDB_DATA_TYPE_NULL;
    //        //        } else {  // todo refactor, taosVariantCreateFromBinary should handle the NULL value
    //        //          if (pCtx->inputType != TSDB_DATA_TYPE_BINARY && pCtx->inputType != TSDB_DATA_TYPE_NCHAR) {
    //        //            taosVariantCreateFromBinary(&pCtx->param[1], (char *)&pQueryAttr->fillVal[i],
    //        pCtx->inputBytes, pCtx->inputType);
    //        //          }
    //        //        }
    //        //      }
    //      } else if (functionId == FUNCTION_TWA) {
    //        //      pCtx->param[1].i = pQueryAttr->window.skey;
    //        //      pCtx->param[1].nType = TSDB_DATA_TYPE_BIGINT;
    //        //      pCtx->param[2].i = pQueryAttr->window.ekey;
    //        //      pCtx->param[2].nType = TSDB_DATA_TYPE_BIGINT;
    //      } else if (functionId == FUNCTION_ARITHM) {
    //        //      pCtx->param[1].pz = (char*) getScalarFuncSupport(pRuntimeEnv->scalarSup, i);
    //      }
    //    }
H
Haojun Liao 已提交
1252 1253
  }

L
Liu Jicong 已提交
1254 1255 1256
  for (int32_t i = 1; i < numOfOutput; ++i) {
    (*rowCellInfoOffset)[i] =
        (int32_t)((*rowCellInfoOffset)[i - 1] + sizeof(SResultRowEntryInfo) + pFuncCtx[i - 1].resDataInfo.interBufSize);
H
Haojun Liao 已提交
1257
  }
H
Haojun Liao 已提交
1258

1259
  setSelectValueColumnInfo(pFuncCtx, numOfOutput);
H
Haojun Liao 已提交
1260 1261 1262
  return pFuncCtx;
}

1263
static void* destroySqlFunctionCtx(SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1264 1265 1266 1267 1268 1269
  if (pCtx == NULL) {
    return NULL;
  }

  for (int32_t i = 0; i < numOfOutput; ++i) {
    for (int32_t j = 0; j < pCtx[i].numOfParams; ++j) {
1270
      taosVariantDestroy(&pCtx[i].param[j].param);
1271 1272 1273
    }

    taosVariantDestroy(&pCtx[i].tag);
1274
    taosMemoryFreeClear(pCtx[i].subsidiaries.pCtx);
H
Haojun Liao 已提交
1275 1276
    taosMemoryFree(pCtx[i].input.pData);
    taosMemoryFree(pCtx[i].input.pColumnDataAgg);
1277 1278
  }

wafwerar's avatar
wafwerar 已提交
1279
  taosMemoryFreeClear(pCtx);
1280 1281 1282
  return NULL;
}

L
Liu Jicong 已提交
1283
bool isTaskKilled(SExecTaskInfo* pTaskInfo) {
1284 1285
  // query has been executed more than tsShellActivityTimer, and the retrieve has not arrived
  // abort current query execution.
L
Liu Jicong 已提交
1286 1287
  if (pTaskInfo->owner != 0 &&
      ((taosGetTimestampSec() - pTaskInfo->cost.start / 1000) > 10 * getMaximumIdleDurationSec())
1288 1289
      /*(!needBuildResAfterQueryComplete(pTaskInfo))*/) {
    assert(pTaskInfo->cost.start != 0);
L
Liu Jicong 已提交
1290 1291 1292
    //    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;
1293 1294 1295 1296 1297
  }

  return false;
}

L
Liu Jicong 已提交
1298
void setTaskKilled(SExecTaskInfo* pTaskInfo) { pTaskInfo->code = TSDB_CODE_TSC_QUERY_CANCELLED; }
1299

L
Liu Jicong 已提交
1300
static bool isCachedLastQuery(STaskAttr* pQueryAttr) {
1301 1302 1303 1304 1305 1306 1307 1308 1309
  for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
    int32_t functionId = getExprFunctionId(&pQueryAttr->pExpr1[i]);
    if (functionId == FUNCTION_LAST || functionId == FUNCTION_LAST_DST) {
      continue;
    }

    return false;
  }

1310 1311
  int32_t order = TSDB_ORDER_ASC;
  if (order != TSDB_ORDER_DESC || !TSWINDOW_IS_EQUAL(pQueryAttr->window, TSWINDOW_DESC_INITIALIZER)) {
1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
    return false;
  }

  if (pQueryAttr->groupbyColumn) {
    return false;
  }

  if (pQueryAttr->interval.interval > 0) {
    return false;
  }

  if (pQueryAttr->numOfFilterCols > 0 || pQueryAttr->havingNum > 0) {
    return false;
  }

  return true;
}

/////////////////////////////////////////////////////////////////////////////////////////////
L
Liu Jicong 已提交
1331
// todo refactor : return window
1332
void getAlignQueryTimeWindow(SInterval* pInterval, int32_t precision, int64_t key, STimeWindow* win) {
H
Haojun Liao 已提交
1333
  win->skey = taosTimeTruncate(key, pInterval, precision);
1334 1335

  /*
H
Haojun Liao 已提交
1336
   * if the realSkey > INT64_MAX - pInterval->interval, the query duration between
1337 1338
   * realSkey and realEkey must be less than one interval.Therefore, no need to adjust the query ranges.
   */
1339 1340
  win->ekey = taosTimeAdd(win->skey, pInterval->interval, pInterval->intervalUnit, precision) - 1;
  if (win->ekey < win->skey) {
1341 1342 1343 1344
    win->ekey = INT64_MAX;
  }
}

L
Liu Jicong 已提交
1345
static int32_t updateBlockLoadStatus(STaskAttr* pQuery, int32_t status) {
1346 1347 1348
  bool hasFirstLastFunc = false;
  bool hasOtherFunc = false;

1349
  if (status == BLK_DATA_DATA_LOAD || status == BLK_DATA_FILTEROUT) {
1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367
    return status;
  }

  for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
    int32_t functionId = getExprFunctionId(&pQuery->pExpr1[i]);

    if (functionId == FUNCTION_TS || functionId == FUNCTION_TS_DUMMY || functionId == FUNCTION_TAG ||
        functionId == FUNCTION_TAG_DUMMY) {
      continue;
    }

    if (functionId == FUNCTION_FIRST_DST || functionId == FUNCTION_LAST_DST) {
      hasFirstLastFunc = true;
    } else {
      hasOtherFunc = true;
    }
  }

1368
  if (hasFirstLastFunc && status == BLK_DATA_NOT_LOAD) {
L
Liu Jicong 已提交
1369
    if (!hasOtherFunc) {
1370
      return BLK_DATA_FILTEROUT;
1371
    } else {
1372
      return BLK_DATA_DATA_LOAD;
1373 1374 1375 1376 1377 1378
    }
  }

  return status;
}

L
Liu Jicong 已提交
1379 1380
// static void updateDataCheckOrder(SQInfo *pQInfo, SQueryTableReq* pQueryMsg, bool stableQuery) {
//   STaskAttr* pQueryAttr = pQInfo->runtimeEnv.pQueryAttr;
H
Haojun Liao 已提交
1381
//
L
Liu Jicong 已提交
1382 1383 1384 1385
//   // in case of point-interpolation query, use asc order scan
//   char msg[] = "QInfo:0x%"PRIx64" scan order changed for %s query, old:%d, new:%d, qrange exchanged, old qrange:%"
//   PRId64
//                "-%" PRId64 ", new qrange:%" PRId64 "-%" PRId64;
H
Haojun Liao 已提交
1386
//
L
Liu Jicong 已提交
1387 1388 1389 1390 1391
//   // todo handle the case the the order irrelevant query type mixed up with order critical query type
//   // descending order query for last_row query
//   if (isFirstLastRowQuery(pQueryAttr)) {
//     //qDebug("QInfo:0x%"PRIx64" scan order changed for last_row query, old:%d, new:%d", pQInfo->qId,
//     pQueryAttr->order.order, TSDB_ORDER_ASC);
H
Haojun Liao 已提交
1392
//
L
Liu Jicong 已提交
1393 1394
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     if (pQueryAttr->window.skey > pQueryAttr->window.ekey) {
wafwerar's avatar
wafwerar 已提交
1395
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1396
//     }
H
Haojun Liao 已提交
1397
//
L
Liu Jicong 已提交
1398 1399 1400
//     pQueryAttr->needReverseScan = false;
//     return;
//   }
H
Haojun Liao 已提交
1401
//
L
Liu Jicong 已提交
1402 1403 1404
//   if (pQueryAttr->groupbyColumn && pQueryAttr->order.order == TSDB_ORDER_DESC) {
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     if (pQueryAttr->window.skey > pQueryAttr->window.ekey) {
wafwerar's avatar
wafwerar 已提交
1405
//       TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
L
Liu Jicong 已提交
1406
//     }
H
Haojun Liao 已提交
1407
//
L
Liu Jicong 已提交
1408 1409 1410 1411
//     pQueryAttr->needReverseScan = false;
//     doUpdateLastKey(pQueryAttr);
//     return;
//   }
H
Haojun Liao 已提交
1412
//
L
Liu Jicong 已提交
1413 1414 1415 1416 1417 1418
//   if (pQueryAttr->pointInterpQuery && pQueryAttr->interval.interval == 0) {
//     if (!QUERY_IS_ASC_QUERY(pQueryAttr)) {
//       //qDebug(msg, pQInfo->qId, "interp", pQueryAttr->order.order, TSDB_ORDER_ASC, pQueryAttr->window.skey,
//       pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey); TSWAP(pQueryAttr->window.skey,
//       pQueryAttr->window.ekey, TSKEY);
//     }
H
Haojun Liao 已提交
1419
//
L
Liu Jicong 已提交
1420 1421 1422
//     pQueryAttr->order.order = TSDB_ORDER_ASC;
//     return;
//   }
H
Haojun Liao 已提交
1423
//
L
Liu Jicong 已提交
1424 1425 1426 1427
//   if (pQueryAttr->interval.interval == 0) {
//     if (onlyFirstQuery(pQueryAttr)) {
//       if (!QUERY_IS_ASC_QUERY(pQueryAttr)) {
//         //qDebug(msg, pQInfo->qId, "only-first", pQueryAttr->order.order, TSDB_ORDER_ASC, pQueryAttr->window.skey,
H
Haojun Liao 已提交
1428 1429
////               pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey);
//
wafwerar's avatar
wafwerar 已提交
1430
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1431 1432 1433 1434 1435 1436 1437 1438 1439 1440
//        doUpdateLastKey(pQueryAttr);
//      }
//
//      pQueryAttr->order.order = TSDB_ORDER_ASC;
//      pQueryAttr->needReverseScan = false;
//    } else if (onlyLastQuery(pQueryAttr) && notContainSessionOrStateWindow(pQueryAttr)) {
//      if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//        //qDebug(msg, pQInfo->qId, "only-last", pQueryAttr->order.order, TSDB_ORDER_DESC, pQueryAttr->window.skey,
////               pQueryAttr->window.ekey, pQueryAttr->window.ekey, pQueryAttr->window.skey);
//
wafwerar's avatar
wafwerar 已提交
1441
//        TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453
//        doUpdateLastKey(pQueryAttr);
//      }
//
//      pQueryAttr->order.order = TSDB_ORDER_DESC;
//      pQueryAttr->needReverseScan = false;
//    }
//
//  } else {  // interval query
//    if (stableQuery) {
//      if (onlyFirstQuery(pQueryAttr)) {
//        if (!QUERY_IS_ASC_QUERY(pQueryAttr)) {
//          //qDebug(msg, pQInfo->qId, "only-first stable", pQueryAttr->order.order, TSDB_ORDER_ASC,
L
Liu Jicong 已提交
1454 1455
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1456
//
wafwerar's avatar
wafwerar 已提交
1457
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1458 1459 1460 1461 1462 1463 1464 1465
//          doUpdateLastKey(pQueryAttr);
//        }
//
//        pQueryAttr->order.order = TSDB_ORDER_ASC;
//        pQueryAttr->needReverseScan = false;
//      } else if (onlyLastQuery(pQueryAttr)) {
//        if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//          //qDebug(msg, pQInfo->qId, "only-last stable", pQueryAttr->order.order, TSDB_ORDER_DESC,
L
Liu Jicong 已提交
1466 1467
////                 pQueryAttr->window.skey, pQueryAttr->window.ekey, pQueryAttr->window.ekey,
/// pQueryAttr->window.skey);
H
Haojun Liao 已提交
1468
//
wafwerar's avatar
wafwerar 已提交
1469
//          TSWAP(pQueryAttr->window.skey, pQueryAttr->window.ekey);
H
Haojun Liao 已提交
1470 1471 1472 1473 1474 1475 1476 1477 1478
//          doUpdateLastKey(pQueryAttr);
//        }
//
//        pQueryAttr->order.order = TSDB_ORDER_DESC;
//        pQueryAttr->needReverseScan = false;
//      }
//    }
//  }
//}
1479

H
Haojun Liao 已提交
1480 1481
static void getIntermediateBufInfo(STaskRuntimeEnv* pRuntimeEnv, int32_t* ps, int32_t* rowsize) {
  STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
L
Liu Jicong 已提交
1482
  int32_t    MIN_ROWS_PER_PAGE = 4;
1483

L
Liu Jicong 已提交
1484 1485
  *rowsize = (int32_t)(pQueryAttr->resultRowSize *
                       getRowNumForMultioutput(pQueryAttr, pQueryAttr->topBotQuery, pQueryAttr->stableQuery));
1486 1487 1488 1489
  int32_t overhead = sizeof(SFilePage);

  // one page contains at least two rows
  *ps = DEFAULT_INTERN_BUF_PAGE_SIZE;
L
Liu Jicong 已提交
1490
  while (((*rowsize) * MIN_ROWS_PER_PAGE) > (*ps) - overhead) {
1491 1492 1493 1494
    *ps = ((*ps) << 1u);
  }
}

L
Liu Jicong 已提交
1495 1496 1497
// static FORCE_INLINE bool doFilterByBlockStatistics(STaskRuntimeEnv* pRuntimeEnv, SDataStatis *pDataStatis,
// SqlFunctionCtx *pCtx, int32_t numOfRows) {
//   STaskAttr* pQueryAttr = pRuntimeEnv->pQueryAttr;
1498
//
L
Liu Jicong 已提交
1499 1500 1501
//   if (pDataStatis == NULL || pQueryAttr->pFilters == NULL) {
//     return true;
//   }
1502
//
L
Liu Jicong 已提交
1503 1504
//   return filterRangeExecute(pQueryAttr->pFilters, pDataStatis, pQueryAttr->numOfCols, numOfRows);
// }
1505

H
Haojun Liao 已提交
1506
static bool overlapWithTimeWindow(STaskAttr* pQueryAttr, SDataBlockInfo* pBlockInfo) {
1507 1508
  STimeWindow w = {0};

dengyihao's avatar
dengyihao 已提交
1509 1510
  TSKEY sk = TMIN(pQueryAttr->window.skey, pQueryAttr->window.ekey);
  TSKEY ek = TMAX(pQueryAttr->window.skey, pQueryAttr->window.ekey);
1511

1512
  if (true) {
L
Liu Jicong 已提交
1513
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.skey, sk, ek, &w);
1514 1515 1516 1517 1518 1519
    assert(w.ekey >= pBlockInfo->window.skey);

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

L
Liu Jicong 已提交
1520 1521
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1522 1523 1524 1525 1526 1527 1528 1529 1530 1531
      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
1532
    //    getAlignQueryTimeWindow(pQueryAttr, pBlockInfo->window.ekey, sk, ek, &w);
1533 1534 1535 1536 1537 1538
    assert(w.skey <= pBlockInfo->window.ekey);

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

L
Liu Jicong 已提交
1539 1540
    while (1) {
      //      getNextTimeWindow(pQueryAttr, &w);
1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604
      if (w.ekey < pBlockInfo->window.skey) {
        break;
      }

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

  return false;
}

void doCompactSDataBlock(SSDataBlock* pBlock, int32_t numOfRows, int8_t* p) {
  int32_t len = 0;
  int32_t start = 0;
  for (int32_t j = 0; j < numOfRows; ++j) {
    if (p[j] == 1) {
      len++;
    } else {
      if (len > 0) {
        int32_t cstart = j - len;
        for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
          SColumnInfoData* pColumnInfoData = taosArrayGet(pBlock->pDataBlock, i);

          int16_t bytes = pColumnInfoData->info.bytes;
          memmove(((char*)pColumnInfoData->pData) + start * bytes, pColumnInfoData->pData + cstart * bytes,
                  len * bytes);
        }

        start += len;
        len = 0;
      }
    }
  }

  if (len > 0) {
    int32_t cstart = numOfRows - len;
    for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
      SColumnInfoData* pColumnInfoData = taosArrayGet(pBlock->pDataBlock, i);

      int16_t bytes = pColumnInfoData->info.bytes;
      memmove(pColumnInfoData->pData + start * bytes, pColumnInfoData->pData + cstart * bytes, len * bytes);
    }

    start += len;
    len = 0;
  }

  pBlock->info.rows = start;
  pBlock->pBlockAgg = NULL;  // clean the block statistics info

  if (start > 0) {
    SColumnInfoData* pColumnInfoData = taosArrayGet(pBlock->pDataBlock, 0);
    if (pColumnInfoData->info.type == TSDB_DATA_TYPE_TIMESTAMP &&
        pColumnInfoData->info.colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
      pBlock->info.window.skey = *(int64_t*)pColumnInfoData->pData;
      pBlock->info.window.ekey = *(int64_t*)(pColumnInfoData->pData + TSDB_KEYSIZE * (start - 1));
    }
  }
}

static SColumnInfo* doGetTagColumnInfoById(SColumnInfo* pTagColList, int32_t numOfTags, int16_t colId);
L
Liu Jicong 已提交
1605
static void         doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes);
1606 1607

static uint32_t doFilterByBlockTimeWindow(STableScanInfo* pTableScanInfo, SSDataBlock* pBlock) {
H
Haojun Liao 已提交
1608
  SqlFunctionCtx* pCtx = pTableScanInfo->pCtx;
1609
  uint32_t        status = BLK_DATA_NOT_LOAD;
1610 1611 1612 1613

  int32_t numOfOutput = pTableScanInfo->numOfOutput;
  for (int32_t i = 0; i < numOfOutput; ++i) {
    int32_t functionId = pCtx[i].functionId;
H
Haojun Liao 已提交
1614
    int32_t colId = pTableScanInfo->pExpr[i].base.pParam[0].pCol->colId;
1615 1616 1617

    // group by + first/last should not apply the first/last block filter
    if (functionId < 0) {
1618
      status |= BLK_DATA_DATA_LOAD;
1619 1620
      return status;
    } else {
L
Liu Jicong 已提交
1621
      //      status |= aAggs[functionId].dataReqFunc(&pTableScanInfo->pCtx[i], &pBlock->info.window, colId);
1622
      //      if ((status & BLK_DATA_DATA_LOAD) == BLK_DATA_DATA_LOAD) {
L
Liu Jicong 已提交
1623 1624
      //        return status;
      //      }
1625 1626 1627 1628 1629 1630
    }
  }

  return status;
}

L
Liu Jicong 已提交
1631 1632
int32_t loadDataBlockOnDemand(SExecTaskInfo* pTaskInfo, STableScanInfo* pTableScanInfo, SSDataBlock* pBlock,
                              uint32_t* status) {
1633
  *status = BLK_DATA_NOT_LOAD;
1634

H
Haojun Liao 已提交
1635
  pBlock->pDataBlock = NULL;
L
Liu Jicong 已提交
1636
  pBlock->pBlockAgg = NULL;
H
Haojun Liao 已提交
1637

L
Liu Jicong 已提交
1638 1639
  //  int64_t groupId = pRuntimeEnv->current->groupIndex;
  //  bool    ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
1640

H
Haojun Liao 已提交
1641
  STaskCostInfo* pCost = &pTaskInfo->cost;
1642

1643 1644
//  pCost->totalBlocks += 1;
//  pCost->totalRows += pBlock->info.rows;
H
Haojun Liao 已提交
1645
#if 0
1646 1647 1648
  // 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
1649
      (QUERY_IS_INTERVAL_QUERY(pQueryAttr) && overlapWithTimeWindow(pTaskInfo, &pBlock->info))) {
1650
    (*status) = BLK_DATA_DATA_LOAD;
1651 1652 1653
  }

  // check if this data block is required to load
1654
  if ((*status) != BLK_DATA_DATA_LOAD) {
1655 1656 1657 1658 1659 1660 1661
    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
1662
      bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685
      TSKEY k = ascQuery? pBlock->info.window.skey : pBlock->info.window.ekey;

      STimeWindow win = getActiveTimeWindow(pTableScanInfo->pResultRowInfo, k, pQueryAttr);
      if (pQueryAttr->pointInterpQuery) {
        needFilter = chkWindowOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, &win, masterScan, &pResult, groupId,
                                    pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
                                    pTableScanInfo->rowCellInfoOffset);
      } else {
        if (setResultOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pBlock->info.uid, &win, masterScan, &pResult, groupId,
                                    pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
                                    pTableScanInfo->rowCellInfoOffset) != TSDB_CODE_SUCCESS) {
          longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
        }
      }
    } else if (pQueryAttr->stableQuery && (!pQueryAttr->tsCompQuery) && (!pQueryAttr->diffQuery)) { // stable aggregate, not interval aggregate or normal column aggregate
      doSetTableGroupOutputBuf(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pTableScanInfo->pCtx,
                               pTableScanInfo->rowCellInfoOffset, pTableScanInfo->numOfOutput,
                               pRuntimeEnv->current->groupIndex);
    }

    if (needFilter) {
      (*status) = doFilterByBlockTimeWindow(pTableScanInfo, pBlock);
    } else {
1686
      (*status) = BLK_DATA_DATA_LOAD;
1687 1688 1689 1690
    }
  }

  SDataBlockInfo* pBlockInfo = &pBlock->info;
H
Haojun Liao 已提交
1691
//  *status = updateBlockLoadStatus(pRuntimeEnv->pQueryAttr, *status);
1692

1693
  if ((*status) == BLK_DATA_NOT_LOAD || (*status) == BLK_DATA_FILTEROUT) {
1694 1695
    //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//           pBlockInfo->window.ekey, pBlockInfo->rows);
1696
    pCost->skipBlocks += 1;
1697
  } else if ((*status) == BLK_DATA_SMA_LOAD) {
1698 1699
    // this function never returns error?
    pCost->loadBlockStatis += 1;
1700
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1701 1702

    if (pBlock->pBlockAgg == NULL) {  // data block statistics does not exist, load data block
1703
//      pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1704 1705 1706
      pCost->totalCheckedRows += pBlock->info.rows;
    }
  } else {
1707
    assert((*status) == BLK_DATA_DATA_LOAD);
1708 1709 1710

    // load the data block statistics to perform further filter
    pCost->loadBlockStatis += 1;
1711
//    tsdbRetrieveDataBlockStatisInfo(pTableScanInfo->pTsdbReadHandle, &pBlock->pBlockAgg);
1712 1713 1714 1715 1716 1717

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

H
Haojun Liao 已提交
1718
          bool  masterScan = IS_MAIN_SCAN(pRuntimeEnv);
1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735
          TSKEY k = ascQuery? pBlock->info.window.skey : pBlock->info.window.ekey;

          STimeWindow win = getActiveTimeWindow(pTableScanInfo->pResultRowInfo, k, pQueryAttr);
          if (setResultOutputBufByKey(pRuntimeEnv, pTableScanInfo->pResultRowInfo, pBlock->info.uid, &win, masterScan, &pResult, groupId,
                                      pTableScanInfo->pCtx, pTableScanInfo->numOfOutput,
                                      pTableScanInfo->rowCellInfoOffset) != TSDB_CODE_SUCCESS) {
            longjmp(pRuntimeEnv->env, TSDB_CODE_QRY_OUT_OF_MEMORY);
          }
        }
      }
      bool load = false;
      for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
        int32_t functionId = pTableScanInfo->pCtx[i].functionId;
        if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM) {
//          load = topbot_datablock_filter(&pTableScanInfo->pCtx[i], (char*)&(pBlock->pBlockAgg[i].min),
//                                         (char*)&(pBlock->pBlockAgg[i].max));
          if (!load) { // current block has been discard due to filter applied
1736
            pCost->skipBlocks += 1;
1737 1738
            //qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId,
//                   pBlockInfo->window.skey, pBlockInfo->window.ekey, pBlockInfo->rows);
1739
            (*status) = BLK_DATA_FILTEROUT;
1740 1741 1742 1743 1744 1745 1746 1747
            return TSDB_CODE_SUCCESS;
          }
        }
      }
    }

    // current block has been discard due to filter applied
//    if (!doFilterByBlockStatistics(pRuntimeEnv, pBlock->pBlockAgg, pTableScanInfo->pCtx, pBlockInfo->rows)) {
1748
//      pCost->skipBlocks += 1;
1749 1750
//      qDebug("QInfo:0x%"PRIx64" data block discard, brange:%" PRId64 "-%" PRId64 ", rows:%d", pQInfo->qId, pBlockInfo->window.skey,
//             pBlockInfo->window.ekey, pBlockInfo->rows);
1751
//      (*status) = BLK_DATA_FILTEROUT;
1752 1753 1754 1755 1756
//      return TSDB_CODE_SUCCESS;
//    }

    pCost->totalCheckedRows += pBlockInfo->rows;
    pCost->loadBlocks += 1;
1757
//    pBlock->pDataBlock = tsdbRetrieveDataBlock(pTableScanInfo->pTsdbReadHandle, NULL);
1758 1759 1760 1761 1762 1763 1764
//    if (pBlock->pDataBlock == NULL) {
//      return terrno;
//    }

//    if (pQueryAttr->pFilters != NULL) {
//      filterSetColFieldData(pQueryAttr->pFilters, pBlock->info.numOfCols, pBlock->pDataBlock);
//    }
1765

1766 1767 1768 1769
//    if (pQueryAttr->pFilters != NULL || pRuntimeEnv->pTsBuf != NULL) {
//      filterColRowsInDataBlock(pRuntimeEnv, pBlock, ascQuery);
//    }
  }
H
Haojun Liao 已提交
1770
#endif
1771 1772 1773 1774
  return TSDB_CODE_SUCCESS;
}

/*
H
Haojun Liao 已提交
1775
 * set tag value in SqlFunctionCtx
1776 1777
 * e.g.,tag information into input buffer
 */
L
Liu Jicong 已提交
1778
static void doSetTagValueInParam(void* pTable, int32_t tagColId, SVariant* tag, int16_t type, int16_t bytes) {
1779 1780 1781
  taosVariantDestroy(tag);

  char* val = NULL;
L
Liu Jicong 已提交
1782 1783 1784 1785 1786 1787
  //  if (tagColId == TSDB_TBNAME_COLUMN_INDEX) {
  //    val = tsdbGetTableName(pTable);
  //    assert(val != NULL);
  //  } else {
  //    val = tsdbGetTableTagVal(pTable, tagColId, type, bytes);
  //  }
1788 1789 1790 1791 1792 1793 1794 1795

  if (val == NULL || isNull(val, type)) {
    tag->nType = TSDB_DATA_TYPE_NULL;
    return;
  }

  if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR) {
    int32_t maxLen = bytes - VARSTR_HEADER_SIZE;
L
Liu Jicong 已提交
1796
    int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
1797
    taosVariantCreateFromBinary(tag, varDataVal(val), len, type);
L
Liu Jicong 已提交
1798
    // taosVariantCreateFromBinary(tag, varDataVal(val), varDataLen(val), type);
1799 1800 1801 1802 1803 1804 1805 1806
  } else {
    taosVariantCreateFromBinary(tag, val, bytes, type);
  }
}

static SColumnInfo* doGetTagColumnInfoById(SColumnInfo* pTagColList, int32_t numOfTags, int16_t colId) {
  assert(pTagColList != NULL && numOfTags > 0);

L
Liu Jicong 已提交
1807
  for (int32_t i = 0; i < numOfTags; ++i) {
1808 1809 1810 1811 1812 1813 1814 1815
    if (pTagColList[i].colId == colId) {
      return &pTagColList[i];
    }
  }

  return NULL;
}

L
Liu Jicong 已提交
1816 1817
void setTagValue(SOperatorInfo* pOperatorInfo, void* pTable, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
  SExprInfo* pExpr = pOperatorInfo->pExpr;
1818
  SExprInfo* pExprInfo = &pExpr[0];
L
Liu Jicong 已提交
1819
  int32_t    functionId = getExprFunctionId(pExprInfo);
1820
#if 0
1821 1822 1823
  if (pQueryAttr->numOfOutput == 1 && functionId == FUNCTION_TS_COMP && pQueryAttr->stableQuery) {
    assert(pExprInfo->base.numOfParams == 1);

L
Liu Jicong 已提交
1824
    //    int16_t      tagColId = (int16_t)pExprInfo->base.param[0].i;
H
Haojun Liao 已提交
1825
    int16_t      tagColId = -1;
1826 1827 1828
    SColumnInfo* pColInfo = doGetTagColumnInfoById(pQueryAttr->tagColList, pQueryAttr->numOfTags, tagColId);

    doSetTagValueInParam(pTable, tagColId, &pCtx[0].tag, pColInfo->type, pColInfo->bytes);
H
Haojun Liao 已提交
1829

1830 1831 1832 1833 1834 1835 1836 1837 1838
  } else {
    // set tag value, by which the results are aggregated.
    int32_t offset = 0;
    memset(pRuntimeEnv->tagVal, 0, pQueryAttr->tagLen);

    for (int32_t idx = 0; idx < numOfOutput; ++idx) {
      SExprInfo* pLocalExprInfo = &pExpr[idx];

      // ts_comp column required the tag value for join filter
H
Haojun Liao 已提交
1839
      if (!TSDB_COL_IS_TAG(pLocalExprInfo->base.pParam[0].pCol->flag)) {
1840 1841 1842 1843
        continue;
      }

      // todo use tag column index to optimize performance
L
Liu Jicong 已提交
1844 1845
      doSetTagValueInParam(pTable, pLocalExprInfo->base.pParam[0].pCol->colId, &pCtx[idx].tag,
                           pLocalExprInfo->base.resSchema.type, pLocalExprInfo->base.resSchema.bytes);
1846

L
Liu Jicong 已提交
1847 1848 1849
      if (IS_NUMERIC_TYPE(pLocalExprInfo->base.resSchema.type) ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_BOOL ||
          pLocalExprInfo->base.resSchema.type == TSDB_DATA_TYPE_TIMESTAMP) {
1850 1851 1852 1853
        memcpy(pRuntimeEnv->tagVal + offset, &pCtx[idx].tag.i, pLocalExprInfo->base.resSchema.bytes);
      } else {
        if (pCtx[idx].tag.pz != NULL) {
          memcpy(pRuntimeEnv->tagVal + offset, pCtx[idx].tag.pz, pCtx[idx].tag.nLen);
L
Liu Jicong 已提交
1854
        }
1855 1856 1857 1858 1859 1860 1861 1862 1863 1864
      }

      offset += pLocalExprInfo->base.resSchema.bytes;
    }
  }

  // set the tsBuf start position before check each data block
  if (pRuntimeEnv->pTsBuf != NULL) {
    setCtxTagForJoin(pRuntimeEnv, &pCtx[0], pExprInfo, pTable);
  }
1865
#endif
1866 1867
}

H
Haojun Liao 已提交
1868
void copyToSDataBlock(SSDataBlock* pBlock, int32_t* offset, SGroupResInfo* pGroupResInfo, SDiskbasedBuf* pResBuf) {
1869 1870 1871 1872 1873 1874 1875
  pBlock->info.rows = 0;

  int32_t code = TSDB_CODE_SUCCESS;
  while (pGroupResInfo->currentGroup < pGroupResInfo->totalGroup) {
    // all results in current group have been returned to client, try next group
    if ((pGroupResInfo->pRows == NULL) || taosArrayGetSize(pGroupResInfo->pRows) == 0) {
      assert(pGroupResInfo->index == 0);
L
Liu Jicong 已提交
1876 1877 1878
      //      if ((code = mergeIntoGroupResult(&pGroupResInfo, pRuntimeEnv, offset)) != TSDB_CODE_SUCCESS) {
      return;
      //      }
1879 1880
    }

L
Liu Jicong 已提交
1881
    //    doCopyToSDataBlock(pResBuf, pGroupResInfo, TSDB_ORDER_ASC, pBlock, );
1882 1883 1884 1885 1886 1887 1888 1889 1890 1891

    // current data are all dumped to result buffer, clear it
    if (!hasRemainDataInCurrentGroup(pGroupResInfo)) {
      cleanupGroupResInfo(pGroupResInfo);
      if (!incNextGroup(pGroupResInfo)) {
        break;
      }
    }

    // enough results in data buffer, return
L
Liu Jicong 已提交
1892 1893 1894
    //    if (pBlock->info.rows >= threshold) {
    //      break;
    //    }
1895 1896 1897
  }
}

L
Liu Jicong 已提交
1898
static void updateTableQueryInfoForReverseScan(STableQueryInfo* pTableQueryInfo) {
1899 1900 1901 1902
  if (pTableQueryInfo == NULL) {
    return;
  }

wafwerar's avatar
wafwerar 已提交
1903
  //  TSWAP(pTableQueryInfo->win.skey, pTableQueryInfo->win.ekey);
L
Liu Jicong 已提交
1904
  //  pTableQueryInfo->lastKey = pTableQueryInfo->win.skey;
1905

L
Liu Jicong 已提交
1906 1907
  //  SWITCH_ORDER(pTableQueryInfo->cur.order);
  //  pTableQueryInfo->cur.vgroupIndex = -1;
1908 1909

  // set the index to be the end slot of result rows array
dengyihao's avatar
dengyihao 已提交
1910 1911 1912 1913 1914 1915
  //  SResultRowInfo* pResultRowInfo = &pTableQueryInfo->resInfo;
  //  if (pResultRowInfo->size > 0) {
  //    pResultRowInfo->curPos = pResultRowInfo->size - 1;
  //  } else {
  //    pResultRowInfo->curPos = -1;
  //  }
1916 1917
}

H
Haojun Liao 已提交
1918
void initResultRow(SResultRow* pResultRow) {
X
Xiaoyu Wang 已提交
1919
  //  pResultRow->pEntryInfo = (struct SResultRowEntryInfo*)((char*)pResultRow + sizeof(SResultRow));
1920 1921 1922 1923 1924
}

/*
 * The start of each column SResultRowEntryInfo is denote by RowCellInfoOffset.
 * Note that in case of top/bottom query, the whole multiple rows of result is treated as only one row of results.
H
Haojun Liao 已提交
1925 1926 1927
 * +------------+-----------------result column 1------------+------------------result column 2-----------+
 * | SResultRow | SResultRowEntryInfo | intermediate buffer1 | SResultRowEntryInfo | intermediate buffer 2|
 * +------------+--------------------------------------------+--------------------------------------------+
1928 1929
 *           offset[0]                                  offset[1]                                   offset[2]
 */
1930
// TODO refactor: some function move away
X
Xiaoyu Wang 已提交
1931 1932
void setFunctionResultOutput(SOptrBasicInfo* pInfo, SAggSupporter* pSup, int32_t stage, int32_t numOfExprs,
                             SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
1933 1934 1935
  SqlFunctionCtx* pCtx = pInfo->pCtx;
  SSDataBlock*    pDataBlock = pInfo->pRes;
  int32_t*        rowCellInfoOffset = pInfo->rowCellInfoOffset;
H
Haojun Liao 已提交
1936

H
Haojun Liao 已提交
1937
  SResultRowInfo* pResultRowInfo = &pInfo->resultRowInfo;
H
Haojun Liao 已提交
1938
  initResultRowInfo(pResultRowInfo, 16);
H
Haojun Liao 已提交
1939

L
Liu Jicong 已提交
1940 1941
  int64_t     tid = 0;
  int64_t     groupId = 0;
1942 1943
  SResultRow* pRow = doSetResultOutBufByKey(pSup->pResultBuf, pResultRowInfo, (char*)&tid, sizeof(tid), true, groupId,
                                            pTaskInfo, false, pSup);
H
Haojun Liao 已提交
1944

H
Haojun Liao 已提交
1945
  ASSERT(pDataBlock->info.numOfCols == numOfExprs);
H
Haojun Liao 已提交
1946 1947 1948 1949
  for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
    struct SResultRowEntryInfo* pEntry = getResultCell(pRow, i, rowCellInfoOffset);
    cleanupResultRowEntry(pEntry);

L
Liu Jicong 已提交
1950
    pCtx[i].resultInfo = pEntry;
1951
    pCtx[i].scanFlag = stage;
H
Haojun Liao 已提交
1952 1953

    // set the timestamp output buffer for top/bottom/diff query
L
Liu Jicong 已提交
1954 1955
    //    int32_t fid = pCtx[i].functionId;
    //    if (fid == FUNCTION_TOP || fid == FUNCTION_BOTTOM || fid == FUNCTION_DIFF || fid == FUNCTION_DERIVATIVE) {
H
Haojun Liao 已提交
1956
    //      if (i > 0) pCtx[i].pTsOutput = pCtx[i-1].pOutput;
L
Liu Jicong 已提交
1957
    //    }
H
Haojun Liao 已提交
1958 1959 1960 1961 1962
  }

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

L
Liu Jicong 已提交
1963
void updateOutputBuf(SOptrBasicInfo* pBInfo, int32_t* bufCapacity, int32_t numOfInputRows) {
1964 1965
  SSDataBlock* pDataBlock = pBInfo->pRes;

L
Liu Jicong 已提交
1966
  int32_t newSize = pDataBlock->info.rows + numOfInputRows + 5;  // extra output buffer
1967
  if ((*bufCapacity) < newSize) {
L
Liu Jicong 已提交
1968 1969
    for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
      SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1970

wafwerar's avatar
wafwerar 已提交
1971
      char* p = taosMemoryRealloc(pColInfo->pData, newSize * pColInfo->info.bytes);
1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984
      if (p != NULL) {
        pColInfo->pData = p;

        // it starts from the tail of the previously generated results.
        pBInfo->pCtx[i].pOutput = pColInfo->pData;
        (*bufCapacity) = newSize;
      } else {
        // longjmp
      }
    }
  }

  for (int32_t i = 0; i < pDataBlock->info.numOfCols; ++i) {
L
Liu Jicong 已提交
1985
    SColumnInfoData* pColInfo = taosArrayGet(pDataBlock->pDataBlock, i);
1986 1987 1988 1989 1990
    pBInfo->pCtx[i].pOutput = pColInfo->pData + pColInfo->info.bytes * pDataBlock->info.rows;

    // set the correct pointer after the memory buffer reallocated.
    int32_t functionId = pBInfo->pCtx[i].functionId;

L
Liu Jicong 已提交
1991 1992
    if (functionId == FUNCTION_TOP || functionId == FUNCTION_BOTTOM || functionId == FUNCTION_DIFF ||
        functionId == FUNCTION_DERIVATIVE) {
dengyihao's avatar
dengyihao 已提交
1993
      //      if (i > 0) pBInfo->pCtx[i].pTsOutput = pBInfo->pCtx[i - 1].pOutput;
1994 1995 1996 1997
    }
  }
}

H
Haojun Liao 已提交
1998
void copyTsColoum(SSDataBlock* pRes, SqlFunctionCtx* pCtx, int32_t numOfOutput) {
1999 2000
  bool    needCopyTs = false;
  int32_t tsNum = 0;
L
Liu Jicong 已提交
2001
  char*   src = NULL;
2002 2003 2004 2005
  for (int32_t i = 0; i < numOfOutput; i++) {
    int32_t functionId = pCtx[i].functionId;
    if (functionId == FUNCTION_DIFF || functionId == FUNCTION_DERIVATIVE) {
      needCopyTs = true;
L
Liu Jicong 已提交
2006 2007
      if (i > 0 && pCtx[i - 1].functionId == FUNCTION_TS_DUMMY) {
        SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i - 1);  // find ts data
2008 2009
        src = pColRes->pData;
      }
L
Liu Jicong 已提交
2010
    } else if (functionId == FUNCTION_TS_DUMMY) {
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
      tsNum++;
    }
  }

  if (!needCopyTs) return;
  if (tsNum < 2) return;
  if (src == NULL) return;

  for (int32_t i = 0; i < numOfOutput; i++) {
    int32_t functionId = pCtx[i].functionId;
L
Liu Jicong 已提交
2021
    if (functionId == FUNCTION_TS_DUMMY) {
2022 2023 2024 2025 2026 2027
      SColumnInfoData* pColRes = taosArrayGet(pRes->pDataBlock, i);
      memcpy(pColRes->pData, src, pColRes->info.bytes * pRes->info.rows);
    }
  }
}

H
Haojun Liao 已提交
2028
void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size) {
2029 2030
  for (int32_t j = 0; j < size; ++j) {
    struct SResultRowEntryInfo* pResInfo = GET_RES_INFO(&pCtx[j]);
dengyihao's avatar
dengyihao 已提交
2031 2032
    if (isRowEntryInitialized(pResInfo) || fmIsPseudoColumnFunc(pCtx[j].functionId) || pCtx[j].functionId == -1 ||
        fmIsScalarFunc(pCtx[j].functionId)) {
2033 2034 2035
      continue;
    }

H
Haojun Liao 已提交
2036
    pCtx[j].fpSet.init(&pCtx[j], pCtx[j].resultInfo);
2037 2038 2039
  }
}

L
Liu Jicong 已提交
2040
void setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status) {
2041
  if (status == TASK_NOT_COMPLETED) {
H
Haojun Liao 已提交
2042
    pTaskInfo->status = status;
2043 2044
  } else {
    // QUERY_NOT_COMPLETED is not compatible with any other status, so clear its position first
2045
    CLEAR_QUERY_STATUS(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
2046
    pTaskInfo->status |= status;
2047 2048 2049
  }
}

2050
// todo merged with the build group result.
2051 2052
void finalizeMultiTupleQueryResult(int32_t numOfOutput, SDiskbasedBuf* pBuf, SResultRowInfo* pResultRowInfo,
                                   int32_t* rowCellInfoOffset) {
2053 2054 2055
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
    SResultRowPosition* pPos = &pResultRowInfo->pPosition[i];

L
Liu Jicong 已提交
2056
    SFilePage*  bufPage = getBufPage(pBuf, pPos->pageId);
2057
    SResultRow* pRow = (SResultRow*)((char*)bufPage + pPos->offset);
H
Haojun Liao 已提交
2058 2059

    // TODO ignore the close status anyway.
dengyihao's avatar
dengyihao 已提交
2060 2061 2062
    //    if (!isResultRowClosed(pRow)) {
    //      continue;
    //    }
2063

2064
    for (int32_t j = 0; j < numOfOutput; ++j) {
2065
      struct SResultRowEntryInfo* pResInfo = getResultCell(pRow, j, rowCellInfoOffset);
H
Haojun Liao 已提交
2066
      if (!isRowEntryInitialized(pResInfo)) {
2067 2068 2069 2070 2071 2072
        continue;
      }

      if (pRow->numOfRows < pResInfo->numOfRes) {
        pRow->numOfRows = pResInfo->numOfRes;
      }
2073
    }
2074 2075 2076

    releaseBufPage(pBuf, bufPage);
  }
2077 2078
}

2079
STableQueryInfo* createTableQueryInfo(void* buf, STimeWindow win) {
L
Liu Jicong 已提交
2080
  STableQueryInfo* pTableQueryInfo = buf;
2081 2082 2083 2084
  pTableQueryInfo->lastKey = win.skey;
  return pTableQueryInfo;
}

L
Liu Jicong 已提交
2085
void destroyTableQueryInfoImpl(STableQueryInfo* pTableQueryInfo) {
2086 2087 2088 2089
  if (pTableQueryInfo == NULL) {
    return;
  }

L
Liu Jicong 已提交
2090
  //  taosVariantDestroy(&pTableQueryInfo->tag);
dengyihao's avatar
dengyihao 已提交
2091
  //  cleanupResultRowInfo(&pTableQueryInfo->resInfo);
2092 2093
}

2094
void setResultRowInitCtx(SResultRow* pResult, SqlFunctionCtx* pCtx, int32_t numOfOutput, int32_t* rowCellInfoOffset) {
2095 2096 2097 2098 2099 2100 2101
  for (int32_t i = 0; i < numOfOutput; ++i) {
    pCtx[i].resultInfo = getResultCell(pResult, i, rowCellInfoOffset);

    struct SResultRowEntryInfo* pResInfo = pCtx[i].resultInfo;
    if (isRowEntryCompleted(pResInfo) && isRowEntryInitialized(pResInfo)) {
      continue;
    }
2102 2103 2104 2105 2106

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

2107 2108 2109 2110 2111 2112
    if (!pResInfo->initialized) {
      if (pCtx[i].functionId != -1) {
        pCtx[i].fpSet.init(&pCtx[i], pResInfo);
      } else {
        pResInfo->initialized = true;
      }
2113 2114 2115 2116
    }
  }
}

2117 2118 2119 2120 2121 2122
void doFilter(const SNode* pFilterNode, SSDataBlock* pBlock) {
  if (pFilterNode == NULL) {
    return;
  }

  SFilterInfo* filter = NULL;
H
Haojun Liao 已提交
2123

H
Haojun Liao 已提交
2124
  // todo move to the initialization function
H
Haojun Liao 已提交
2125
  int32_t code = filterInitFromNode((SNode*)pFilterNode, &filter, 0);
2126 2127 2128 2129 2130 2131

  SFilterColumnParam param1 = {.numOfCols = pBlock->info.numOfCols, .pDataBlock = pBlock->pDataBlock};
  code = filterSetDataFromSlotId(filter, &param1);

  int8_t* rowRes = NULL;
  bool    keep = filterExecute(filter, pBlock, &rowRes, NULL, param1.numOfCols);
D
dapan1121 已提交
2132
  filterFreeInfo(filter);
2133

2134
  SSDataBlock* px = createOneDataBlock(pBlock, false);
2135 2136
  blockDataEnsureCapacity(px, pBlock->info.rows);

H
Haojun Liao 已提交
2137
  // todo extract method
2138 2139 2140 2141
  int32_t numOfRow = 0;
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
    SColumnInfoData* pDst = taosArrayGet(px->pDataBlock, i);
    SColumnInfoData* pSrc = taosArrayGet(pBlock->pDataBlock, i);
D
dapan1121 已提交
2142 2143 2144 2145 2146 2147 2148 2149 2150
    if (keep) {
      colDataAssign(pDst, pSrc, pBlock->info.rows);
      numOfRow = pBlock->info.rows;
    } else if (NULL != rowRes) {
      numOfRow = 0;
      for (int32_t j = 0; j < pBlock->info.rows; ++j) {
        if (rowRes[j] == 0) {
          continue;
        }
2151

D
dapan1121 已提交
2152 2153 2154 2155 2156 2157
        if (colDataIsNull_s(pSrc, j)) {
          colDataAppendNULL(pDst, numOfRow);
        } else {
          colDataAppend(pDst, numOfRow, colDataGetData(pSrc, j), false);
        }
        numOfRow += 1;
H
Haojun Liao 已提交
2158
      }
D
dapan1121 已提交
2159 2160
    } else {
      numOfRow = 0;
2161 2162 2163 2164 2165 2166
    }

    *pSrc = *pDst;
  }

  pBlock->info.rows = numOfRow;
2167
  blockDataUpdateTsWindow(pBlock);
2168 2169
}

dengyihao's avatar
dengyihao 已提交
2170 2171
void doSetTableGroupOutputBuf(SAggOperatorInfo* pAggInfo, int32_t numOfOutput, uint64_t groupId,
                              SExecTaskInfo* pTaskInfo) {
2172 2173 2174
  // for simple group by query without interval, all the tables belong to one group result.
  int64_t uid = 0;

2175 2176
  SResultRowInfo* pResultRowInfo = &pAggInfo->binfo.resultRowInfo;
  SqlFunctionCtx* pCtx = pAggInfo->binfo.pCtx;
L
Liu Jicong 已提交
2177
  int32_t*        rowCellInfoOffset = pAggInfo->binfo.rowCellInfoOffset;
2178

2179
  SResultRow* pResultRow = doSetResultOutBufByKey(pAggInfo->aggSup.pResultBuf, pResultRowInfo, (char*)&groupId,
L
Liu Jicong 已提交
2180
                                                  sizeof(groupId), true, groupId, pTaskInfo, false, &pAggInfo->aggSup);
L
Liu Jicong 已提交
2181
  assert(pResultRow != NULL);
2182 2183 2184 2185 2186 2187

  /*
   * 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
2188 2189
    int32_t ret =
        addNewWindowResultBuf(pResultRow, pAggInfo->aggSup.pResultBuf, groupId, pAggInfo->binfo.pRes->info.rowSize);
2190 2191 2192 2193 2194
    if (ret != TSDB_CODE_SUCCESS) {
      return;
    }
  }

2195
  setResultRowInitCtx(pResultRow, pCtx, numOfOutput, rowCellInfoOffset);
2196 2197
}

H
Haojun Liao 已提交
2198 2199
void setExecutionContext(int32_t numOfOutput, uint64_t groupId, SExecTaskInfo* pTaskInfo, SAggOperatorInfo* pAggInfo) {
  if (pAggInfo->groupId != INT32_MIN && pAggInfo->groupId == groupId) {
2200 2201 2202
    return;
  }

H
Haojun Liao 已提交
2203
  doSetTableGroupOutputBuf(pAggInfo, numOfOutput, groupId, pTaskInfo);
2204 2205

  // record the current active group id
H
Haojun Liao 已提交
2206
  pAggInfo->groupId = groupId;
2207 2208 2209 2210 2211 2212 2213 2214 2215 2216
}

/**
 * For interval query of both super table and table, copy the data in ascending order, since the output results are
 * ordered in SWindowResutl already. While handling the group by query for both table and super table,
 * all group result are completed already.
 *
 * @param pQInfo
 * @param result
 */
2217
int32_t doCopyToSDataBlock(SExecTaskInfo* taskInfo, SSDataBlock* pBlock, SExprInfo* pExprInfo, SDiskbasedBuf* pBuf, SGroupResInfo* pGroupResInfo,
2218
                           int32_t orderType, int32_t* rowCellOffset, SqlFunctionCtx* pCtx, int32_t numOfExprs) {
2219
  int32_t numOfRows = getNumOfTotalRes(pGroupResInfo);
L
Liu Jicong 已提交
2220
  int32_t numOfResult = pBlock->info.rows;  // there are already exists result rows
2221 2222

  int32_t start = 0;
2223
  int32_t step = 1;
2224

L
Liu Jicong 已提交
2225
  // qDebug("QInfo:0x%"PRIx64" start to copy data from windowResInfo to output buf", GET_TASKID(pRuntimeEnv));
2226 2227 2228 2229 2230 2231 2232 2233 2234
  assert(orderType == TSDB_ORDER_ASC || orderType == TSDB_ORDER_DESC);

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

  for (int32_t i = start; (i < numOfRows) && (i >= 0); i += step) {
L
Liu Jicong 已提交
2235 2236
    SResKeyPos* pPos = taosArrayGetP(pGroupResInfo->pRows, i);
    SFilePage*  page = getBufPage(pBuf, pPos->pos.pageId);
2237

2238
    SResultRow* pRow = (SResultRow*)((char*)page + pPos->pos.offset);
2239 2240 2241 2242 2243
    if (pRow->numOfRows == 0) {
      pGroupResInfo->index += 1;
      continue;
    }

2244
    // TODO copy multiple rows?
2245
    int32_t numOfRowsToCopy = pRow->numOfRows;
2246
    if (numOfResult + numOfRowsToCopy >= pBlock->info.capacity) {
2247 2248 2249 2250 2251
      break;
    }

    pGroupResInfo->index += 1;

2252
    for (int32_t j = 0; j < numOfExprs; ++j) {
2253 2254
      int32_t slotId = pExprInfo[j].base.resSchema.slotId;

2255
      pCtx[j].resultInfo = getResultCell(pRow, j, rowCellOffset);
2256
      if (pCtx[j].fpSet.finalize) {
2257
        int32_t code = pCtx[j].fpSet.finalize(&pCtx[j], pBlock);
2258
        if (TAOS_FAILED(code)) {
2259 2260 2261 2262
          qError("%s build result data block error, code %s", GET_TASKID(taskInfo), tstrerror(code));
          taskInfo->code = code;
          longjmp(taskInfo->env, code);
        }
2263 2264
      } else if (strcmp(pCtx[j].pExpr->pExpr->_function.functionName, "_select_value") == 0) {
        // do nothing, todo refactor
2265
      } else {
2266 2267 2268 2269 2270 2271 2272
        // expand the result into multiple rows. E.g., _wstartts, top(k, 20)
        // the _wstartts needs to copy to 20 following rows, since the results of top-k expands to 20 different rows.
          SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, slotId);
          char* in = GET_ROWCELL_INTERBUF(pCtx[j].resultInfo);
          for(int32_t k = 0; k < pRow->numOfRows; ++k) {
            colDataAppend(pColInfoData, pBlock->info.rows + k, in, pCtx[j].resultInfo->isNullRes);
          }
2273
      }
2274 2275
    }

2276 2277
    releaseBufPage(pBuf, page);

2278 2279
    pBlock->info.rows += pRow->numOfRows;
    if (pBlock->info.rows >= pBlock->info.capacity) {  // output buffer is full
2280 2281 2282 2283
      break;
    }
  }

L
Liu Jicong 已提交
2284
  // qDebug("QInfo:0x%"PRIx64" copy data to query buf completed", GET_TASKID(pRuntimeEnv));
2285
  blockDataUpdateTsWindow(pBlock);
2286 2287 2288
  return 0;
}

2289
void doBuildResultDatablock(SOperatorInfo* pOperator, SOptrBasicInfo* pbInfo, SGroupResInfo* pGroupResInfo, SDiskbasedBuf* pBuf) {
2290 2291
  assert(pGroupResInfo->currentGroup <= pGroupResInfo->totalGroup);

2292 2293 2294 2295
  SExprInfo* pExprInfo = pOperator->pExpr;
  int32_t numOfExprs = pOperator->numOfExprs;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;

X
Xiaoyu Wang 已提交
2296 2297 2298
  int32_t*        rowCellOffset = pbInfo->rowCellInfoOffset;
  SSDataBlock*    pBlock = pbInfo->pRes;
  SqlFunctionCtx* pCtx = pbInfo->pCtx;
2299

2300
  blockDataCleanup(pBlock);
2301 2302 2303 2304
  if (!hasRemainDataInCurrentGroup(pGroupResInfo)) {
    return;
  }

2305
  int32_t orderType = TSDB_ORDER_ASC;
2306
  doCopyToSDataBlock(pTaskInfo, pBlock, pExprInfo, pBuf, pGroupResInfo, orderType, rowCellOffset, pCtx, numOfExprs);
2307

H
Haojun Liao 已提交
2308 2309
  // add condition (pBlock->info.rows >= 1) just to runtime happy
  blockDataUpdateTsWindow(pBlock);
2310 2311
}

L
Liu Jicong 已提交
2312 2313
static void updateNumOfRowsInResultRows(SqlFunctionCtx* pCtx, int32_t numOfOutput, SResultRowInfo* pResultRowInfo,
                                        int32_t* rowCellInfoOffset) {
2314
  // update the number of result for each, only update the number of rows for the corresponding window result.
L
Liu Jicong 已提交
2315 2316 2317
  //  if (QUERY_IS_INTERVAL_QUERY(pQueryAttr)) {
  //    return;
  //  }
H
Haojun Liao 已提交
2318
#if 0
2319
  for (int32_t i = 0; i < pResultRowInfo->size; ++i) {
L
Liu Jicong 已提交
2320
    SResultRow* pResult = pResultRowInfo->pResult[i];
2321 2322 2323 2324 2325 2326 2327

    for (int32_t j = 0; j < numOfOutput; ++j) {
      int32_t functionId = pCtx[j].functionId;
      if (functionId == FUNCTION_TS || functionId == FUNCTION_TAG || functionId == FUNCTION_TAGPRJ) {
        continue;
      }

2328 2329
      SResultRowEntryInfo* pCell = getResultCell(pResult, j, rowCellInfoOffset);
      pResult->numOfRows = (uint16_t)(TMAX(pResult->numOfRows, pCell->numOfRes));
2330 2331
    }
  }
H
Haojun Liao 已提交
2332
#endif
2333 2334
}

L
Liu Jicong 已提交
2335
static int32_t compressQueryColData(SColumnInfoData* pColRes, int32_t numOfRows, char* data, int8_t compressed) {
2336 2337
  int32_t colSize = pColRes->info.bytes * numOfRows;
  return (*(tDataTypes[pColRes->info.type].compFunc))(pColRes->pData, colSize, numOfRows, data,
L
Liu Jicong 已提交
2338
                                                      colSize + COMP_OVERFLOW_BYTES, compressed, NULL, 0);
2339 2340
}

2341
int32_t doFillTimeIntervalGapsInResults(struct SFillInfo* pFillInfo, SSDataBlock* pBlock, int32_t capacity) {
L
Liu Jicong 已提交
2342 2343 2344 2345
  //  for(int32_t i = 0; i < pFillInfo->numOfCols; ++i) {
  //    SColumnInfoData* pColInfoData = taosArrayGet(pOutput->pDataBlock, i);
  //    p[i] = pColInfoData->pData + (pColInfoData->info.bytes * pOutput->info.rows);
  //  }
2346

2347 2348
  int32_t numOfRows = (int32_t)taosFillResultDataBlock(pFillInfo, pBlock, capacity - pBlock->info.rows);
  pBlock->info.rows += numOfRows;
2349

2350
  return pBlock->info.rows;
2351 2352
}

2353
void publishOperatorProfEvent(SOperatorInfo* pOperator, EQueryProfEventType eventType) {
2354 2355
  SQueryProfEvent event = {0};

L
Liu Jicong 已提交
2356 2357
  event.eventType = eventType;
  event.eventTime = taosGetTimestampUs();
2358
  event.operatorType = pOperator->operatorType;
dengyihao's avatar
dengyihao 已提交
2359 2360 2361
  //    if (pQInfo->summary.queryProfEvents) {
  //      taosArrayPush(pQInfo->summary.queryProfEvents, &event);
  //    }
2362 2363
}

L
Liu Jicong 已提交
2364
void publishQueryAbortEvent(SExecTaskInfo* pTaskInfo, int32_t code) {
2365 2366 2367 2368 2369
  SQueryProfEvent event;
  event.eventType = QUERY_PROF_QUERY_ABORT;
  event.eventTime = taosGetTimestampUs();
  event.abortCode = code;

2370 2371
  if (pTaskInfo->cost.queryProfEvents) {
    taosArrayPush(pTaskInfo->cost.queryProfEvents, &event);
2372 2373 2374
  }
}

L
Liu Jicong 已提交
2375
typedef struct {
2376 2377 2378 2379 2380 2381 2382
  uint8_t operatorType;
  int64_t beginTime;
  int64_t endTime;
  int64_t selfTime;
  int64_t descendantsTime;
} SOperatorStackItem;

L
Liu Jicong 已提交
2383 2384
static void doOperatorExecProfOnce(SOperatorStackItem* item, SQueryProfEvent* event, SArray* opStack,
                                   SHashObj* profResults) {
2385 2386 2387 2388 2389 2390 2391 2392
  item->endTime = event->eventTime;
  item->selfTime = (item->endTime - item->beginTime) - (item->descendantsTime);

  for (int32_t j = 0; j < taosArrayGetSize(opStack); ++j) {
    SOperatorStackItem* ancestor = taosArrayGet(opStack, j);
    ancestor->descendantsTime += item->selfTime;
  }

L
Liu Jicong 已提交
2393
  uint8_t              operatorType = item->operatorType;
2394 2395 2396 2397 2398 2399 2400 2401 2402
  SOperatorProfResult* result = taosHashGet(profResults, &operatorType, sizeof(operatorType));
  if (result != NULL) {
    result->sumRunTimes++;
    result->sumSelfTime += item->selfTime;
  } else {
    SOperatorProfResult opResult;
    opResult.operatorType = operatorType;
    opResult.sumSelfTime = item->selfTime;
    opResult.sumRunTimes = 1;
L
Liu Jicong 已提交
2403
    taosHashPut(profResults, &(operatorType), sizeof(operatorType), &opResult, sizeof(opResult));
2404 2405 2406
  }
}

2407
void calculateOperatorProfResults(void) {
dengyihao's avatar
dengyihao 已提交
2408 2409 2410 2411 2412 2413 2414 2415 2416
  //  if (pQInfo->summary.queryProfEvents == NULL) {
  //    // qDebug("QInfo:0x%"PRIx64" query prof events array is null", pQInfo->qId);
  //    return;
  //  }
  //
  //  if (pQInfo->summary.operatorProfResults == NULL) {
  //    // qDebug("QInfo:0x%"PRIx64" operator prof results hash is null", pQInfo->qId);
  //    return;
  //  }
2417 2418 2419 2420 2421

  SArray* opStack = taosArrayInit(32, sizeof(SOperatorStackItem));
  if (opStack == NULL) {
    return;
  }
2422
#if 0
L
Liu Jicong 已提交
2423
  size_t    size = taosArrayGetSize(pQInfo->summary.queryProfEvents);
2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444
  SHashObj* profResults = pQInfo->summary.operatorProfResults;

  for (int i = 0; i < size; ++i) {
    SQueryProfEvent* event = taosArrayGet(pQInfo->summary.queryProfEvents, i);
    if (event->eventType == QUERY_PROF_BEFORE_OPERATOR_EXEC) {
      SOperatorStackItem opItem;
      opItem.operatorType = event->operatorType;
      opItem.beginTime = event->eventTime;
      opItem.descendantsTime = 0;
      taosArrayPush(opStack, &opItem);
    } else if (event->eventType == QUERY_PROF_AFTER_OPERATOR_EXEC) {
      SOperatorStackItem* item = taosArrayPop(opStack);
      assert(item->operatorType == event->operatorType);
      doOperatorExecProfOnce(item, event, opStack, profResults);
    } else if (event->eventType == QUERY_PROF_QUERY_ABORT) {
      SOperatorStackItem* item;
      while ((item = taosArrayPop(opStack)) != NULL) {
        doOperatorExecProfOnce(item, event, opStack, profResults);
      }
    }
  }
2445
#endif
2446 2447 2448
  taosArrayDestroy(opStack);
}

L
Liu Jicong 已提交
2449 2450
void queryCostStatis(SExecTaskInfo* pTaskInfo) {
  STaskCostInfo* pSummary = &pTaskInfo->cost;
2451

L
Liu Jicong 已提交
2452 2453 2454
  //  uint64_t hashSize = taosHashGetMemSize(pQInfo->runtimeEnv.pResultRowHashTable);
  //  hashSize += taosHashGetMemSize(pRuntimeEnv->tableqinfoGroupInfo.map);
  //  pSummary->hashSize = hashSize;
2455 2456 2457 2458

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

L
Liu Jicong 已提交
2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469
  //  SResultRowPool* p = pTaskInfo->pool;
  //  if (p != NULL) {
  //    pSummary->winInfoSize = getResultRowPoolMemSize(p);
  //    pSummary->numOfTimeWindows = getNumOfAllocatedResultRows(p);
  //  } else {
  //    pSummary->winInfoSize = 0;
  //    pSummary->numOfTimeWindows = 0;
  //  }
  //
  //  calculateOperatorProfResults(pQInfo);

2470 2471
  SFileBlockLoadRecorder* pRecorder = pSummary->pRecoder;
  if (pSummary->pRecoder != NULL) {
X
Xiaoyu Wang 已提交
2472 2473
    qDebug("%s :cost summary: elapsed time:%" PRId64 " us, first merge:%" PRId64
           " us, total blocks:%d, "
2474 2475 2476 2477
           "load block statis:%d, load data block:%d, total rows:%" PRId64 ", check rows:%" PRId64,
           GET_TASKID(pTaskInfo), pSummary->elapsedTime, pSummary->firstStageMergeTime, pRecorder->totalBlocks,
           pRecorder->loadBlockStatis, pRecorder->loadBlocks, pRecorder->totalRows, pRecorder->totalCheckedRows);
  }
L
Liu Jicong 已提交
2478 2479 2480
  // 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);
2481 2482 2483 2484

  if (pSummary->operatorProfResults) {
    SOperatorProfResult* opRes = taosHashIterate(pSummary->operatorProfResults, NULL);
    while (opRes != NULL) {
L
Liu Jicong 已提交
2485 2486
      // qDebug("QInfo:0x%" PRIx64 " :cost summary: operator : %d, exec times: %" PRId64 ", self time: %" PRId64,
      //             pQInfo->qId, opRes->operatorType, opRes->sumRunTimes, opRes->sumSelfTime);
2487 2488 2489 2490 2491
      opRes = taosHashIterate(pSummary->operatorProfResults, opRes);
    }
  }
}

L
Liu Jicong 已提交
2492 2493 2494
// static void updateOffsetVal(STaskRuntimeEnv *pRuntimeEnv, SDataBlockInfo *pBlockInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
2495
//
L
Liu Jicong 已提交
2496
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2497
//
L
Liu Jicong 已提交
2498 2499 2500 2501
//   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;
//   }
2502
//
L
Liu Jicong 已提交
2503 2504 2505 2506 2507
//   if (QUERY_IS_ASC_QUERY(pQueryAttr)) {
//     pQueryAttr->pos = (int32_t)pQueryAttr->limit.offset;
//   } else {
//     pQueryAttr->pos = pBlockInfo->rows - (int32_t)pQueryAttr->limit.offset - 1;
//   }
2508
//
L
Liu Jicong 已提交
2509
//   assert(pQueryAttr->pos >= 0 && pQueryAttr->pos <= pBlockInfo->rows - 1);
2510
//
L
Liu Jicong 已提交
2511 2512
//   SArray *         pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//   SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2513
//
L
Liu Jicong 已提交
2514 2515
//   // update the pQueryAttr->limit.offset value, and pQueryAttr->pos value
//   TSKEY *keys = (TSKEY *) pColInfoData->pData;
2516
//
L
Liu Jicong 已提交
2517 2518 2519
//   // update the offset value
//   pTableQueryInfo->lastKey = keys[pQueryAttr->pos];
//   pQueryAttr->limit.offset = 0;
2520
//
L
Liu Jicong 已提交
2521
//   int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
2522
//
L
Liu Jicong 已提交
2523 2524 2525 2526
//   //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);
// }
2527

L
Liu Jicong 已提交
2528 2529
// void skipBlocks(STaskRuntimeEnv *pRuntimeEnv) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
2530
//
L
Liu Jicong 已提交
2531 2532 2533
//   if (pQueryAttr->limit.offset <= 0 || pQueryAttr->numOfFilterCols > 0) {
//     return;
//   }
2534
//
L
Liu Jicong 已提交
2535 2536
//   pQueryAttr->pos = 0;
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(pQueryAttr->order.order);
2537
//
L
Liu Jicong 已提交
2538 2539
//   STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
//   TsdbQueryHandleT pTsdbReadHandle = pRuntimeEnv->pTsdbReadHandle;
2540
//
L
Liu Jicong 已提交
2541 2542 2543 2544 2545
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pTsdbReadHandle)) {
//     if (isTaskKilled(pRuntimeEnv->qinfo)) {
//       longjmp(pRuntimeEnv->env, TSDB_CODE_TSC_QUERY_CANCELLED);
//     }
2546
//
L
Liu Jicong 已提交
2547
//     tsdbRetrieveDataBlockInfo(pTsdbReadHandle, &blockInfo);
2548
//
L
Liu Jicong 已提交
2549 2550 2551 2552
//     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;
2553
//
L
Liu Jicong 已提交
2554 2555 2556 2557 2558 2559 2560
//       //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;
//     }
//   }
2561
//
L
Liu Jicong 已提交
2562 2563 2564 2565 2566 2567 2568 2569 2570
//   if (terrno != TSDB_CODE_SUCCESS) {
//     longjmp(pRuntimeEnv->env, terrno);
//   }
// }

// static TSKEY doSkipIntervalProcess(STaskRuntimeEnv* pRuntimeEnv, STimeWindow* win, SDataBlockInfo* pBlockInfo,
// STableQueryInfo* pTableQueryInfo) {
//   STaskAttr *pQueryAttr = pRuntimeEnv->pQueryAttr;
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
2571
//
L
Liu Jicong 已提交
2572 2573 2574
//   assert(pQueryAttr->limit.offset == 0);
//   STimeWindow tw = *win;
//   getNextTimeWindow(pQueryAttr, &tw);
2575
//
L
Liu Jicong 已提交
2576 2577
//   if ((tw.skey <= pBlockInfo->window.ekey && QUERY_IS_ASC_QUERY(pQueryAttr)) ||
//       (tw.ekey >= pBlockInfo->window.skey && !QUERY_IS_ASC_QUERY(pQueryAttr))) {
2578
//
L
Liu Jicong 已提交
2579 2580 2581 2582
//     // 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);
2583
//
L
Liu Jicong 已提交
2584 2585 2586 2587
//     tw = *win;
//     int32_t startPos =
//         getNextQualifiedWindow(pQueryAttr, &tw, pBlockInfo, pColInfoData->pData, binarySearchForKey, -1);
//     assert(startPos >= 0);
2588
//
L
Liu Jicong 已提交
2589 2590
//     // set the abort info
//     pQueryAttr->pos = startPos;
2591
//
L
Liu Jicong 已提交
2592 2593 2594 2595
//     // reset the query start timestamp
//     pTableQueryInfo->win.skey = ((TSKEY *)pColInfoData->pData)[startPos];
//     pQueryAttr->window.skey = pTableQueryInfo->win.skey;
//     TSKEY key = pTableQueryInfo->win.skey;
2596
//
L
Liu Jicong 已提交
2597 2598
//     pWindowResInfo->prevSKey = tw.skey;
//     int32_t index = pRuntimeEnv->resultRowInfo.curIndex;
2599
//
L
Liu Jicong 已提交
2600 2601
//     int32_t numOfRes = tableApplyFunctionsOnBlock(pRuntimeEnv, pBlockInfo, NULL, binarySearchForKey, pDataBlock);
//     pRuntimeEnv->resultRowInfo.curIndex = index;  // restore the window index
2602
//
L
Liu Jicong 已提交
2603 2604 2605 2606
//     //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);
2607
//
L
Liu Jicong 已提交
2608 2609 2610 2611 2612
//     return key;
//   } else {  // do nothing
//     pQueryAttr->window.skey      = tw.skey;
//     pWindowResInfo->prevSKey = tw.skey;
//     pTableQueryInfo->lastKey = tw.skey;
2613
//
L
Liu Jicong 已提交
2614 2615
//     return tw.skey;
//   }
2616
//
L
Liu Jicong 已提交
2617 2618 2619 2620 2621 2622 2623 2624 2625 2626
//   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);
//   }
2627
//
L
Liu Jicong 已提交
2628 2629 2630 2631 2632
//   // 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;
//   }
2633
//
L
Liu Jicong 已提交
2634 2635 2636 2637 2638 2639 2640
//   /*
//    * 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);
2641
//
L
Liu Jicong 已提交
2642 2643
//   STimeWindow w = TSWINDOW_INITIALIZER;
//   bool ascQuery = QUERY_IS_ASC_QUERY(pQueryAttr);
2644
//
L
Liu Jicong 已提交
2645 2646
//   SResultRowInfo *pWindowResInfo = &pRuntimeEnv->resultRowInfo;
//   STableQueryInfo *pTableQueryInfo = pRuntimeEnv->current;
2647
//
L
Liu Jicong 已提交
2648 2649 2650
//   SDataBlockInfo blockInfo = SDATA_BLOCK_INITIALIZER;
//   while (tsdbNextDataBlock(pRuntimeEnv->pTsdbReadHandle)) {
//     tsdbRetrieveDataBlockInfo(pRuntimeEnv->pTsdbReadHandle, &blockInfo);
2651
//
L
Liu Jicong 已提交
2652 2653 2654 2655 2656 2657 2658 2659 2660
//     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;
//     }
2661
//
L
Liu Jicong 已提交
2662 2663
//     // the first time window
//     STimeWindow win = getActiveTimeWindow(pWindowResInfo, pWindowResInfo->prevSKey, pQueryAttr);
2664
//
L
Liu Jicong 已提交
2665 2666
//     while (pQueryAttr->limit.offset > 0) {
//       STimeWindow tw = win;
2667
//
L
Liu Jicong 已提交
2668 2669 2670
//       if ((win.ekey <= blockInfo.window.ekey && ascQuery) || (win.ekey >= blockInfo.window.skey && !ascQuery)) {
//         pQueryAttr->limit.offset -= 1;
//         pWindowResInfo->prevSKey = win.skey;
2671
//
L
Liu Jicong 已提交
2672 2673 2674 2675 2676 2677
//         // 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;
//         }
//       }
2678
//
L
Liu Jicong 已提交
2679 2680 2681 2682
//       if (pQueryAttr->limit.offset == 0) {
//         *start = doSkipIntervalProcess(pRuntimeEnv, &win, &blockInfo, pTableQueryInfo);
//         return true;
//       }
2683
//
L
Liu Jicong 已提交
2684 2685
//       // current window does not ended in current data block, try next data block
//       getNextTimeWindow(pQueryAttr, &tw);
2686
//
L
Liu Jicong 已提交
2687 2688 2689 2690 2691 2692 2693 2694 2695
//       /*
//        * 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)) {
2696
//
L
Liu Jicong 已提交
2697 2698
//         SArray *pDataBlock = tsdbRetrieveDataBlock(pRuntimeEnv->pTsdbReadHandle, NULL);
//         SColumnInfoData *pColInfoData = taosArrayGet(pDataBlock, 0);
2699
//
L
Liu Jicong 已提交
2700 2701 2702
//         if ((win.ekey > blockInfo.window.ekey && ascQuery) || (win.ekey < blockInfo.window.skey && !ascQuery)) {
//           pQueryAttr->limit.offset -= 1;
//         }
2703
//
L
Liu Jicong 已提交
2704 2705 2706 2707 2708 2709 2710 2711
//         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);
2712
//
L
Liu Jicong 已提交
2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723
//           // 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.
//       }
//     }
//   }
2724
//
L
Liu Jicong 已提交
2725 2726 2727 2728
//   // check for error
//   if (terrno != TSDB_CODE_SUCCESS) {
//     longjmp(pRuntimeEnv->env, terrno);
//   }
2729
//
L
Liu Jicong 已提交
2730 2731
//   return true;
// }
2732

2733
int32_t appendDownstream(SOperatorInfo* p, SOperatorInfo** pDownstream, int32_t num) {
H
Haojun Liao 已提交
2734
  if (p->pDownstream == NULL) {
H
Haojun Liao 已提交
2735
    assert(p->numOfDownstream == 0);
2736 2737
  }

wafwerar's avatar
wafwerar 已提交
2738
  p->pDownstream = taosMemoryCalloc(1, num * POINTER_BYTES);
2739 2740 2741 2742 2743 2744 2745
  if (p->pDownstream == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

  memcpy(p->pDownstream, pDownstream, num * POINTER_BYTES);
  p->numOfDownstream = num;
  return TSDB_CODE_SUCCESS;
2746 2747 2748 2749
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo);

2750
static void doTableQueryInfoTimeWindowCheck(SExecTaskInfo* pTaskInfo, STableQueryInfo* pTableQueryInfo, int32_t order) {
H
Haojun Liao 已提交
2751 2752
#if 0
    if (order == TSDB_ORDER_ASC) {
2753 2754
    assert(
        (pTableQueryInfo->win.skey <= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2755 2756
        (pTableQueryInfo->lastKey >= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey >= pTaskInfo->window.skey && pTableQueryInfo->win.ekey <= pTaskInfo->window.ekey));
2757 2758 2759
  } else {
    assert(
        (pTableQueryInfo->win.skey >= pTableQueryInfo->win.ekey) &&
H
Haojun Liao 已提交
2760 2761
        (pTableQueryInfo->lastKey <= pTaskInfo->window.skey) &&
        (pTableQueryInfo->win.skey <= pTaskInfo->window.skey && pTableQueryInfo->win.ekey >= pTaskInfo->window.ekey));
2762
  }
H
Haojun Liao 已提交
2763
#endif
2764 2765
}

L
Liu Jicong 已提交
2766 2767 2768 2769
// static void updateTableIdInfo(STableQueryInfo* pTableQueryInfo, SSDataBlock* pBlock, SHashObj* pTableIdInfo, int32_t
// order) {
//   int32_t step = GET_FORWARD_DIRECTION_FACTOR(order);
//   pTableQueryInfo->lastKey = ((order == TSDB_ORDER_ASC)? pBlock->info.window.ekey:pBlock->info.window.skey) + step;
2770
//
L
Liu Jicong 已提交
2771 2772 2773
//   if (pTableQueryInfo->pTable == NULL) {
//     return;
//   }
2774
//
L
Liu Jicong 已提交
2775 2776 2777 2778 2779 2780 2781 2782 2783
//   STableIdInfo tidInfo = createTableIdInfo(pTableQueryInfo);
//   STableIdInfo *idinfo = taosHashGet(pTableIdInfo, &tidInfo.tid, sizeof(tidInfo.tid));
//   if (idinfo != NULL) {
//     assert(idinfo->tid == tidInfo.tid && idinfo->uid == tidInfo.uid);
//     idinfo->key = tidInfo.key;
//   } else {
//     taosHashPut(pTableIdInfo, &tidInfo.tid, sizeof(tidInfo.tid), &tidInfo, sizeof(STableIdInfo));
//   }
// }
2784

2785
int32_t loadRemoteDataCallback(void* param, const SDataBuf* pMsg, int32_t code) {
L
Liu Jicong 已提交
2786
  SSourceDataInfo* pSourceDataInfo = (SSourceDataInfo*)param;
H
Haojun Liao 已提交
2787 2788
  if (code == TSDB_CODE_SUCCESS) {
    pSourceDataInfo->pRsp = pMsg->pData;
2789

H
Haojun Liao 已提交
2790 2791
    SRetrieveTableRsp* pRsp = pSourceDataInfo->pRsp;
    pRsp->numOfRows = htonl(pRsp->numOfRows);
dengyihao's avatar
dengyihao 已提交
2792
    pRsp->compLen = htonl(pRsp->compLen);
2793
    pRsp->numOfCols = htonl(pRsp->numOfCols);
dengyihao's avatar
dengyihao 已提交
2794
    pRsp->useconds = htobe64(pRsp->useconds);
H
Haojun Liao 已提交
2795 2796 2797
  } else {
    pSourceDataInfo->code = code;
  }
H
Haojun Liao 已提交
2798

H
Haojun Liao 已提交
2799
  pSourceDataInfo->status = EX_SOURCE_DATA_READY;
2800
  tsem_post(&pSourceDataInfo->pEx->ready);
wmmhello's avatar
wmmhello 已提交
2801
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
2802 2803 2804 2805
}

static void destroySendMsgInfo(SMsgSendInfo* pMsgBody) {
  assert(pMsgBody != NULL);
wafwerar's avatar
wafwerar 已提交
2806 2807
  taosMemoryFreeClear(pMsgBody->msgInfo.pData);
  taosMemoryFreeClear(pMsgBody);
H
Haojun Liao 已提交
2808 2809
}

S
Shengliang Guan 已提交
2810
void qProcessFetchRsp(void* parent, SRpcMsg* pMsg, SEpSet* pEpSet) {
L
Liu Jicong 已提交
2811
  SMsgSendInfo* pSendInfo = (SMsgSendInfo*)pMsg->ahandle;
H
Haojun Liao 已提交
2812 2813 2814 2815 2816
  assert(pMsg->ahandle != NULL);

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

  if (pMsg->contLen > 0) {
wafwerar's avatar
wafwerar 已提交
2817
    buf.pData = taosMemoryCalloc(1, pMsg->contLen);
H
Haojun Liao 已提交
2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828
    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);
2829 2830
}

L
Liu Jicong 已提交
2831
static int32_t doSendFetchDataRequest(SExchangeInfo* pExchangeInfo, SExecTaskInfo* pTaskInfo, int32_t sourceIndex) {
2832
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
2833

wafwerar's avatar
wafwerar 已提交
2834
  SResFetchReq* pMsg = taosMemoryCalloc(1, sizeof(SResFetchReq));
2835 2836 2837 2838
  if (NULL == pMsg) {
    pTaskInfo->code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return pTaskInfo->code;
  }
2839

L
Liu Jicong 已提交
2840 2841
  SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, sourceIndex);
  SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, sourceIndex);
2842

L
Liu Jicong 已提交
2843 2844
  qDebug("%s build fetch msg and send to vgId:%d, ep:%s, taskId:0x%" PRIx64 ", %d/%" PRIzu, GET_TASKID(pTaskInfo),
         pSource->addr.nodeId, pSource->addr.epSet.eps[0].fqdn, pSource->taskId, sourceIndex, totalSources);
2845 2846 2847 2848 2849 2850 2851

  pMsg->header.vgId = htonl(pSource->addr.nodeId);
  pMsg->sId = htobe64(pSource->schedId);
  pMsg->taskId = htobe64(pSource->taskId);
  pMsg->queryId = htobe64(pTaskInfo->id.queryId);

  // send the fetch remote task result reques
wafwerar's avatar
wafwerar 已提交
2852
  SMsgSendInfo* pMsgSendInfo = taosMemoryCalloc(1, sizeof(SMsgSendInfo));
2853
  if (NULL == pMsgSendInfo) {
wafwerar's avatar
wafwerar 已提交
2854
    taosMemoryFreeClear(pMsg);
2855 2856 2857
    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
2858 2859
  }

2860 2861 2862 2863 2864
  pMsgSendInfo->param = pDataInfo;
  pMsgSendInfo->msgInfo.pData = pMsg;
  pMsgSendInfo->msgInfo.len = sizeof(SResFetchReq);
  pMsgSendInfo->msgType = TDMT_VND_FETCH;
  pMsgSendInfo->fp = loadRemoteDataCallback;
2865

2866
  int64_t transporterId = 0;
L
Liu Jicong 已提交
2867
  int32_t code = asyncSendMsgToServer(pExchangeInfo->pTransporter, &pSource->addr.epSet, &transporterId, pMsgSendInfo);
2868 2869 2870
  return TSDB_CODE_SUCCESS;
}

2871
// NOTE: sources columns are more than the destination SSDatablock columns.
2872
void relocateColumnData(SSDataBlock* pBlock, const SArray* pColMatchInfo, SArray* pCols) {
2873 2874 2875
  size_t numOfSrcCols = taosArrayGetSize(pCols);

  int32_t i = 0, j = 0;
L
Liu Jicong 已提交
2876
  while (i < numOfSrcCols && j < taosArrayGetSize(pColMatchInfo)) {
2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895
    SColumnInfoData* p = taosArrayGet(pCols, i);
    SColMatchInfo*   pmInfo = taosArrayGet(pColMatchInfo, j);
    if (!pmInfo->output) {
      j++;
      continue;
    }

    if (p->info.colId == pmInfo->colId) {
      taosArraySet(pBlock->pDataBlock, pmInfo->targetSlotId, p);
      i++;
      j++;
    } else if (p->info.colId < pmInfo->colId) {
      i++;
    } else {
      ASSERT(0);
    }
  }
}

2896
int32_t setSDataBlockFromFetchRsp(SSDataBlock* pRes, SLoadRemoteDataInfo* pLoadInfo, int32_t numOfRows, char* pData,
dengyihao's avatar
dengyihao 已提交
2897 2898
                                  int32_t compLen, int32_t numOfOutput, int64_t startTs, uint64_t* total,
                                  SArray* pColList) {
H
Haojun Liao 已提交
2899
  if (pColList == NULL) {  // data from other sources
H
Haojun Liao 已提交
2900 2901
    blockDataEnsureCapacity(pRes, numOfRows);

L
Liu Jicong 已提交
2902
    int32_t dataLen = *(int32_t*)pData;
2903 2904
    pData += sizeof(int32_t);

L
Liu Jicong 已提交
2905
    pRes->info.groupId = *(uint64_t*)pData;
2906 2907
    pData += sizeof(uint64_t);

H
Haojun Liao 已提交
2908 2909
    int32_t* colLen = (int32_t*)pData;

2910
    char* pStart = pData + sizeof(int32_t) * numOfOutput;
2911
    for (int32_t i = 0; i < numOfOutput; ++i) {
H
Haojun Liao 已提交
2912
      colLen[i] = htonl(colLen[i]);
2913
      ASSERT(colLen[i] >= 0);
H
Haojun Liao 已提交
2914

2915
      SColumnInfoData* pColInfoData = taosArrayGet(pRes->pDataBlock, i);
H
Haojun Liao 已提交
2916 2917 2918
      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
        pColInfoData->varmeta.length = colLen[i];
        pColInfoData->varmeta.allocLen = colLen[i];
2919

L
Liu Jicong 已提交
2920 2921
        memcpy(pColInfoData->varmeta.offset, pStart, sizeof(int32_t) * numOfRows);
        pStart += sizeof(int32_t) * numOfRows;
H
Haojun Liao 已提交
2922

2923 2924 2925
        if (colLen[i] > 0) {
          pColInfoData->pData = taosMemoryMalloc(colLen[i]);
        }
H
Haojun Liao 已提交
2926 2927 2928
      } else {
        memcpy(pColInfoData->nullbitmap, pStart, BitmapLen(numOfRows));
        pStart += BitmapLen(numOfRows);
2929
      }
H
Haojun Liao 已提交
2930

2931 2932 2933 2934
      if (colLen[i] > 0) {
        memcpy(pColInfoData->pData, pStart, colLen[i]);
      }

L
Liu Jicong 已提交
2935 2936
      // TODO setting this flag to true temporarily so aggregate function on stable will
      // examine NULL value for non-primary key column
2937
      pColInfoData->hasNull = true;
H
Haojun Liao 已提交
2938
      pStart += colLen[i];
2939
    }
H
Haojun Liao 已提交
2940
  } else {  // extract data according to pColList
2941
    ASSERT(numOfOutput == taosArrayGetSize(pColList));
2942 2943 2944 2945 2946
    char* pStart = pData;

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

2947
    // todo refactor:extract method
2948
    SSysTableSchema* pSchema = (SSysTableSchema*)pStart;
dengyihao's avatar
dengyihao 已提交
2949
    for (int32_t i = 0; i < numOfCols; ++i) {
2950 2951 2952 2953 2954 2955 2956
      SSysTableSchema* p = (SSysTableSchema*)pStart;

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

H
Haojun Liao 已提交
2957 2958 2959 2960
    SSDataBlock* pBlock = taosMemoryCalloc(1, sizeof(SSDataBlock));
    pBlock->pDataBlock = taosArrayInit(numOfCols, sizeof(SColumnInfoData));
    pBlock->info.numOfCols = numOfCols;

dengyihao's avatar
dengyihao 已提交
2961
    for (int32_t i = 0; i < numOfCols; ++i) {
2962 2963 2964 2965 2966
      SColumnInfoData idata = {0};
      idata.info.type = pSchema[i].type;
      idata.info.bytes = pSchema[i].bytes;
      idata.info.colId = pSchema[i].colId;

H
Haojun Liao 已提交
2967
      taosArrayPush(pBlock->pDataBlock, &idata);
2968
      if (IS_VAR_DATA_TYPE(idata.info.type)) {
H
Haojun Liao 已提交
2969
        pBlock->info.hasVarCol = true;
2970 2971 2972
      }
    }

H
Haojun Liao 已提交
2973
    blockDataEnsureCapacity(pBlock, numOfRows);
2974

L
Liu Jicong 已提交
2975 2976
    int32_t  dataLen = *(int32_t*)pStart;
    uint64_t groupId = *(uint64_t*)(pStart + sizeof(int32_t));
2977 2978
    pStart += sizeof(int32_t) + sizeof(uint64_t);

L
Liu Jicong 已提交
2979
    int32_t* colLen = (int32_t*)(pStart);
2980 2981 2982 2983
    pStart += sizeof(int32_t) * numOfCols;

    for (int32_t i = 0; i < numOfCols; ++i) {
      colLen[i] = htonl(colLen[i]);
2984
      ASSERT(colLen[i] >= 0);
2985

H
Haojun Liao 已提交
2986
      SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002
      if (IS_VAR_DATA_TYPE(pColInfoData->info.type)) {
        pColInfoData->varmeta.length = colLen[i];
        pColInfoData->varmeta.allocLen = colLen[i];

        memcpy(pColInfoData->varmeta.offset, pStart, sizeof(int32_t) * numOfRows);
        pStart += sizeof(int32_t) * numOfRows;

        pColInfoData->pData = taosMemoryMalloc(colLen[i]);
      } else {
        memcpy(pColInfoData->nullbitmap, pStart, BitmapLen(numOfRows));
        pStart += BitmapLen(numOfRows);
      }

      memcpy(pColInfoData->pData, pStart, colLen[i]);
      pStart += colLen[i];
    }
H
Haojun Liao 已提交
3003 3004

    // data from mnode
H
Haojun Liao 已提交
3005 3006 3007
    relocateColumnData(pRes, pColList, pBlock->pDataBlock);
    taosArrayDestroy(pBlock->pDataBlock);
    taosMemoryFree(pBlock);
X
Xiaoyu Wang 已提交
3008
    //    blockDataDestroy(pBlock);
3009
  }
3010

H
Haojun Liao 已提交
3011
  pRes->info.rows = numOfRows;
3012
  blockDataUpdateTsWindow(pRes);
3013

3014
  int64_t el = taosGetTimestampUs() - startTs;
3015

H
Haojun Liao 已提交
3016 3017
  pLoadInfo->totalRows += numOfRows;
  pLoadInfo->totalSize += compLen;
3018

H
Haojun Liao 已提交
3019 3020 3021
  if (total != NULL) {
    *total += numOfRows;
  }
3022

H
Haojun Liao 已提交
3023
  pLoadInfo->totalElapsed += el;
3024 3025
  return TSDB_CODE_SUCCESS;
}
3026

L
Liu Jicong 已提交
3027 3028
static void* setAllSourcesCompleted(SOperatorInfo* pOperator, int64_t startTs) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
3029
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3030

L
Liu Jicong 已提交
3031
  int64_t              el = taosGetTimestampUs() - startTs;
H
Haojun Liao 已提交
3032 3033
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
  pLoadInfo->totalElapsed += el;
H
Haojun Liao 已提交
3034

3035
  size_t totalSources = taosArrayGetSize(pExchangeInfo->pSources);
L
Liu Jicong 已提交
3036 3037 3038
  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);
3039 3040 3041 3042 3043

  doSetOperatorCompleted(pOperator);
  return NULL;
}

L
Liu Jicong 已提交
3044 3045
static SSDataBlock* concurrentlyLoadRemoteDataImpl(SOperatorInfo* pOperator, SExchangeInfo* pExchangeInfo,
                                                   SExecTaskInfo* pTaskInfo) {
3046 3047 3048 3049 3050 3051 3052 3053 3054
  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);

3055
      if (pDataInfo->status == EX_SOURCE_DATA_EXHAUSTED) {
3056
        completed += 1;
H
Haojun Liao 已提交
3057 3058
        continue;
      }
3059

3060
      if (pDataInfo->status != EX_SOURCE_DATA_READY) {
3061 3062 3063
        continue;
      }

L
Liu Jicong 已提交
3064
      SRetrieveTableRsp*     pRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3065
      SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, i);
3066

L
Liu Jicong 已提交
3067
      SSDataBlock*         pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3068
      SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3069
      if (pRsp->numOfRows == 0) {
L
Liu Jicong 已提交
3070 3071
        qDebug("%s vgId:%d, taskID:0x%" PRIx64 " index:%d completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
               " try next",
3072
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, i + 1, pDataInfo->totalRows,
H
Haojun Liao 已提交
3073
               pExchangeInfo->loadInfo.totalRows);
3074
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3075 3076 3077
        completed += 1;
        continue;
      }
H
Haojun Liao 已提交
3078

H
Haojun Liao 已提交
3079
      SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
X
Xiaoyu Wang 已提交
3080 3081
      code = setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
                                       pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3082
      if (code != 0) {
3083 3084 3085
        goto _error;
      }

3086 3087 3088
      if (pRsp->completed == 1) {
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
               ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
L
Liu Jicong 已提交
3089 3090
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
               pLoadInfo->totalRows, pLoadInfo->totalSize, i + 1, totalSources);
3091
        pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3092
      } else {
dengyihao's avatar
dengyihao 已提交
3093 3094
        qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
               ", totalBytes:%" PRIu64,
H
Haojun Liao 已提交
3095 3096
               GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
               pLoadInfo->totalSize);
3097 3098
      }

3099 3100
      if (pDataInfo->status != EX_SOURCE_DATA_EXHAUSTED) {
        pDataInfo->status = EX_SOURCE_DATA_NOT_READY;
3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119
        code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
        if (code != TSDB_CODE_SUCCESS) {
          goto _error;
        }
      }

      return pExchangeInfo->pResult;
    }

    if (completed == totalSources) {
      return setAllSourcesCompleted(pOperator, startTs);
    }
  }

_error:
  pTaskInfo->code = code;
  return NULL;
}

L
Liu Jicong 已提交
3120 3121 3122
static SSDataBlock* concurrentlyLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
H
Haojun Liao 已提交
3123

H
Haojun Liao 已提交
3124 3125 3126
  if (pOperator->status == OP_RES_TO_RETURN) {
    return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
  }
3127

L
Liu Jicong 已提交
3128
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3129 3130 3131
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3132
  for (int32_t i = 0; i < totalSources; ++i) {
3133 3134
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3135
      return NULL;
3136 3137 3138 3139
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3140 3141
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3142 3143

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3144 3145
  pOperator->status = OP_RES_TO_RETURN;
  return concurrentlyLoadRemoteDataImpl(pOperator, pExchangeInfo, pTaskInfo);
3146 3147
}

L
Liu Jicong 已提交
3148 3149 3150
static int32_t prepareConcurrentlyLoad(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3151

L
Liu Jicong 已提交
3152
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3153 3154 3155
  int64_t startTs = taosGetTimestampUs();

  // Asynchronously send all fetch requests to all sources.
L
Liu Jicong 已提交
3156
  for (int32_t i = 0; i < totalSources; ++i) {
3157 3158
    int32_t code = doSendFetchDataRequest(pExchangeInfo, pTaskInfo, i);
    if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3159 3160
      pTaskInfo->code = code;
      return code;
3161 3162 3163 3164
    }
  }

  int64_t endTs = taosGetTimestampUs();
L
Liu Jicong 已提交
3165 3166
  qDebug("%s send all fetch request to %" PRIzu " sources completed, elapsed:%" PRId64, GET_TASKID(pTaskInfo),
         totalSources, endTs - startTs);
3167 3168

  tsem_wait(&pExchangeInfo->ready);
H
Haojun Liao 已提交
3169
  pOperator->cost.openCost = taosGetTimestampUs() - startTs;
3170

H
Haojun Liao 已提交
3171
  return TSDB_CODE_SUCCESS;
3172 3173
}

L
Liu Jicong 已提交
3174 3175 3176
static SSDataBlock* seqLoadRemoteData(SOperatorInfo* pOperator) {
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3177

L
Liu Jicong 已提交
3178
  size_t  totalSources = taosArrayGetSize(pExchangeInfo->pSources);
3179
  int64_t startTs = taosGetTimestampUs();
3180

L
Liu Jicong 已提交
3181
  while (1) {
3182 3183
    if (pExchangeInfo->current >= totalSources) {
      return setAllSourcesCompleted(pOperator, startTs);
3184
    }
3185

3186 3187 3188
    doSendFetchDataRequest(pExchangeInfo, pTaskInfo, pExchangeInfo->current);
    tsem_wait(&pExchangeInfo->ready);

dengyihao's avatar
dengyihao 已提交
3189
    SSourceDataInfo*       pDataInfo = taosArrayGet(pExchangeInfo->pSourceDataInfo, pExchangeInfo->current);
X
Xiaoyu Wang 已提交
3190
    SDownstreamSourceNode* pSource = taosArrayGet(pExchangeInfo->pSources, pExchangeInfo->current);
3191

H
Haojun Liao 已提交
3192
    if (pDataInfo->code != TSDB_CODE_SUCCESS) {
dengyihao's avatar
dengyihao 已提交
3193 3194
      qError("%s vgId:%d, taskID:0x%" PRIx64 " error happens, code:%s", GET_TASKID(pTaskInfo), pSource->addr.nodeId,
             pSource->taskId, tstrerror(pDataInfo->code));
H
Haojun Liao 已提交
3195 3196 3197 3198
      pOperator->pTaskInfo->code = pDataInfo->code;
      return NULL;
    }

L
Liu Jicong 已提交
3199
    SRetrieveTableRsp*   pRsp = pDataInfo->pRsp;
H
Haojun Liao 已提交
3200
    SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
3201
    if (pRsp->numOfRows == 0) {
dengyihao's avatar
dengyihao 已提交
3202 3203
      qDebug("%s vgId:%d, taskID:0x%" PRIx64 " %d of total completed, rowsOfSource:%" PRIu64 ", totalRows:%" PRIu64
             " try next",
3204
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pExchangeInfo->current + 1,
H
Haojun Liao 已提交
3205
             pDataInfo->totalRows, pLoadInfo->totalRows);
H
Haojun Liao 已提交
3206

3207
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3208 3209 3210
      pExchangeInfo->current += 1;
      continue;
    }
H
Haojun Liao 已提交
3211

L
Liu Jicong 已提交
3212
    SSDataBlock*       pRes = pExchangeInfo->pResult;
H
Haojun Liao 已提交
3213
    SRetrieveTableRsp* pTableRsp = pDataInfo->pRsp;
L
Liu Jicong 已提交
3214 3215
    int32_t            code =
        setSDataBlockFromFetchRsp(pExchangeInfo->pResult, pLoadInfo, pTableRsp->numOfRows, pTableRsp->data,
3216
                                  pTableRsp->compLen, pTableRsp->numOfCols, startTs, &pDataInfo->totalRows, NULL);
3217 3218

    if (pRsp->completed == 1) {
H
Haojun Liao 已提交
3219
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, rowsOfSource:%" PRIu64
L
Liu Jicong 已提交
3220 3221 3222
             ", totalRows:%" PRIu64 ", totalBytes:%" PRIu64 " try next %d/%" PRIzu,
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pDataInfo->totalRows,
             pLoadInfo->totalRows, pLoadInfo->totalSize, pExchangeInfo->current + 1, totalSources);
3223

3224
      pDataInfo->status = EX_SOURCE_DATA_EXHAUSTED;
3225 3226
      pExchangeInfo->current += 1;
    } else {
L
Liu Jicong 已提交
3227 3228 3229 3230
      qDebug("%s fetch msg rsp from vgId:%d, taskId:0x%" PRIx64 " numOfRows:%d, totalRows:%" PRIu64
             ", totalBytes:%" PRIu64,
             GET_TASKID(pTaskInfo), pSource->addr.nodeId, pSource->taskId, pRes->info.rows, pLoadInfo->totalRows,
             pLoadInfo->totalSize);
3231 3232 3233 3234
    }

    return pExchangeInfo->pResult;
  }
3235 3236
}

L
Liu Jicong 已提交
3237
static int32_t prepareLoadRemoteData(SOperatorInfo* pOperator) {
3238
  if (OPTR_IS_OPENED(pOperator)) {
H
Haojun Liao 已提交
3239 3240 3241
    return TSDB_CODE_SUCCESS;
  }

L
Liu Jicong 已提交
3242
  SExchangeInfo* pExchangeInfo = pOperator->info;
H
Haojun Liao 已提交
3243 3244 3245 3246 3247 3248 3249 3250 3251
  if (pExchangeInfo->seqLoadData) {
    // do nothing for sequentially load data
  } else {
    int32_t code = prepareConcurrentlyLoad(pOperator);
    if (code != TSDB_CODE_SUCCESS) {
      return code;
    }
  }

3252
  OPTR_SET_OPENED(pOperator);
H
Haojun Liao 已提交
3253 3254 3255
  return TSDB_CODE_SUCCESS;
}

3256
static SSDataBlock* doLoadRemoteData(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3257 3258
  SExchangeInfo* pExchangeInfo = pOperator->info;
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3259

3260
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
3261
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
3262 3263
    return NULL;
  }
3264

L
Liu Jicong 已提交
3265
  size_t               totalSources = taosArrayGetSize(pExchangeInfo->pSources);
H
Haojun Liao 已提交
3266
  SLoadRemoteDataInfo* pLoadInfo = &pExchangeInfo->loadInfo;
H
Haojun Liao 已提交
3267

3268
  if (pOperator->status == OP_EXEC_DONE) {
L
Liu Jicong 已提交
3269 3270 3271
    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);
3272 3273 3274 3275 3276 3277
    return NULL;
  }

  if (pExchangeInfo->seqLoadData) {
    return seqLoadRemoteData(pOperator);
  } else {
H
Haojun Liao 已提交
3278
    return concurrentlyLoadRemoteData(pOperator);
3279
  }
H
Haojun Liao 已提交
3280

3281
#if 0
H
Haojun Liao 已提交
3282
  _error:
wafwerar's avatar
wafwerar 已提交
3283 3284
  taosMemoryFreeClear(pMsg);
  taosMemoryFreeClear(pMsgSendInfo);
H
Haojun Liao 已提交
3285 3286 3287

  terrno = pTaskInfo->code;
  return NULL;
3288
#endif
H
Haojun Liao 已提交
3289
}
3290

H
Haojun Liao 已提交
3291
static int32_t initDataSource(int32_t numOfSources, SExchangeInfo* pInfo) {
3292
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
H
Haojun Liao 已提交
3293 3294
  if (pInfo->pSourceDataInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
3295 3296
  }

L
Liu Jicong 已提交
3297
  for (int32_t i = 0; i < numOfSources; ++i) {
3298
    SSourceDataInfo dataInfo = {0};
H
Haojun Liao 已提交
3299
    dataInfo.status = EX_SOURCE_DATA_NOT_READY;
L
Liu Jicong 已提交
3300 3301
    dataInfo.pEx = pInfo;
    dataInfo.index = i;
3302

H
Haojun Liao 已提交
3303 3304 3305 3306 3307 3308 3309 3310 3311 3312
    void* ret = taosArrayPush(pInfo->pSourceDataInfo, &dataInfo);
    if (ret == NULL) {
      taosArrayDestroy(pInfo->pSourceDataInfo);
      return TSDB_CODE_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
}

3313
SOperatorInfo* createExchangeOperatorInfo(void *pTransporter, const SNodeList* pSources, SSDataBlock* pBlock,
3314
                                          SExecTaskInfo* pTaskInfo) {
L
Liu Jicong 已提交
3315
  SExchangeInfo* pInfo = taosMemoryCalloc(1, sizeof(SExchangeInfo));
wafwerar's avatar
wafwerar 已提交
3316
  SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
3317 3318

  if (pInfo == NULL || pOperator == NULL) {
wafwerar's avatar
wafwerar 已提交
3319 3320
    taosMemoryFreeClear(pInfo);
    taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3321 3322
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
    return NULL;
H
Haojun Liao 已提交
3323 3324
  }

H
Haojun Liao 已提交
3325
  size_t numOfSources = LIST_LENGTH(pSources);
H
Haojun Liao 已提交
3326
  pInfo->pSources = taosArrayInit(numOfSources, sizeof(SDownstreamSourceNode));
H
Haojun Liao 已提交
3327 3328 3329
  pInfo->pSourceDataInfo = taosArrayInit(numOfSources, sizeof(SSourceDataInfo));
  if (pInfo->pSourceDataInfo == NULL || pInfo->pSources == NULL) {
    goto _error;
H
Haojun Liao 已提交
3330 3331
  }

L
Liu Jicong 已提交
3332 3333
  for (int32_t i = 0; i < numOfSources; ++i) {
    SNodeListNode* pNode = nodesListGetNode((SNodeList*)pSources, i);
H
Haojun Liao 已提交
3334 3335
    taosArrayPush(pInfo->pSources, pNode);
  }
3336

H
Haojun Liao 已提交
3337 3338 3339
  int32_t code = initDataSource(numOfSources, pInfo);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3340
  }
H
Haojun Liao 已提交
3341

dengyihao's avatar
dengyihao 已提交
3342
  pInfo->pResult = pBlock;
3343 3344 3345
  pInfo->seqLoadData = true;

  tsem_init(&pInfo->ready, 0, 0);
3346

dengyihao's avatar
dengyihao 已提交
3347
  pOperator->name = "ExchangeOperator";
X
Xiaoyu Wang 已提交
3348
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_EXCHANGE;
3349
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
3350 3351
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3352
  pOperator->numOfExprs = pBlock->info.numOfCols;
dengyihao's avatar
dengyihao 已提交
3353
  pOperator->pTaskInfo = pTaskInfo;
3354

L
Liu Jicong 已提交
3355 3356
  pOperator->fpSet = createOperatorFpSet(prepareLoadRemoteData, doLoadRemoteData, NULL, NULL,
                                         destroyExchangeOperatorInfo, NULL, NULL, NULL);
3357
  pInfo->pTransporter = pTransporter;
3358

3359
  return pOperator;
H
Haojun Liao 已提交
3360

L
Liu Jicong 已提交
3361
_error:
H
Haojun Liao 已提交
3362
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3363
    destroyExchangeOperatorInfo(pInfo, numOfSources);
H
Haojun Liao 已提交
3364 3365
  }

wafwerar's avatar
wafwerar 已提交
3366 3367
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
3368
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
H
Haojun Liao 已提交
3369
  return NULL;
3370 3371
}

dengyihao's avatar
dengyihao 已提交
3372 3373
static int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                                const char* pKey);
3374

3375
static void destroySortedMergeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
3376
  SSortedMergeOperatorInfo* pInfo = (SSortedMergeOperatorInfo*)param;
H
Haojun Liao 已提交
3377
  taosArrayDestroy(pInfo->pSortInfo);
3378 3379 3380
  taosArrayDestroy(pInfo->groupInfo);

  if (pInfo->pSortHandle != NULL) {
H
Haojun Liao 已提交
3381
    tsortDestroySortHandle(pInfo->pSortHandle);
3382 3383
  }

H
Haojun Liao 已提交
3384
  blockDataDestroy(pInfo->binfo.pRes);
H
Haojun Liao 已提交
3385
  cleanupAggSup(&pInfo->aggSup);
3386
}
H
Haojun Liao 已提交
3387

L
Liu Jicong 已提交
3388
static bool needToMerge(SSDataBlock* pBlock, SArray* groupInfo, char** buf, int32_t rowIndex) {
3389 3390 3391 3392
  size_t size = taosArrayGetSize(groupInfo);
  if (size == 0) {
    return true;
  }
3393

3394 3395
  for (int32_t i = 0; i < size; ++i) {
    int32_t* index = taosArrayGet(groupInfo, i);
3396

3397
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
L
Liu Jicong 已提交
3398
    bool             isNull = colDataIsNull(pColInfo, rowIndex, pBlock->info.rows, NULL);
3399

3400 3401 3402
    if ((isNull && buf[i] != NULL) || (!isNull && buf[i] == NULL)) {
      return false;
    }
3403

3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416
    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;
      }
3417 3418 3419
    }
  }

3420
  return 0;
3421 3422
}

L
Liu Jicong 已提交
3423 3424 3425
static void doMergeResultImpl(SSortedMergeOperatorInfo* pInfo, SqlFunctionCtx* pCtx, int32_t numOfExpr,
                              int32_t rowIndex) {
  for (int32_t j = 0; j < numOfExpr; ++j) {  // TODO set row index
3426
    pCtx[j].startRow = rowIndex;
3427 3428
  }

3429 3430
  for (int32_t j = 0; j < numOfExpr; ++j) {
    int32_t functionId = pCtx[j].functionId;
L
Liu Jicong 已提交
3431 3432 3433 3434 3435 3436 3437 3438 3439
    //    pCtx[j].fpSet->addInput(&pCtx[j]);

    //    if (functionId < 0) {
    //      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
    //      doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_MERGE);
    //    } else {
    //      assert(!TSDB_FUNC_IS_SCALAR(functionId));
    //      aAggs[functionId].mergeFunc(&pCtx[j]);
    //    }
3440
  }
3441
}
3442

L
Liu Jicong 已提交
3443 3444
static void doFinalizeResultImpl(SqlFunctionCtx* pCtx, int32_t numOfExpr) {
  for (int32_t j = 0; j < numOfExpr; ++j) {
3445 3446 3447 3448
    int32_t functionId = pCtx[j].functionId;
    //    if (functionId == FUNC_TAG_DUMMY || functionId == FUNC_TS_DUMMY) {
    //      continue;
    //    }
3449

3450 3451 3452 3453
    //    if (functionId < 0) {
    //      SUdfInfo* pUdfInfo = taosArrayGet(pInfo->udfInfo, -1 * functionId - 1);
    //      doInvokeUdf(pUdfInfo, &pCtx[j], 0, TSDB_UDF_FUNC_FINALIZE);
    //    } else {
dengyihao's avatar
dengyihao 已提交
3454
    //    pCtx[j].fpSet.finalize(&pCtx[j]);
3455 3456
  }
}
3457

3458
static bool saveCurrentTuple(char** rowColData, SArray* pColumnList, SSDataBlock* pBlock, int32_t rowIndex) {
L
Liu Jicong 已提交
3459
  int32_t size = (int32_t)taosArrayGetSize(pColumnList);
3460

L
Liu Jicong 已提交
3461 3462
  for (int32_t i = 0; i < size; ++i) {
    int32_t*         index = taosArrayGet(pColumnList, i);
3463
    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, *index);
H
Haojun Liao 已提交
3464

3465 3466 3467
    char* data = colDataGetData(pColInfo, rowIndex);
    memcpy(rowColData[i], data, colDataGetLength(pColInfo, rowIndex));
  }
3468

3469 3470
  return true;
}
3471

3472 3473
static void doMergeImpl(SOperatorInfo* pOperator, int32_t numOfExpr, SSDataBlock* pBlock) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
3474

3475
  SqlFunctionCtx* pCtx = pInfo->binfo.pCtx;
L
Liu Jicong 已提交
3476
  for (int32_t i = 0; i < pBlock->info.numOfCols; ++i) {
3477 3478
    pCtx[i].size = 1;
  }
3479

L
Liu Jicong 已提交
3480
  for (int32_t i = 0; i < pBlock->info.rows; ++i) {
3481 3482 3483 3484 3485 3486 3487 3488 3489
    if (!pInfo->hasGroupVal) {
      ASSERT(i == 0);
      doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
      pInfo->hasGroupVal = saveCurrentTuple(pInfo->groupVal, pInfo->groupInfo, pBlock, i);
    } else {
      if (needToMerge(pBlock, pInfo->groupInfo, pInfo->groupVal, i)) {
        doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
      } else {
        doFinalizeResultImpl(pCtx, numOfExpr);
3490 3491
        int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfExprs, NULL);
        //        setTagValueForMultipleRows(pCtx, pOperator->numOfExprs, numOfRows);
3492

3493
        // TODO check for available buffer;
H
Haojun Liao 已提交
3494

3495 3496 3497 3498 3499
        // next group info data
        pInfo->binfo.pRes->info.rows += numOfRows;
        for (int32_t j = 0; j < numOfExpr; ++j) {
          if (pCtx[j].functionId < 0) {
            continue;
3500
          }
3501

H
Haojun Liao 已提交
3502
          pCtx[j].fpSet.process(&pCtx[j]);
3503
        }
3504 3505 3506

        doMergeResultImpl(pInfo, pCtx, numOfExpr, i);
        pInfo->hasGroupVal = saveCurrentTuple(pInfo->groupVal, pInfo->groupInfo, pBlock, i);
H
Haojun Liao 已提交
3507
      }
3508 3509 3510 3511
    }
  }
}

3512 3513
static SSDataBlock* doMerge(SOperatorInfo* pOperator) {
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
L
Liu Jicong 已提交
3514
  SSortHandle*              pHandle = pInfo->pSortHandle;
3515

3516
  SSDataBlock* pDataBlock = createOneDataBlock(pInfo->binfo.pRes, false);
3517
  blockDataEnsureCapacity(pDataBlock, pOperator->resultInfo.capacity);
3518

L
Liu Jicong 已提交
3519
  while (1) {
3520
    blockDataCleanup(pDataBlock);
3521
    while (1) {
H
Haojun Liao 已提交
3522
      STupleHandle* pTupleHandle = tsortNextTuple(pHandle);
3523 3524
      if (pTupleHandle == NULL) {
        break;
3525
      }
3526

3527 3528
      // build datablock for merge for one group
      appendOneRowToDataBlock(pDataBlock, pTupleHandle);
3529
      if (pDataBlock->info.rows >= pOperator->resultInfo.capacity) {
3530 3531
        break;
      }
3532
    }
3533

3534 3535 3536
    if (pDataBlock->info.rows == 0) {
      break;
    }
3537

3538
    setInputDataBlock(pOperator, pInfo->binfo.pCtx, pDataBlock, TSDB_ORDER_ASC, MAIN_SCAN, true);
L
Liu Jicong 已提交
3539 3540
    //  updateOutputBuf(&pInfo->binfo, &pAggInfo->bufCapacity, pBlock->info.rows * pAggInfo->resultRowFactor,
    //  pOperator->pRuntimeEnv, true);
3541
    doMergeImpl(pOperator, pOperator->numOfExprs, pDataBlock);
3542 3543
    // flush to tuple store, and after all data have been handled, return to upstream node or sink node
  }
3544

3545 3546 3547
  doFinalizeResultImpl(pInfo->binfo.pCtx, pOperator->numOfExprs);
  int32_t numOfRows = getNumOfResult(pInfo->binfo.pCtx, pOperator->numOfExprs, NULL);
  //        setTagValueForMultipleRows(pCtx, pOperator->numOfExprs, numOfRows);
3548

3549
  // TODO check for available buffer;
3550

3551 3552
  // next group info data
  pInfo->binfo.pRes->info.rows += numOfRows;
L
Liu Jicong 已提交
3553
  return (pInfo->binfo.pRes->info.rows > 0) ? pInfo->binfo.pRes : NULL;
3554
}
3555

3556
static SSDataBlock* doSortedMerge(SOperatorInfo* pOperator) {
3557 3558
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
3559 3560
  }

L
Liu Jicong 已提交
3561
  SExecTaskInfo*            pTaskInfo = pOperator->pTaskInfo;
3562
  SSortedMergeOperatorInfo* pInfo = pOperator->info;
H
Haojun Liao 已提交
3563
  if (pOperator->status == OP_RES_TO_RETURN) {
3564
    return getSortedBlockData(pInfo->pSortHandle, pInfo->binfo.pRes, pOperator->resultInfo.capacity, NULL);
3565 3566
  }

3567
  int32_t numOfBufPage = pInfo->sortBufSize / pInfo->bufPageSize;
dengyihao's avatar
dengyihao 已提交
3568 3569
  pInfo->pSortHandle = tsortCreateSortHandle(pInfo->pSortInfo, NULL, SORT_MULTISOURCE_MERGE, pInfo->bufPageSize,
                                             numOfBufPage, pInfo->binfo.pRes, "GET_TASKID(pTaskInfo)");
H
Haojun Liao 已提交
3570

3571
  tsortSetFetchRawDataFp(pInfo->pSortHandle, loadNextDataBlock, NULL, NULL);
3572

L
Liu Jicong 已提交
3573
  for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
wmmhello's avatar
wmmhello 已提交
3574
    SSortSource* ps = taosMemoryCalloc(1, sizeof(SSortSource));
H
Haojun Liao 已提交
3575
    ps->param = pOperator->pDownstream[i];
H
Haojun Liao 已提交
3576
    tsortAddSource(pInfo->pSortHandle, ps);
3577 3578
  }

H
Haojun Liao 已提交
3579
  int32_t code = tsortOpen(pInfo->pSortHandle);
3580
  if (code != TSDB_CODE_SUCCESS) {
3581
    longjmp(pTaskInfo->env, terrno);
3582 3583
  }

H
Haojun Liao 已提交
3584
  pOperator->status = OP_RES_TO_RETURN;
3585
  return doMerge(pOperator);
3586
}
3587

L
Liu Jicong 已提交
3588 3589
static int32_t initGroupCol(SExprInfo* pExprInfo, int32_t numOfCols, SArray* pGroupInfo,
                            SSortedMergeOperatorInfo* pInfo) {
3590 3591
  if (pGroupInfo == NULL || taosArrayGetSize(pGroupInfo) == 0) {
    return 0;
H
Haojun Liao 已提交
3592 3593
  }

3594 3595 3596 3597 3598 3599 3600 3601
  int32_t len = 0;
  SArray* plist = taosArrayInit(3, sizeof(SColumn));
  pInfo->groupInfo = taosArrayInit(3, sizeof(int32_t));

  if (plist == NULL || pInfo->groupInfo == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  }

L
Liu Jicong 已提交
3602 3603
  size_t numOfGroupCol = taosArrayGetSize(pInfo->groupInfo);
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3604
    SColumn* pCol = taosArrayGet(pGroupInfo, i);
L
Liu Jicong 已提交
3605
    for (int32_t j = 0; j < numOfCols; ++j) {
H
Haojun Liao 已提交
3606
      SExprInfo* pe = &pExprInfo[j];
3607
      if (pe->base.resSchema.slotId == pCol->colId) {
3608 3609
        taosArrayPush(plist, pCol);
        taosArrayPush(pInfo->groupInfo, &j);
H
Haojun Liao 已提交
3610
        len += pCol->bytes;
3611 3612
        break;
      }
H
Haojun Liao 已提交
3613 3614 3615
    }
  }

3616
  ASSERT(taosArrayGetSize(pGroupInfo) == taosArrayGetSize(plist));
H
Haojun Liao 已提交
3617

wafwerar's avatar
wafwerar 已提交
3618
  pInfo->groupVal = taosMemoryCalloc(1, (POINTER_BYTES * numOfGroupCol + len));
3619 3620 3621 3622
  if (pInfo->groupVal == NULL) {
    taosArrayDestroy(plist);
    return TSDB_CODE_OUT_OF_MEMORY;
  }
H
Haojun Liao 已提交
3623

3624
  int32_t offset = 0;
L
Liu Jicong 已提交
3625 3626
  char*   start = (char*)(pInfo->groupVal + (POINTER_BYTES * numOfGroupCol));
  for (int32_t i = 0; i < numOfGroupCol; ++i) {
3627 3628
    pInfo->groupVal[i] = start + offset;
    SColumn* pCol = taosArrayGet(plist, i);
H
Haojun Liao 已提交
3629
    offset += pCol->bytes;
3630
  }
H
Haojun Liao 已提交
3631

3632
  taosArrayDestroy(plist);
H
Haojun Liao 已提交
3633

3634 3635
  return TSDB_CODE_SUCCESS;
}
H
Haojun Liao 已提交
3636

L
Liu Jicong 已提交
3637 3638 3639
SOperatorInfo* createSortedMergeOperatorInfo(SOperatorInfo** downstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
                                             int32_t num, SArray* pSortInfo, SArray* pGroupInfo,
                                             SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
3640
  SSortedMergeOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SSortedMergeOperatorInfo));
L
Liu Jicong 已提交
3641
  SOperatorInfo*            pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
3642
  if (pInfo == NULL || pOperator == NULL) {
3643
    goto _error;
3644
  }
H
Haojun Liao 已提交
3645

3646
  pInfo->binfo.pCtx = createSqlFunctionCtx(pExprInfo, num, &pInfo->binfo.rowCellInfoOffset);
3647
  initResultRowInfo(&pInfo->binfo.resultRowInfo, (int32_t)1);
H
Haojun Liao 已提交
3648

3649 3650 3651
  if (pInfo->binfo.pCtx == NULL || pInfo->binfo.pRes == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
3652

dengyihao's avatar
dengyihao 已提交
3653
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
3654
  int32_t code = doInitAggInfoSup(&pInfo->aggSup, pInfo->binfo.pCtx, num, keyBufSize, pTaskInfo->id.str);
3655 3656 3657
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3658

H
Haojun Liao 已提交
3659
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, num, pTaskInfo);
H
Haojun Liao 已提交
3660
  code = initGroupCol(pExprInfo, num, pGroupInfo, pInfo);
3661 3662 3663
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
H
Haojun Liao 已提交
3664

L
Liu Jicong 已提交
3665 3666 3667 3668 3669
  //  pInfo->resultRowFactor = (int32_t)(getRowNumForMultioutput(pRuntimeEnv->pQueryAttr,
  //      pRuntimeEnv->pQueryAttr->topBotQuery, false));
  pInfo->sortBufSize = 1024 * 16;  // 1MB
  pInfo->bufPageSize = 1024;
  pInfo->pSortInfo = pSortInfo;
H
Haojun Liao 已提交
3670

3671
  pOperator->resultInfo.capacity = blockDataGetCapacityInRow(pInfo->binfo.pRes, pInfo->bufPageSize);
H
Haojun Liao 已提交
3672

L
Liu Jicong 已提交
3673
  pOperator->name = "SortedMerge";
X
Xiaoyu Wang 已提交
3674
  // pOperator->operatorType = OP_SortedMerge;
3675
  pOperator->blocking = true;
L
Liu Jicong 已提交
3676 3677
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
3678
  pOperator->numOfExprs = num;
L
Liu Jicong 已提交
3679
  pOperator->pExpr = pExprInfo;
H
Haojun Liao 已提交
3680

L
Liu Jicong 已提交
3681
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
3682

3683 3684
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doSortedMerge, NULL, NULL, destroySortedMergeOperatorInfo,
                                         NULL, NULL, NULL);
3685 3686 3687
  code = appendDownstream(pOperator, downstream, numOfDownstream);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
3688
  }
H
Haojun Liao 已提交
3689

3690
  return pOperator;
H
Haojun Liao 已提交
3691

L
Liu Jicong 已提交
3692
_error:
3693
  if (pInfo != NULL) {
H
Haojun Liao 已提交
3694
    destroySortedMergeOperatorInfo(pInfo, num);
H
Haojun Liao 已提交
3695 3696
  }

wafwerar's avatar
wafwerar 已提交
3697 3698
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
3699 3700
  terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
  return NULL;
H
Haojun Liao 已提交
3701 3702
}

3703
int32_t getTableScanInfo(SOperatorInfo* pOperator, int32_t *order, int32_t* scanFlag) {
3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714
  // todo add more information about exchange operation
  if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_EXCHANGE) {
    *order = TSDB_ORDER_ASC;
    *scanFlag = MAIN_SCAN;
    return TSDB_CODE_SUCCESS;
  } else if (pOperator->operatorType == QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN) {
    STableScanInfo* pTableScanInfo = pOperator->info;
    *order = pTableScanInfo->cond.order;
    *scanFlag = pTableScanInfo->scanFlag;
    return TSDB_CODE_SUCCESS;
  } else {
H
Haojun Liao 已提交
3715
    if (pOperator->pDownstream == NULL || pOperator->pDownstream[0] == NULL) {
3716
      return TSDB_CODE_INVALID_PARA;
H
Haojun Liao 已提交
3717
    } else {
3718
      return getTableScanInfo(pOperator->pDownstream[0], order, scanFlag);
3719 3720 3721
    }
  }
}
3722 3723

// this is a blocking operator
L
Liu Jicong 已提交
3724
static int32_t doOpenAggregateOptr(SOperatorInfo* pOperator) {
H
Haojun Liao 已提交
3725 3726
  if (OPTR_IS_OPENED(pOperator)) {
    return TSDB_CODE_SUCCESS;
3727 3728
  }

H
Haojun Liao 已提交
3729
  SExecTaskInfo*    pTaskInfo = pOperator->pTaskInfo;
3730
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3731

dengyihao's avatar
dengyihao 已提交
3732
  SOptrBasicInfo* pInfo = &pAggInfo->binfo;
H
Haojun Liao 已提交
3733
  SOperatorInfo* downstream = pOperator->pDownstream[0];
3734

3735 3736 3737
  int32_t order = TSDB_ORDER_ASC;
  int32_t scanFlag = MAIN_SCAN;

H
Haojun Liao 已提交
3738
  while (1) {
H
Haojun Liao 已提交
3739
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
3740
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
3741
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
3742 3743 3744 3745 3746

    if (pBlock == NULL) {
      break;
    }

3747 3748 3749 3750
    int32_t code = getTableScanInfo(pOperator, &order, &scanFlag);
    if (code != TSDB_CODE_SUCCESS) {
      longjmp(pTaskInfo->env, code);
    }
3751

3752 3753
    // there is an scalar expression that needs to be calculated before apply the group aggregation.
    if (pAggInfo->pScalarExprInfo != NULL) {
3754
      code = projectApplyFunctions(pAggInfo->pScalarExprInfo, pBlock, pBlock, pAggInfo->pScalarCtx,
3755
                                   pAggInfo->numOfScalarExpr, NULL);
3756
      if (code != TSDB_CODE_SUCCESS) {
3757
        longjmp(pTaskInfo->env, code);
3758
      }
3759 3760
    }

3761
    // the pDataBlock are always the same one, no need to call this again
3762
    setExecutionContext(pOperator->numOfExprs, pBlock->info.groupId, pTaskInfo, pAggInfo);
3763
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, scanFlag, true);
3764 3765 3766 3767
    code = doAggregateImpl(pOperator, 0, pInfo->pCtx);
    if (code != 0) {
      longjmp(pTaskInfo->env, code);
    }
3768

dengyihao's avatar
dengyihao 已提交
3769
#if 0  // test for encode/decode result info
3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780
    if(pOperator->encodeResultRow){
      char *result = NULL;
      int32_t length = 0;
      SAggSupporter   *pSup = &pAggInfo->aggSup;
      pOperator->encodeResultRow(pOperator, pSup, pInfo, &result, &length);
      taosHashClear(pSup->pResultRowHashTable);
      pInfo->resultRowInfo.size = 0;
      pOperator->decodeResultRow(pOperator, pSup, pInfo, result, length);
      if(result){
        taosMemoryFree(result);
      }
3781
    }
3782
#endif
3783 3784
  }

H
Haojun Liao 已提交
3785
  closeAllResultRows(&pAggInfo->binfo.resultRowInfo);
3786 3787
  finalizeMultiTupleQueryResult(pOperator->numOfExprs, pAggInfo->aggSup.pResultBuf, &pAggInfo->binfo.resultRowInfo,
                                pAggInfo->binfo.rowCellInfoOffset);
H
Haojun Liao 已提交
3788

3789
  initGroupedResultInfo(&pAggInfo->groupResInfo, pAggInfo->aggSup.pResultRowHashTable, false);
H
Haojun Liao 已提交
3790 3791 3792 3793
  OPTR_SET_OPENED(pOperator);
  return TSDB_CODE_SUCCESS;
}

3794
static SSDataBlock* getAggregateResult(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
3795
  SAggOperatorInfo* pAggInfo = pOperator->info;
H
Haojun Liao 已提交
3796 3797 3798 3799 3800 3801
  SOptrBasicInfo*   pInfo = &pAggInfo->binfo;

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

L
Liu Jicong 已提交
3802
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
3803
  pTaskInfo->code = pOperator->fpSet._openFn(pOperator);
H
Haojun Liao 已提交
3804 3805 3806 3807
  if (pTaskInfo->code != TSDB_CODE_SUCCESS) {
    return NULL;
  }

H
Haojun Liao 已提交
3808
  blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
3809
  doBuildResultDatablock(pOperator, pInfo, &pAggInfo->groupResInfo, pAggInfo->aggSup.pResultBuf);
H
Haojun Liao 已提交
3810 3811 3812
  if (pInfo->pRes->info.rows == 0 || !hasRemainDataInCurrentGroup(&pAggInfo->groupResInfo)) {
    doSetOperatorCompleted(pOperator);
  }
3813

H
Haojun Liao 已提交
3814
  doSetOperatorCompleted(pOperator);
L
Liu Jicong 已提交
3815
  return (blockDataGetNumOfRows(pInfo->pRes) != 0) ? pInfo->pRes : NULL;
3816 3817
}

dengyihao's avatar
dengyihao 已提交
3818 3819
void aggEncodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char** result,
                        int32_t* length) {
wmmhello's avatar
wmmhello 已提交
3820
  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
3821
  size_t  keyLen = sizeof(uint64_t) * 2;  // estimate the key length
wmmhello's avatar
wmmhello 已提交
3822
  int32_t totalSize = sizeof(int32_t) + size * (sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize);
wafwerar's avatar
wafwerar 已提交
3823
  *result = taosMemoryCalloc(1, totalSize);
L
Liu Jicong 已提交
3824
  if (*result == NULL) {
3825
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
wmmhello's avatar
wmmhello 已提交
3826 3827 3828
  }
  *(int32_t*)(*result) = size;
  int32_t offset = sizeof(int32_t);
3829 3830

  // prepare memory
3831
  SResultRowPosition* pos = &pInfo->resultRowInfo.cur;
dengyihao's avatar
dengyihao 已提交
3832 3833
  void*               pPage = getBufPage(pSup->pResultBuf, pos->pageId);
  SResultRow*         pRow = (SResultRow*)((char*)pPage + pos->offset);
3834 3835 3836
  setBufPageDirty(pPage, true);
  releaseBufPage(pSup->pResultBuf, pPage);

dengyihao's avatar
dengyihao 已提交
3837
  void* pIter = taosHashIterate(pSup->pResultRowHashTable, NULL);
wmmhello's avatar
wmmhello 已提交
3838
  while (pIter) {
dengyihao's avatar
dengyihao 已提交
3839
    void*               key = taosHashGetKey(pIter, &keyLen);
3840
    SResultRowPosition* p1 = (SResultRowPosition*)pIter;
3841

dengyihao's avatar
dengyihao 已提交
3842
    pPage = (SFilePage*)getBufPage(pSup->pResultBuf, p1->pageId);
3843
    pRow = (SResultRow*)((char*)pPage + p1->offset);
3844 3845
    setBufPageDirty(pPage, true);
    releaseBufPage(pSup->pResultBuf, pPage);
wmmhello's avatar
wmmhello 已提交
3846 3847 3848

    // recalculate the result size
    int32_t realTotalSize = offset + sizeof(int32_t) + keyLen + sizeof(int32_t) + pSup->resultRowSize;
L
Liu Jicong 已提交
3849 3850 3851
    if (realTotalSize > totalSize) {
      char* tmp = taosMemoryRealloc(*result, realTotalSize);
      if (tmp == NULL) {
wmmhello's avatar
wmmhello 已提交
3852
        terrno = TSDB_CODE_OUT_OF_MEMORY;
wafwerar's avatar
wafwerar 已提交
3853
        taosMemoryFree(*result);
wmmhello's avatar
wmmhello 已提交
3854
        *result = NULL;
3855
        longjmp(pOperator->pTaskInfo->env, TSDB_CODE_OUT_OF_MEMORY);
L
Liu Jicong 已提交
3856
      } else {
wmmhello's avatar
wmmhello 已提交
3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868
        *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);
3869
    memcpy(*result + offset, pRow, pSup->resultRowSize);
wmmhello's avatar
wmmhello 已提交
3870 3871 3872 3873 3874
    offset += pSup->resultRowSize;

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

L
Liu Jicong 已提交
3875
  if (length) {
wmmhello's avatar
wmmhello 已提交
3876 3877 3878 3879 3880
    *length = offset;
  }
  return;
}

dengyihao's avatar
dengyihao 已提交
3881 3882
bool aggDecodeResultRow(SOperatorInfo* pOperator, SAggSupporter* pSup, SOptrBasicInfo* pInfo, char* result,
                        int32_t length) {
L
Liu Jicong 已提交
3883
  if (!result || length <= 0) {
wmmhello's avatar
wmmhello 已提交
3884 3885 3886 3887 3888 3889 3890
    return false;
  }

  //  int32_t size = taosHashGetSize(pSup->pResultRowHashTable);
  int32_t count = *(int32_t*)(result);

  int32_t offset = sizeof(int32_t);
L
Liu Jicong 已提交
3891
  while (count-- > 0 && length > offset) {
wmmhello's avatar
wmmhello 已提交
3892 3893 3894
    int32_t keyLen = *(int32_t*)(result + offset);
    offset += sizeof(int32_t);

L
Liu Jicong 已提交
3895 3896 3897
    uint64_t    tableGroupId = *(uint64_t*)(result + offset);
    SResultRow* resultRow = getNewResultRow_rv(pSup->pResultBuf, tableGroupId, pSup->resultRowSize);
    if (!resultRow) {
3898
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3899
    }
3900

wmmhello's avatar
wmmhello 已提交
3901
    // add a new result set for a new group
3902 3903
    SResultRowPosition pos = {.pageId = resultRow->pageId, .offset = resultRow->offset};
    taosHashPut(pSup->pResultRowHashTable, result + offset, keyLen, &pos, sizeof(SResultRowPosition));
wmmhello's avatar
wmmhello 已提交
3904 3905 3906

    offset += keyLen;
    int32_t valueLen = *(int32_t*)(result + offset);
L
Liu Jicong 已提交
3907
    if (valueLen != pSup->resultRowSize) {
3908
      longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3909 3910 3911 3912 3913 3914 3915 3916 3917 3918
    }
    offset += sizeof(int32_t);
    int32_t pageId = resultRow->pageId;
    int32_t pOffset = resultRow->offset;
    memcpy(resultRow, result + offset, valueLen);
    resultRow->pageId = pageId;
    resultRow->offset = pOffset;
    offset += valueLen;

    initResultRow(resultRow);
3919
    prepareResultListBuffer(&pInfo->resultRowInfo, pOperator->pTaskInfo->env);
dengyihao's avatar
dengyihao 已提交
3920
    //    pInfo->resultRowInfo.cur = pInfo->resultRowInfo.size;
L
Liu Jicong 已提交
3921 3922
    //    pInfo->resultRowInfo.pPosition[pInfo->resultRowInfo.size++] =
    //        (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
dengyihao's avatar
dengyihao 已提交
3923
    pInfo->resultRowInfo.cur = (SResultRowPosition){.pageId = resultRow->pageId, .offset = resultRow->offset};
wmmhello's avatar
wmmhello 已提交
3924 3925
  }

L
Liu Jicong 已提交
3926
  if (offset != length) {
3927
    longjmp(pOperator->pTaskInfo->env, TSDB_CODE_TSC_INVALID_INPUT);
wmmhello's avatar
wmmhello 已提交
3928 3929 3930 3931
  }
  return true;
}

3932 3933
enum {
  PROJECT_RETRIEVE_CONTINUE = 0x1,
L
Liu Jicong 已提交
3934
  PROJECT_RETRIEVE_DONE = 0x2,
3935 3936 3937 3938 3939
};

static int32_t handleLimitOffset(SOperatorInfo* pOperator, SSDataBlock* pBlock) {
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
L
Liu Jicong 已提交
3940
  SSDataBlock*          pRes = pInfo->pRes;
3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988

  if (pProjectInfo->curSOffset > 0) {
    if (pProjectInfo->groupId == 0) {  // it is the first group
      pProjectInfo->groupId = pBlock->info.groupId;
      blockDataCleanup(pInfo->pRes);
      return PROJECT_RETRIEVE_CONTINUE;
    } else if (pProjectInfo->groupId != pBlock->info.groupId) {
      pProjectInfo->curSOffset -= 1;

      // ignore data block in current group
      if (pProjectInfo->curSOffset > 0) {
        blockDataCleanup(pInfo->pRes);
        return PROJECT_RETRIEVE_CONTINUE;
      }
    }

    // set current group id of the project operator
    pProjectInfo->groupId = pBlock->info.groupId;
  }

  if (pProjectInfo->groupId != 0 && pProjectInfo->groupId != pBlock->info.groupId) {
    pProjectInfo->curGroupOutput += 1;
    if ((pProjectInfo->slimit.limit > 0) && (pProjectInfo->slimit.limit <= pProjectInfo->curGroupOutput)) {
      pOperator->status = OP_EXEC_DONE;
      blockDataCleanup(pRes);

      return PROJECT_RETRIEVE_DONE;
    }

    // reset the value for a new group data
    pProjectInfo->curOffset = 0;
    pProjectInfo->curOutput = 0;
  }

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

  // set current group id
  pProjectInfo->groupId = pBlock->info.groupId;

  if (pProjectInfo->curOffset >= pRes->info.rows) {
    pProjectInfo->curOffset -= pRes->info.rows;
    blockDataCleanup(pRes);
    return PROJECT_RETRIEVE_CONTINUE;
  } else if (pProjectInfo->curOffset < pRes->info.rows && pProjectInfo->curOffset > 0) {
    blockDataTrimFirstNRows(pRes, pProjectInfo->curOffset);
    pProjectInfo->curOffset = 0;
  }

3989 3990 3991
  // check for the limitation in each group
  if (pProjectInfo->limit.limit > 0 && pProjectInfo->curOutput + pRes->info.rows >= pProjectInfo->limit.limit) {
    pRes->info.rows = (int32_t)(pProjectInfo->limit.limit - pProjectInfo->curOutput);
3992 3993 3994 3995 3996

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

3997
    return PROJECT_RETRIEVE_DONE;
3998
  }
3999

4000
  // todo optimize performance
4001 4002
  // If there are slimit/soffset value exists, multi-round result can not be packed into one group, since the
  // they may not belong to the same group the limit/offset value is not valid in this case.
L
Liu Jicong 已提交
4003 4004
  if (pRes->info.rows >= pOperator->resultInfo.threshold || pProjectInfo->slimit.offset != -1 ||
      pProjectInfo->slimit.limit != -1) {
4005
    return PROJECT_RETRIEVE_DONE;
L
Liu Jicong 已提交
4006
  } else {  // not full enough, continue to accumulate the output data in the buffer.
4007 4008 4009 4010
    return PROJECT_RETRIEVE_CONTINUE;
  }
}

4011
static SSDataBlock* doProjectOperation(SOperatorInfo* pOperator) {
4012
  SProjectOperatorInfo* pProjectInfo = pOperator->info;
L
Liu Jicong 已提交
4013
  SOptrBasicInfo*       pInfo = &pProjectInfo->binfo;
4014 4015

  SSDataBlock* pRes = pInfo->pRes;
4016
  blockDataCleanup(pRes);
4017

4018
  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
4019 4020 4021
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }
dengyihao's avatar
dengyihao 已提交
4022

H
Haojun Liao 已提交
4023
#if 0
4024 4025 4026 4027 4028 4029
  if (pProjectInfo->existDataBlock) {  // TODO refactor
    SSDataBlock* pBlock = pProjectInfo->existDataBlock;
    pProjectInfo->existDataBlock = NULL;
    *newgroup = true;

    // todo dynamic set tags
L
Liu Jicong 已提交
4030
    //    if (pTableQueryInfo != NULL) {
4031
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
4032
    //    }
4033 4034

    // the pDataBlock are always the same one, no need to call this again
H
Haojun Liao 已提交
4035
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, TSDB_ORDER_ASC);
4036

H
Haojun Liao 已提交
4037
    blockDataEnsureCapacity(pInfo->pRes, pBlock->info.rows);
4038
    projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
4039
    if (pRes->info.rows >= pProjectInfo->binfo.capacity * 0.8) {
4040 4041
      copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfExprs);
      resetResultRowEntryResult(pInfo->pCtx, pOperator->numOfExprs);
4042 4043 4044
      return pRes;
    }
  }
H
Haojun Liao 已提交
4045
#endif
4046

4047 4048 4049
  int32_t order = 0;
  int32_t scanFlag = 0;

H
Haojun Liao 已提交
4050 4051
  SOperatorInfo* downstream = pOperator->pDownstream[0];

L
Liu Jicong 已提交
4052
  while (1) {
H
Haojun Liao 已提交
4053
    // The downstream exec may change the value of the newgroup, so use a local variable instead.
H
Haojun Liao 已提交
4054
    publishOperatorProfEvent(downstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4055
    SSDataBlock* pBlock = downstream->fpSet.getNextFn(downstream);
H
Haojun Liao 已提交
4056
    publishOperatorProfEvent(downstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4057 4058

    if (pBlock == NULL) {
4059
      setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
4060 4061 4062 4063
      break;
    }

    // Return result of the previous group in the firstly.
4064
    if (false) {
4065 4066 4067
      if (pRes->info.rows > 0) {
        pProjectInfo->existDataBlock = pBlock;
        break;
L
Liu Jicong 已提交
4068
      } else {  // init output buffer for a new group data
4069
        initCtxOutputBuffer(pInfo->pCtx, pOperator->numOfExprs);
4070 4071 4072
      }
    }

4073 4074
    // todo set tags

H
Haojun Liao 已提交
4075 4076
    //    STableQueryInfo* pTableQueryInfo = pRuntimeEnv->current;
    //    if (pTableQueryInfo != NULL) {
4077
    //      setTagValue(pOperator, pTableQueryInfo->pTable, pInfo->pCtx, pOperator->numOfExprs);
H
Haojun Liao 已提交
4078
    //    }
4079 4080

    // the pDataBlock are always the same one, no need to call this again
4081
    int32_t code = getTableScanInfo(pOperator->pDownstream[0], &order, &scanFlag);
4082

4083
    setInputDataBlock(pOperator, pInfo->pCtx, pBlock, order, scanFlag, false);
4084 4085
    blockDataEnsureCapacity(pInfo->pRes, pInfo->pRes->info.rows + pBlock->info.rows);

4086 4087 4088
    code = projectApplyFunctions(pOperator->pExpr, pInfo->pRes, pBlock, pInfo->pCtx, pOperator->numOfExprs, pProjectInfo->pPseudoColInfo);
    if (code != TSDB_CODE_SUCCESS) {
      longjmp(pTaskInfo->env, code);
4089 4090
    }

4091 4092
    int32_t status = handleLimitOffset(pOperator, pBlock);
    if (status == PROJECT_RETRIEVE_CONTINUE) {
H
Haojun Liao 已提交
4093
      continue;
L
Liu Jicong 已提交
4094
    } else if (status == PROJECT_RETRIEVE_DONE) {
4095 4096 4097
      break;
    }
  }
dengyihao's avatar
dengyihao 已提交
4098

H
Haojun Liao 已提交
4099
  pProjectInfo->curOutput += pInfo->pRes->info.rows;
H
Haojun Liao 已提交
4100

4101
  //  copyTsColoum(pRes, pInfo->pCtx, pOperator->numOfExprs);
L
Liu Jicong 已提交
4102
  return (pInfo->pRes->info.rows > 0) ? pInfo->pRes : NULL;
4103 4104
}

L
Liu Jicong 已提交
4105 4106
static void doHandleRemainBlockForNewGroupImpl(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                               SExecTaskInfo* pTaskInfo) {
4107
  pInfo->totalInputRows = pInfo->existNewGroupBlock->info.rows;
H
Haojun Liao 已提交
4108

L
Liu Jicong 已提交
4109 4110
  int64_t ekey = Q_STATUS_EQUAL(pTaskInfo->status, TASK_COMPLETED) ? pTaskInfo->window.ekey
                                                                   : pInfo->existNewGroupBlock->info.window.ekey;
4111 4112
  taosResetFillInfo(pInfo->pFillInfo, getFillInfoStart(pInfo->pFillInfo));

4113
  taosFillSetStartInfo(pInfo->pFillInfo, pInfo->existNewGroupBlock->info.rows, ekey);
4114 4115
  taosFillSetInputDataBlock(pInfo->pFillInfo, pInfo->existNewGroupBlock);

4116
  doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, pResultInfo->capacity);
4117 4118 4119 4120
  pInfo->existNewGroupBlock = NULL;
  *newgroup = true;
}

L
Liu Jicong 已提交
4121 4122
static void doHandleRemainBlockFromNewGroup(SFillOperatorInfo* pInfo, SResultInfo* pResultInfo, bool* newgroup,
                                            SExecTaskInfo* pTaskInfo) {
4123 4124
  if (taosFillHasMoreResults(pInfo->pFillInfo)) {
    *newgroup = false;
4125
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pInfo->pRes, (int32_t)pResultInfo->capacity);
H
Haojun Liao 已提交
4126
    if (pInfo->pRes->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult)) {
4127 4128 4129 4130 4131 4132
      return;
    }
  }

  // handle the cached new group data block
  if (pInfo->existNewGroupBlock) {
4133
    doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4134 4135 4136
  }
}

4137
static SSDataBlock* doFill(SOperatorInfo* pOperator) {
L
Liu Jicong 已提交
4138 4139
  SFillOperatorInfo* pInfo = pOperator->info;
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
4140

H
Haojun Liao 已提交
4141
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4142 4143 4144
  SSDataBlock* pResBlock = pInfo->pRes;

  blockDataCleanup(pResBlock);
4145 4146 4147 4148
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
  }

4149
  // todo handle different group data interpolation
X
Xiaoyu Wang 已提交
4150 4151
  bool  n = false;
  bool* newgroup = &n;
4152
  doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4153 4154
  if (pResBlock->info.rows > pResultInfo->threshold || (!pInfo->multigroupResult && pResBlock->info.rows > 0)) {
    return pResBlock;
H
Haojun Liao 已提交
4155
  }
4156

H
Haojun Liao 已提交
4157
  SOperatorInfo* pDownstream = pOperator->pDownstream[0];
L
Liu Jicong 已提交
4158
  while (1) {
H
Haojun Liao 已提交
4159
    publishOperatorProfEvent(pDownstream, QUERY_PROF_BEFORE_OPERATOR_EXEC);
4160
    SSDataBlock* pBlock = pDownstream->fpSet.getNextFn(pDownstream);
H
Haojun Liao 已提交
4161
    publishOperatorProfEvent(pDownstream, QUERY_PROF_AFTER_OPERATOR_EXEC);
4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172

    if (*newgroup) {
      assert(pBlock != NULL);
    }

    if (*newgroup && pInfo->totalInputRows > 0) {  // there are already processed current group data block
      pInfo->existNewGroupBlock = pBlock;
      *newgroup = false;

      // Fill the previous group data block, before handle the data block of new group.
      // Close the fill operation for previous group data block
4173
      taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4174 4175 4176 4177 4178 4179 4180
    } else {
      if (pBlock == NULL) {
        if (pInfo->totalInputRows == 0) {
          pOperator->status = OP_EXEC_DONE;
          return NULL;
        }

4181
        taosFillSetStartInfo(pInfo->pFillInfo, 0, pTaskInfo->window.ekey);
4182 4183 4184 4185 4186 4187 4188
      } else {
        pInfo->totalInputRows += pBlock->info.rows;
        taosFillSetStartInfo(pInfo->pFillInfo, pBlock->info.rows, pBlock->info.window.ekey);
        taosFillSetInputDataBlock(pInfo->pFillInfo, pBlock);
      }
    }

4189 4190
    blockDataEnsureCapacity(pResBlock, pOperator->resultInfo.capacity);
    doFillTimeIntervalGapsInResults(pInfo->pFillInfo, pResBlock, pOperator->resultInfo.capacity);
4191 4192

    // current group has no more result to return
4193
    if (pResBlock->info.rows > 0) {
4194 4195
      // 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
4196 4197
      if (pResBlock->info.rows > pResultInfo->threshold || pBlock == NULL || (!pInfo->multigroupResult)) {
        return pResBlock;
4198 4199
      }

4200
      doHandleRemainBlockFromNewGroup(pInfo, pResultInfo, newgroup, pTaskInfo);
4201 4202
      if (pResBlock->info.rows > pOperator->resultInfo.threshold || pBlock == NULL) {
        return pResBlock;
4203 4204 4205
      }
    } else if (pInfo->existNewGroupBlock) {  // try next group
      assert(pBlock != NULL);
4206
      doHandleRemainBlockForNewGroupImpl(pInfo, pResultInfo, newgroup, pTaskInfo);
4207 4208
      if (pResBlock->info.rows > pResultInfo->threshold) {
        return pResBlock;
4209 4210 4211 4212 4213 4214 4215 4216
      }
    } else {
      return NULL;
    }
  }
}

// todo set the attribute of query scan count
H
Haojun Liao 已提交
4217
static int32_t getNumOfScanTimes(STaskAttr* pQueryAttr) {
L
Liu Jicong 已提交
4218
  for (int32_t i = 0; i < pQueryAttr->numOfOutput; ++i) {
4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232
    int32_t functionId = getExprFunctionId(&pQueryAttr->pExpr1[i]);
    if (functionId == FUNCTION_STDDEV || functionId == FUNCTION_PERCT) {
      return 2;
    }
  }

  return 1;
}

static void destroyOperatorInfo(SOperatorInfo* pOperator) {
  if (pOperator == NULL) {
    return;
  }

4233
  if (pOperator->fpSet.closeFn != NULL) {
4234
    pOperator->fpSet.closeFn(pOperator->info, pOperator->numOfExprs);
4235 4236
  }

H
Haojun Liao 已提交
4237
  if (pOperator->pDownstream != NULL) {
L
Liu Jicong 已提交
4238
    for (int32_t i = 0; i < pOperator->numOfDownstream; ++i) {
H
Haojun Liao 已提交
4239
      destroyOperatorInfo(pOperator->pDownstream[i]);
4240 4241
    }

wafwerar's avatar
wafwerar 已提交
4242
    taosMemoryFreeClear(pOperator->pDownstream);
H
Haojun Liao 已提交
4243
    pOperator->numOfDownstream = 0;
4244 4245
  }

H
Haojun Liao 已提交
4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256
  if (pOperator->pExpr != NULL) {
    for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
      SExprInfo* pExprInfo = &pOperator->pExpr[i];
      if (pExprInfo->pExpr->nodeType == QUERY_NODE_COLUMN) {
        taosMemoryFree(pExprInfo->base.pParam[0].pCol);
      }
      taosMemoryFree(pExprInfo->base.pParam);
      taosMemoryFree(pExprInfo->pExpr);
    }
  }

D
fix bug  
dapan 已提交
4257
  taosMemoryFreeClear(pOperator->pExpr);
wafwerar's avatar
wafwerar 已提交
4258 4259
  taosMemoryFreeClear(pOperator->info);
  taosMemoryFreeClear(pOperator);
4260 4261
}

dengyihao's avatar
dengyihao 已提交
4262 4263
int32_t doInitAggInfoSup(SAggSupporter* pAggSup, SqlFunctionCtx* pCtx, int32_t numOfOutput, size_t keyBufSize,
                         const char* pKey) {
4264 4265
  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);

dengyihao's avatar
dengyihao 已提交
4266 4267
  pAggSup->resultRowSize = getResultRowSize(pCtx, numOfOutput);
  pAggSup->keyBuf = taosMemoryCalloc(1, keyBufSize + POINTER_BYTES + sizeof(int64_t));
4268 4269
  pAggSup->pResultRowHashTable = taosHashInit(10, hashFn, true, HASH_NO_LOCK);

H
Haojun Liao 已提交
4270
  if (pAggSup->keyBuf == NULL || pAggSup->pResultRowHashTable == NULL) {
4271 4272 4273
    return TSDB_CODE_OUT_OF_MEMORY;
  }

H
Haojun Liao 已提交
4274
  uint32_t defaultPgsz = 4096;
X
Xiaoyu Wang 已提交
4275
  while (defaultPgsz < pAggSup->resultRowSize * 4) {
H
Haojun Liao 已提交
4276 4277 4278 4279 4280 4281 4282 4283 4284 4285
    defaultPgsz <<= 1u;
  }

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

  int32_t code = createDiskbasedBuf(&pAggSup->pResultBuf, defaultPgsz, defaultBufsz, pKey, "/tmp/");
H
Haojun Liao 已提交
4286 4287 4288 4289
  if (code != TSDB_CODE_SUCCESS) {
    return code;
  }

4290 4291 4292
  return TSDB_CODE_SUCCESS;
}

4293
void cleanupAggSup(SAggSupporter* pAggSup) {
wafwerar's avatar
wafwerar 已提交
4294
  taosMemoryFreeClear(pAggSup->keyBuf);
4295
  taosHashCleanup(pAggSup->pResultRowHashTable);
H
Haojun Liao 已提交
4296
  destroyDiskbasedBuf(pAggSup->pResultBuf);
4297 4298
}

H
Haojun Liao 已提交
4299
int32_t initAggInfo(SOptrBasicInfo* pBasicInfo, SAggSupporter* pAggSup, SExprInfo* pExprInfo, int32_t numOfCols,
4300
                    SSDataBlock* pResultBlock, size_t keyBufSize, const char* pkey) {
4301
  pBasicInfo->pCtx = createSqlFunctionCtx(pExprInfo, numOfCols, &pBasicInfo->rowCellInfoOffset);
H
Haojun Liao 已提交
4302 4303
  pBasicInfo->pRes = pResultBlock;

4304
  doInitAggInfoSup(pAggSup, pBasicInfo->pCtx, numOfCols, keyBufSize, pkey);
4305

L
Liu Jicong 已提交
4306
  for (int32_t i = 0; i < numOfCols; ++i) {
4307 4308 4309
    pBasicInfo->pCtx[i].pBuf = pAggSup->pResultBuf;
  }

4310
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
4311 4312
}

4313 4314 4315 4316 4317 4318 4319 4320 4321
void initResultSizeInfo(SOperatorInfo* pOperator, int32_t numOfRows) {
  pOperator->resultInfo.capacity = numOfRows;
  pOperator->resultInfo.threshold = numOfRows * 0.75;

  if (pOperator->resultInfo.threshold == 0) {
    pOperator->resultInfo.capacity = numOfRows;
  }
}

H
Haojun Liao 已提交
4322
static STableQueryInfo* initTableQueryInfo(const STableGroupInfo* pTableGroupInfo) {
L
Liu Jicong 已提交
4323 4324 4325 4326
  if (pTableGroupInfo->numOfTables == 0) {
    return NULL;
  }

wafwerar's avatar
wafwerar 已提交
4327
  STableQueryInfo* pTableQueryInfo = taosMemoryCalloc(pTableGroupInfo->numOfTables, sizeof(STableQueryInfo));
H
Haojun Liao 已提交
4328 4329
  if (pTableQueryInfo == NULL) {
    return NULL;
H
Haojun Liao 已提交
4330
  }
H
Haojun Liao 已提交
4331 4332

  int32_t index = 0;
L
Liu Jicong 已提交
4333
  for (int32_t i = 0; i < taosArrayGetSize(pTableGroupInfo->pGroupList); ++i) {
H
Haojun Liao 已提交
4334
    SArray* pa = taosArrayGetP(pTableGroupInfo->pGroupList, i);
L
Liu Jicong 已提交
4335
    for (int32_t j = 0; j < taosArrayGetSize(pa); ++j) {
H
Haojun Liao 已提交
4336
      STableKeyInfo* pk = taosArrayGet(pa, j);
H
Haojun Liao 已提交
4337
      STableQueryInfo* pTQueryInfo = &pTableQueryInfo[index++];
L
Liu Jicong 已提交
4338
      pTQueryInfo->lastKey = pk->lastKey;
H
Haojun Liao 已提交
4339 4340
    }
  }
H
Haojun Liao 已提交
4341 4342

  STimeWindow win = {0, INT64_MAX};
4343
  createTableQueryInfo(pTableQueryInfo, win);
H
Haojun Liao 已提交
4344
  return pTableQueryInfo;
H
Haojun Liao 已提交
4345 4346
}

L
Liu Jicong 已提交
4347
SOperatorInfo* createAggregateOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t numOfCols,
4348
                                           SSDataBlock* pResultBlock, SExprInfo* pScalarExprInfo,
L
Liu Jicong 已提交
4349 4350
                                           int32_t numOfScalarExpr, SExecTaskInfo* pTaskInfo,
                                           const STableGroupInfo* pTableGroupInfo) {
wafwerar's avatar
wafwerar 已提交
4351
  SAggOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SAggOperatorInfo));
L
Liu Jicong 已提交
4352
  SOperatorInfo*    pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4353 4354 4355
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
H
Haojun Liao 已提交
4356

4357
  int32_t numOfRows = 10;
dengyihao's avatar
dengyihao 已提交
4358
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4359 4360

  initResultSizeInfo(pOperator, numOfRows);
dengyihao's avatar
dengyihao 已提交
4361 4362
  int32_t code =
      initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResultBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4363
  pInfo->pTableQueryInfo = initTableQueryInfo(pTableGroupInfo);
L
Liu Jicong 已提交
4364
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4365 4366
    goto _error;
  }
H
Haojun Liao 已提交
4367

H
Haojun Liao 已提交
4368 4369 4370 4371
  int32_t numOfGroup = 10;  // todo replaced with true value
  pInfo->groupId = INT32_MIN;
  initResultRowInfo(&pInfo->binfo.resultRowInfo, numOfGroup);

4372 4373
  pInfo->pScalarExprInfo = pScalarExprInfo;
  pInfo->numOfScalarExpr = numOfScalarExpr;
4374 4375 4376
  if (pInfo->pScalarExprInfo != NULL) {
    pInfo->pScalarCtx = createSqlFunctionCtx(pScalarExprInfo, numOfCols, &pInfo->rowCellInfoOffset);
  }
4377

dengyihao's avatar
dengyihao 已提交
4378
  pOperator->name = "TableAggregate";
X
Xiaoyu Wang 已提交
4379
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_AGG;
4380
  pOperator->blocking = true;
dengyihao's avatar
dengyihao 已提交
4381 4382 4383
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
4384
  pOperator->numOfExprs = numOfCols;
dengyihao's avatar
dengyihao 已提交
4385
  pOperator->pTaskInfo = pTaskInfo;
H
Haojun Liao 已提交
4386

4387 4388
  pOperator->fpSet = createOperatorFpSet(doOpenAggregateOptr, getAggregateResult, NULL, NULL, destroyAggOperatorInfo,
                                         aggEncodeResultRow, aggDecodeResultRow, NULL);
H
Haojun Liao 已提交
4389 4390 4391 4392 4393

  code = appendDownstream(pOperator, &downstream, 1);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4394 4395

  return pOperator;
L
Liu Jicong 已提交
4396
_error:
H
Haojun Liao 已提交
4397
  destroyAggOperatorInfo(pInfo, numOfCols);
wafwerar's avatar
wafwerar 已提交
4398 4399
  taosMemoryFreeClear(pInfo);
  taosMemoryFreeClear(pOperator);
H
Haojun Liao 已提交
4400 4401
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4402 4403
}

H
Haojun Liao 已提交
4404
void doDestroyBasicInfo(SOptrBasicInfo* pInfo, int32_t numOfOutput) {
4405 4406
  assert(pInfo != NULL);

4407
  destroySqlFunctionCtx(pInfo->pCtx, numOfOutput);
wafwerar's avatar
wafwerar 已提交
4408
  taosMemoryFreeClear(pInfo->rowCellInfoOffset);
4409 4410

  cleanupResultRowInfo(&pInfo->resultRowInfo);
H
Haojun Liao 已提交
4411
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
4412 4413
}

H
Haojun Liao 已提交
4414
void destroyBasicOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4415
  SOptrBasicInfo* pInfo = (SOptrBasicInfo*)param;
4416 4417
  doDestroyBasicInfo(pInfo, numOfOutput);
}
H
Haojun Liao 已提交
4418

H
Haojun Liao 已提交
4419
void destroyMergeJoinOperator(void* param, int32_t numOfOutput) {
X
Xiaoyu Wang 已提交
4420
  SJoinOperatorInfo* pJoinOperator = (SJoinOperatorInfo*)param;
H
Haojun Liao 已提交
4421 4422
}

H
Haojun Liao 已提交
4423
void destroyAggOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4424
  SAggOperatorInfo* pInfo = (SAggOperatorInfo*)param;
4425 4426
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
}
4427

H
Haojun Liao 已提交
4428
void destroySFillOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4429
  SFillOperatorInfo* pInfo = (SFillOperatorInfo*)param;
4430
  pInfo->pFillInfo = taosDestroyFillInfo(pInfo->pFillInfo);
H
Haojun Liao 已提交
4431
  pInfo->pRes = blockDataDestroy(pInfo->pRes);
wafwerar's avatar
wafwerar 已提交
4432
  taosMemoryFreeClear(pInfo->p);
4433 4434
}

H
Haojun Liao 已提交
4435
static void destroyProjectOperatorInfo(void* param, int32_t numOfOutput) {
D
fix bug  
dapan 已提交
4436 4437 4438
  if (NULL == param) {
    return;
  }
L
Liu Jicong 已提交
4439
  SProjectOperatorInfo* pInfo = (SProjectOperatorInfo*)param;
4440
  doDestroyBasicInfo(&pInfo->binfo, numOfOutput);
H
Haojun Liao 已提交
4441
  cleanupAggSup(&pInfo->aggSup);
H
Haojun Liao 已提交
4442
  taosArrayDestroy(pInfo->pPseudoColInfo);
4443 4444
}

H
Haojun Liao 已提交
4445
void destroyExchangeOperatorInfo(void* param, int32_t numOfOutput) {
L
Liu Jicong 已提交
4446
  SExchangeInfo* pExInfo = (SExchangeInfo*)param;
H
Haojun Liao 已提交
4447 4448 4449 4450 4451 4452 4453 4454 4455
  taosArrayDestroy(pExInfo->pSources);
  taosArrayDestroy(pExInfo->pSourceDataInfo);
  if (pExInfo->pResult != NULL) {
    blockDataDestroy(pExInfo->pResult);
  }

  tsem_destroy(&pExInfo->ready);
}

H
Haojun Liao 已提交
4456 4457
static SArray* setRowTsColumnOutputInfo(SqlFunctionCtx* pCtx, int32_t numOfCols) {
  SArray* pList = taosArrayInit(4, sizeof(int32_t));
dengyihao's avatar
dengyihao 已提交
4458
  for (int32_t i = 0; i < numOfCols; ++i) {
H
Haojun Liao 已提交
4459 4460 4461 4462 4463 4464 4465 4466
    if (fmIsPseudoColumnFunc(pCtx[i].functionId)) {
      taosArrayPush(pList, &i);
    }
  }

  return pList;
}

L
Liu Jicong 已提交
4467
SOperatorInfo* createProjectOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExprInfo, int32_t num,
dengyihao's avatar
dengyihao 已提交
4468 4469
                                         SSDataBlock* pResBlock, SLimit* pLimit, SLimit* pSlimit,
                                         SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4470
  SProjectOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SProjectOperatorInfo));
L
Liu Jicong 已提交
4471
  SOperatorInfo*        pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4472 4473 4474
  if (pInfo == NULL || pOperator == NULL) {
    goto _error;
  }
4475

dengyihao's avatar
dengyihao 已提交
4476 4477 4478
  pInfo->limit = *pLimit;
  pInfo->slimit = *pSlimit;
  pInfo->curOffset = pLimit->offset;
H
Haojun Liao 已提交
4479 4480
  pInfo->curSOffset = pSlimit->offset;

H
Haojun Liao 已提交
4481
  pInfo->binfo.pRes = pResBlock;
H
Haojun Liao 已提交
4482 4483 4484

  int32_t numOfCols = num;
  int32_t numOfRows = 4096;
dengyihao's avatar
dengyihao 已提交
4485
  size_t  keyBufSize = sizeof(int64_t) + sizeof(int64_t) + POINTER_BYTES;
4486 4487 4488

  initResultSizeInfo(pOperator, numOfRows);
  initAggInfo(&pInfo->binfo, &pInfo->aggSup, pExprInfo, numOfCols, pResBlock, keyBufSize, pTaskInfo->id.str);
H
Haojun Liao 已提交
4489
  setFunctionResultOutput(&pInfo->binfo, &pInfo->aggSup, MAIN_SCAN, numOfCols, pTaskInfo);
H
Haojun Liao 已提交
4490
  pInfo->pPseudoColInfo = setRowTsColumnOutputInfo(pInfo->binfo.pCtx, numOfCols);
4491

dengyihao's avatar
dengyihao 已提交
4492
  pOperator->name = "ProjectOperator";
H
Haojun Liao 已提交
4493
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_PROJECT;
4494
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4495 4496 4497
  pOperator->status = OP_NOT_OPENED;
  pOperator->info = pInfo;
  pOperator->pExpr = pExprInfo;
4498
  pOperator->numOfExprs = num;
4499

L
Liu Jicong 已提交
4500 4501
  pOperator->fpSet = createOperatorFpSet(operatorDummyOpenFn, doProjectOperation, NULL, NULL,
                                         destroyProjectOperatorInfo, NULL, NULL, NULL);
L
Liu Jicong 已提交
4502 4503

  pOperator->pTaskInfo = pTaskInfo;
4504
  int32_t code = appendDownstream(pOperator, &downstream, 1);
H
Haojun Liao 已提交
4505
  if (code != TSDB_CODE_SUCCESS) {
H
Haojun Liao 已提交
4506 4507
    goto _error;
  }
4508 4509

  return pOperator;
H
Haojun Liao 已提交
4510

L
Liu Jicong 已提交
4511
_error:
H
Haojun Liao 已提交
4512 4513
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
4514 4515
}

4516
static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t numOfCols, SNodeListNode* pValNode,
L
Liu Jicong 已提交
4517
                            STimeWindow win, int32_t capacity, const char* id, SInterval* pInterval, int32_t fillType) {
4518
  SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pValNode);
H
Haojun Liao 已提交
4519 4520

  STimeWindow w = TSWINDOW_INITIALIZER;
4521
  getAlignQueryTimeWindow(pInterval, pInterval->precision, win.skey, &w);
H
Haojun Liao 已提交
4522 4523

  int32_t order = TSDB_ORDER_ASC;
4524
  pInfo->pFillInfo = taosCreateFillInfo(order, w.skey, 0, capacity, numOfCols, pInterval, fillType, pColInfo, id);
H
Haojun Liao 已提交
4525

wafwerar's avatar
wafwerar 已提交
4526
  pInfo->p = taosMemoryCalloc(numOfCols, POINTER_BYTES);
H
Haojun Liao 已提交
4527 4528 4529 4530 4531 4532 4533
  if (pInfo->pFillInfo == NULL || pInfo->p == NULL) {
    return TSDB_CODE_OUT_OF_MEMORY;
  } else {
    return TSDB_CODE_SUCCESS;
  }
}

L
Liu Jicong 已提交
4534
SOperatorInfo* createFillOperatorInfo(SOperatorInfo* downstream, SExprInfo* pExpr, int32_t numOfCols,
X
Xiaoyu Wang 已提交
4535 4536 4537
                                      SInterval* pInterval, STimeWindow* pWindow, SSDataBlock* pResBlock,
                                      int32_t fillType, SNodeListNode* pValueNode, bool multigroupResult,
                                      SExecTaskInfo* pTaskInfo) {
wafwerar's avatar
wafwerar 已提交
4538
  SFillOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SFillOperatorInfo));
L
Liu Jicong 已提交
4539
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
H
Haojun Liao 已提交
4540

L
Liu Jicong 已提交
4541
  pInfo->pRes = pResBlock;
4542 4543
  pInfo->multigroupResult = multigroupResult;

4544 4545
  int32_t type = TSDB_FILL_NONE;
  switch (fillType) {
dengyihao's avatar
dengyihao 已提交
4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563
    case FILL_MODE_PREV:
      type = TSDB_FILL_PREV;
      break;
    case FILL_MODE_NONE:
      type = TSDB_FILL_NONE;
      break;
    case FILL_MODE_NULL:
      type = TSDB_FILL_NULL;
      break;
    case FILL_MODE_NEXT:
      type = TSDB_FILL_NEXT;
      break;
    case FILL_MODE_VALUE:
      type = TSDB_FILL_SET_VALUE;
      break;
    case FILL_MODE_LINEAR:
      type = TSDB_FILL_LINEAR;
      break;
4564 4565 4566 4567
    default:
      type = TSDB_FILL_NONE;
  }

H
Haojun Liao 已提交
4568
  SResultInfo* pResultInfo = &pOperator->resultInfo;
4569 4570
  initResultSizeInfo(pOperator, 4096);

X
Xiaoyu Wang 已提交
4571 4572
  int32_t code = initFillInfo(pInfo, pExpr, numOfCols, pValueNode, *pWindow, pResultInfo->capacity, pTaskInfo->id.str,
                              pInterval, type);
4573 4574 4575
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }
4576

dengyihao's avatar
dengyihao 已提交
4577
  pOperator->name = "FillOperator";
4578
  pOperator->blocking = false;
dengyihao's avatar
dengyihao 已提交
4579
  pOperator->status = OP_NOT_OPENED;
4580
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_FILL;
dengyihao's avatar
dengyihao 已提交
4581
  pOperator->pExpr = pExpr;
4582
  pOperator->numOfExprs = numOfCols;
dengyihao's avatar
dengyihao 已提交
4583
  pOperator->info = pInfo;
H
Haojun Liao 已提交
4584

L
Liu Jicong 已提交
4585 4586
  pOperator->fpSet =
      createOperatorFpSet(operatorDummyOpenFn, doFill, NULL, NULL, destroySFillOperatorInfo, NULL, NULL, NULL);
4587
  pOperator->pTaskInfo = pTaskInfo;
4588
  code = appendDownstream(pOperator, &downstream, 1);
4589
  return pOperator;
H
Haojun Liao 已提交
4590

L
Liu Jicong 已提交
4591
_error:
wafwerar's avatar
wafwerar 已提交
4592 4593
  taosMemoryFreeClear(pOperator);
  taosMemoryFreeClear(pInfo);
H
Haojun Liao 已提交
4594
  return NULL;
4595 4596
}

L
Liu Jicong 已提交
4597 4598
static SResSchema createResSchema(int32_t type, int32_t bytes, int32_t slotId, int32_t scale, int32_t precision,
                                  const char* name) {
H
Haojun Liao 已提交
4599
  SResSchema s = {0};
dengyihao's avatar
dengyihao 已提交
4600 4601 4602 4603
  s.scale = scale;
  s.type = type;
  s.bytes = bytes;
  s.slotId = slotId;
H
Haojun Liao 已提交
4604
  s.precision = precision;
H
Haojun Liao 已提交
4605 4606 4607 4608
  strncpy(s.name, name, tListLen(s.name));

  return s;
}
H
Haojun Liao 已提交
4609

4610
static SColumn* createColumn(int32_t blockId, int32_t slotId, int32_t colId, SDataType* pType) {
H
Haojun Liao 已提交
4611 4612 4613 4614 4615 4616
  SColumn* pCol = taosMemoryCalloc(1, sizeof(SColumn));
  if (pCol == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

dengyihao's avatar
dengyihao 已提交
4617
  pCol->slotId = slotId;
4618 4619 4620 4621
  pCol->colId  = colId;
  pCol->bytes  = pType->bytes;
  pCol->type   = pType->type;
  pCol->scale  = pType->scale;
dengyihao's avatar
dengyihao 已提交
4622
  pCol->precision = pType->precision;
H
Haojun Liao 已提交
4623 4624 4625 4626 4627
  pCol->dataBlockId = blockId;

  return pCol;
}

H
Haojun Liao 已提交
4628
SExprInfo* createExprInfo(SNodeList* pNodeList, SNodeList* pGroupKeys, int32_t* numOfExprs) {
H
Haojun Liao 已提交
4629
  int32_t numOfFuncs = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
4630 4631 4632 4633
  int32_t numOfGroupKeys = 0;
  if (pGroupKeys != NULL) {
    numOfGroupKeys = LIST_LENGTH(pGroupKeys);
  }
H
Haojun Liao 已提交
4634

H
Haojun Liao 已提交
4635
  *numOfExprs = numOfFuncs + numOfGroupKeys;
wafwerar's avatar
wafwerar 已提交
4636
  SExprInfo* pExprs = taosMemoryCalloc(*numOfExprs, sizeof(SExprInfo));
H
Haojun Liao 已提交
4637

L
Liu Jicong 已提交
4638
  for (int32_t i = 0; i < (*numOfExprs); ++i) {
H
Haojun Liao 已提交
4639 4640 4641 4642 4643 4644
    STargetNode* pTargetNode = NULL;
    if (i < numOfFuncs) {
      pTargetNode = (STargetNode*)nodesListGetNode(pNodeList, i);
    } else {
      pTargetNode = (STargetNode*)nodesListGetNode(pGroupKeys, i - numOfFuncs);
    }
H
Haojun Liao 已提交
4645

4646
    SExprInfo* pExp = &pExprs[i];
H
Haojun Liao 已提交
4647

wafwerar's avatar
wafwerar 已提交
4648
    pExp->pExpr = taosMemoryCalloc(1, sizeof(tExprNode));
H
Haojun Liao 已提交
4649
    pExp->pExpr->_function.num = 1;
H
Haojun Liao 已提交
4650
    pExp->pExpr->_function.functionId = -1;
H
Haojun Liao 已提交
4651

4652
    int32_t type = nodeType(pTargetNode->pExpr);
H
Haojun Liao 已提交
4653
    // it is a project query, or group by column
4654
    if (type == QUERY_NODE_COLUMN) {
H
Haojun Liao 已提交
4655
      pExp->pExpr->nodeType = QUERY_NODE_COLUMN;
L
Liu Jicong 已提交
4656
      SColumnNode* pColNode = (SColumnNode*)pTargetNode->pExpr;
H
Haojun Liao 已提交
4657

G
Ganlin Zhao 已提交
4658 4659 4660
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

H
Haojun Liao 已提交
4661
      SDataType* pType = &pColNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4662 4663
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pColNode->colName);
4664
      pExp->base.pParam[0].pCol = createColumn(pColNode->dataBlockId, pColNode->slotId, pColNode->colId, pType);
H
Haojun Liao 已提交
4665
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_COLUMN;
4666
    } else if (type == QUERY_NODE_VALUE) {
4667 4668 4669 4670 4671 4672 4673
      pExp->pExpr->nodeType = QUERY_NODE_VALUE;
      SValueNode* pValNode = (SValueNode*)pTargetNode->pExpr;

      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

      SDataType* pType = &pValNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4674 4675
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pValNode->node.aliasName);
4676 4677
      pExp->base.pParam[0].type = FUNC_PARAM_TYPE_VALUE;
      valueNodeToVariant(pValNode, &pExp->base.pParam[0].param);
4678
    } else if (type == QUERY_NODE_FUNCTION) {
H
Haojun Liao 已提交
4679
      pExp->pExpr->nodeType = QUERY_NODE_FUNCTION;
H
Haojun Liao 已提交
4680 4681 4682
      SFunctionNode* pFuncNode = (SFunctionNode*)pTargetNode->pExpr;

      SDataType* pType = &pFuncNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4683 4684
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pFuncNode->node.aliasName);
H
Haojun Liao 已提交
4685

H
Haojun Liao 已提交
4686
      pExp->pExpr->_function.functionId = pFuncNode->funcId;
H
Haojun Liao 已提交
4687
      pExp->pExpr->_function.pFunctNode = pFuncNode;
4688

dengyihao's avatar
dengyihao 已提交
4689 4690
      strncpy(pExp->pExpr->_function.functionName, pFuncNode->functionName,
              tListLen(pExp->pExpr->_function.functionName));
4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703
#if 1
      // todo refactor: add the parameter for tbname function
      if (strcmp(pExp->pExpr->_function.functionName, "tbname") == 0) {
        pFuncNode->pParameterList = nodesMakeList();
        ASSERT(LIST_LENGTH(pFuncNode->pParameterList) == 0);
        SValueNode *res = (SValueNode *)nodesMakeNode(QUERY_NODE_VALUE);
        if (NULL == res) { // todo handle error
        } else {
          res->node.resType = (SDataType) {.bytes = sizeof(int64_t), .type = TSDB_DATA_TYPE_BIGINT};
          nodesListAppend(pFuncNode->pParameterList, res);
        }
      }
#endif
H
Haojun Liao 已提交
4704 4705

      int32_t numOfParam = LIST_LENGTH(pFuncNode->pParameterList);
G
Ganlin Zhao 已提交
4706 4707 4708 4709

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

H
Haojun Liao 已提交
4710
      for (int32_t j = 0; j < numOfParam; ++j) {
4711
        SNode* p1 = nodesListGetNode(pFuncNode->pParameterList, j);
G
Ganlin Zhao 已提交
4712
        if (p1->type == QUERY_NODE_COLUMN) {
dengyihao's avatar
dengyihao 已提交
4713
          SColumnNode* pcn = (SColumnNode*)p1;
G
Ganlin Zhao 已提交
4714 4715

          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_COLUMN;
4716
          pExp->base.pParam[j].pCol = createColumn(pcn->dataBlockId, pcn->slotId, pcn->colId, &pcn->node.resType);
G
Ganlin Zhao 已提交
4717 4718 4719
        } else if (p1->type == QUERY_NODE_VALUE) {
          SValueNode* pvn = (SValueNode*)p1;
          pExp->base.pParam[j].type = FUNC_PARAM_TYPE_VALUE;
4720
          valueNodeToVariant(pvn, &pExp->base.pParam[j].param);
G
Ganlin Zhao 已提交
4721
        }
H
Haojun Liao 已提交
4722
      }
4723
    } else if (type == QUERY_NODE_OPERATOR) {
H
Haojun Liao 已提交
4724
      pExp->pExpr->nodeType = QUERY_NODE_OPERATOR;
L
Liu Jicong 已提交
4725
      SOperatorNode* pNode = (SOperatorNode*)pTargetNode->pExpr;
4726

G
Ganlin Zhao 已提交
4727 4728 4729
      pExp->base.pParam = taosMemoryCalloc(1, sizeof(SFunctParam));
      pExp->base.numOfParams = 1;

4730
      SDataType* pType = &pNode->node.resType;
dengyihao's avatar
dengyihao 已提交
4731 4732
      pExp->base.resSchema = createResSchema(pType->type, pType->bytes, pTargetNode->slotId, pType->scale,
                                             pType->precision, pNode->node.aliasName);
4733 4734 4735
      pExp->pExpr->_optrRoot.pRootNode = pTargetNode->pExpr;
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4736 4737 4738
    }
  }

H
Haojun Liao 已提交
4739
  return pExprs;
H
Haojun Liao 已提交
4740 4741
}

4742
static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPTR_EXEC_MODEL model) {
wafwerar's avatar
wafwerar 已提交
4743
  SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
4744
  setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
H
Haojun Liao 已提交
4745

4746
  pTaskInfo->cost.created = taosGetTimestampMs();
H
Haojun Liao 已提交
4747
  pTaskInfo->id.queryId = queryId;
dengyihao's avatar
dengyihao 已提交
4748
  pTaskInfo->execModel = model;
H
Haojun Liao 已提交
4749

wafwerar's avatar
wafwerar 已提交
4750
  char* p = taosMemoryCalloc(1, 128);
L
Liu Jicong 已提交
4751
  snprintf(p, 128, "TID:0x%" PRIx64 " QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
4752
  pTaskInfo->id.str = p;
H
Haojun Liao 已提交
4753

4754 4755
  return pTaskInfo;
}
H
Haojun Liao 已提交
4756

L
Liu Jicong 已提交
4757 4758
static tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                                      STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId);
H
Haojun Liao 已提交
4759

L
Liu Jicong 已提交
4760 4761
static int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                                  uint64_t queryId, uint64_t taskId);
4762
static SArray* extractTableIdList(const STableGroupInfo* pTableGroupInfo);
H
Haojun Liao 已提交
4763
static SArray* extractColumnInfo(SNodeList* pNodeList);
4764

4765
static SArray* createSortInfo(SNodeList* pNodeList);
4766
static SArray* extractPartitionColInfo(SNodeList* pNodeList);
4767
static void    setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode);
4768

H
Haojun Liao 已提交
4769
SOperatorInfo* createOperatorTree(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SReadHandle* pHandle,
4770
                                  uint64_t queryId, uint64_t taskId, STableGroupInfo* pTableGroupInfo) {
4771 4772
  int32_t type = nodeType(pPhyNode);

X
Xiaoyu Wang 已提交
4773
  if (pPhyNode->pChildren == NULL || LIST_LENGTH(pPhyNode->pChildren) == 0) {
H
Haojun Liao 已提交
4774
    if (QUERY_NODE_PHYSICAL_PLAN_TABLE_SCAN == type) {
dengyihao's avatar
dengyihao 已提交
4775
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4776

H
Haojun Liao 已提交
4777
      tsdbReaderT pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableGroupInfo, (uint64_t)queryId, taskId);
4778
      if (pDataReader == NULL && terrno != 0) {
4779 4780
        return NULL;
      }
4781

4782
      SOperatorInfo* pOperator = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
4783

4784 4785
      STableScanInfo* pScanInfo = pOperator->info;
      pTaskInfo->cost.pRecoder = &pScanInfo->readRecorder;
4786

4787
      return pOperator;
H
Haojun Liao 已提交
4788
    } else if (QUERY_NODE_PHYSICAL_PLAN_EXCHANGE == type) {
H
Haojun Liao 已提交
4789
      SExchangePhysiNode* pExchange = (SExchangePhysiNode*)pPhyNode;
4790
      SSDataBlock*        pResBlock = createResDataBlock(pExchange->node.pOutputDataBlockDesc);
4791
      return createExchangeOperatorInfo(pHandle->pMsgCb->clientRpc, pExchange->pSrcEndPoints, pResBlock, pTaskInfo);
H
Haojun Liao 已提交
4792
    } else if (QUERY_NODE_PHYSICAL_PLAN_STREAM_SCAN == type) {
H
Haojun Liao 已提交
4793
      SScanPhysiNode* pScanPhyNode = (SScanPhysiNode*)pPhyNode;  // simple child table.
5
54liuyao 已提交
4794
      STableScanPhysiNode* pTableScanNode = (STableScanPhysiNode*)pPhyNode;
4795

4796
      int32_t numOfCols = 0;
5
54liuyao 已提交
4797 4798 4799 4800 4801 4802 4803 4804

      tsdbReaderT pDataReader = NULL;
      if (pHandle->vnode) {
        pDataReader = doCreateDataReader(pTableScanNode, pHandle, pTableGroupInfo, (uint64_t)queryId, taskId);
      } else {
        doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo, 
            queryId, taskId);
      }
4805

5
54liuyao 已提交
4806 4807 4808 4809 4810 4811
      if (pDataReader == NULL && terrno != 0) {
        qDebug("pDataReader is     NULL");
        // return NULL;
      } else {
        qDebug("pDataReader is not NULL");
      }
H
Haojun Liao 已提交
4812

4813
      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;
4814
      SOperatorInfo* pOperatorDumy = createTableScanOperatorInfo(pTableScanNode, pDataReader, pHandle, pTaskInfo);
5
54liuyao 已提交
4815 4816

      SArray* tableIdList = extractTableIdList(pTableGroupInfo);
X
Xiaoyu Wang 已提交
4817
      SSDataBlock*        pResBlock = createResDataBlock(pDescNode);
4818

X
Xiaoyu Wang 已提交
4819
      SArray*        pCols = extractColMatchInfo(pScanPhyNode->pScanCols, pDescNode, &numOfCols, COL_MATCH_FROM_COL_ID);
5
54liuyao 已提交
4820
      SOperatorInfo* pOperator = createStreamScanOperatorInfo(pHandle->reader, pDataReader, pResBlock, pCols, tableIdList, pTaskInfo,
5
54liuyao 已提交
4821
                                                              pScanPhyNode->node.pConditions, pOperatorDumy);
4822
      taosArrayDestroy(tableIdList);
H
Haojun Liao 已提交
4823
      return pOperator;
H
Haojun Liao 已提交
4824
    } else if (QUERY_NODE_PHYSICAL_PLAN_SYSTABLE_SCAN == type) {
L
Liu Jicong 已提交
4825
      SSystemTableScanPhysiNode* pSysScanPhyNode = (SSystemTableScanPhysiNode*)pPhyNode;
L
Liu Jicong 已提交
4826
      SScanPhysiNode*            pScanNode = &pSysScanPhyNode->scan;
H
Haojun Liao 已提交
4827

4828 4829 4830
      SDataBlockDescNode* pDescNode = pScanNode->node.pOutputDataBlockDesc;

      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4831

4832
      int32_t numOfOutputCols = 0;
4833
      SArray* colList = extractColMatchInfo(pScanNode->pScanCols, pDescNode, &numOfOutputCols, COL_MATCH_FROM_COL_ID);
L
Liu Jicong 已提交
4834
      SOperatorInfo* pOperator = createSysTableScanOperatorInfo(
L
Liu Jicong 已提交
4835 4836
          pHandle, pResBlock, &pScanNode->tableName, pScanNode->node.pConditions, pSysScanPhyNode->mgmtEpSet, colList,
          pTaskInfo, pSysScanPhyNode->showRewrite, pSysScanPhyNode->accountId);
H
Haojun Liao 已提交
4837
      return pOperator;
4838
    } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
X
Xiaoyu Wang 已提交
4839
      STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
4840 4841 4842 4843

      SDataBlockDescNode* pDescNode = pScanPhyNode->node.pOutputDataBlockDesc;

      SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4844

X
Xiaoyu Wang 已提交
4845 4846
      int32_t code = doCreateTableGroup(pHandle->meta, pScanPhyNode->tableType, pScanPhyNode->uid, pTableGroupInfo,
                                        queryId, taskId);
4847 4848 4849 4850
      if (code != TSDB_CODE_SUCCESS) {
        return NULL;
      }

X
Xiaoyu Wang 已提交
4851
      int32_t    num = 0;
4852 4853 4854
      SExprInfo* pExprInfo = createExprInfo(pScanPhyNode->pScanPseudoCols, NULL, &num);

      int32_t numOfOutputCols = 0;
X
Xiaoyu Wang 已提交
4855 4856
      SArray* colList =
          extractColMatchInfo(pScanPhyNode->pScanPseudoCols, pDescNode, &numOfOutputCols, COL_MATCH_FROM_COL_ID);
4857

X
Xiaoyu Wang 已提交
4858 4859
      SOperatorInfo* pOperator =
          createTagScanOperatorInfo(pHandle, pExprInfo, num, pResBlock, colList, pTableGroupInfo, pTaskInfo);
4860
      return pOperator;
H
Haojun Liao 已提交
4861 4862
    } else {
      ASSERT(0);
H
Haojun Liao 已提交
4863 4864 4865
    }
  }

4866 4867
  int32_t num = 0;
  size_t  size = LIST_LENGTH(pPhyNode->pChildren);
H
Haojun Liao 已提交
4868

4869
  SOperatorInfo** ops = taosMemoryCalloc(size, POINTER_BYTES);
dengyihao's avatar
dengyihao 已提交
4870
  for (int32_t i = 0; i < size; ++i) {
4871 4872
    SPhysiNode* pChildNode = (SPhysiNode*)nodesListGetNode(pPhyNode->pChildren, i);
    ops[i] = createOperatorTree(pChildNode, pTaskInfo, pHandle, queryId, taskId, pTableGroupInfo);
4873 4874 4875
    if (ops[i] == NULL) {
      return NULL;
    }
4876
  }
H
Haojun Liao 已提交
4877

4878
  SOperatorInfo* pOptr = NULL;
H
Haojun Liao 已提交
4879
  if (QUERY_NODE_PHYSICAL_PLAN_PROJECT == type) {
dengyihao's avatar
dengyihao 已提交
4880 4881
    SProjectPhysiNode* pProjPhyNode = (SProjectPhysiNode*)pPhyNode;
    SExprInfo*         pExprInfo = createExprInfo(pProjPhyNode->pProjections, NULL, &num);
H
Haojun Liao 已提交
4882

4883
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
dengyihao's avatar
dengyihao 已提交
4884 4885
    SLimit       limit = {.limit = pProjPhyNode->limit, .offset = pProjPhyNode->offset};
    SLimit       slimit = {.limit = pProjPhyNode->slimit, .offset = pProjPhyNode->soffset};
4886
    pOptr = createProjectOperatorInfo(ops[0], pExprInfo, num, pResBlock, &limit, &slimit, pTaskInfo);
H
Haojun Liao 已提交
4887 4888 4889
  } else if (QUERY_NODE_PHYSICAL_PLAN_AGG == type) {
    SAggPhysiNode* pAggNode = (SAggPhysiNode*)pPhyNode;
    SExprInfo*     pExprInfo = createExprInfo(pAggNode->pAggFuncs, pAggNode->pGroupKeys, &num);
4890
    SSDataBlock*   pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4891

dengyihao's avatar
dengyihao 已提交
4892
    int32_t    numOfScalarExpr = 0;
4893 4894 4895 4896 4897
    SExprInfo* pScalarExprInfo = NULL;
    if (pAggNode->pExprs != NULL) {
      pScalarExprInfo = createExprInfo(pAggNode->pExprs, NULL, &numOfScalarExpr);
    }

H
Haojun Liao 已提交
4898 4899
    if (pAggNode->pGroupKeys != NULL) {
      SArray* pColList = extractColumnInfo(pAggNode->pGroupKeys);
dengyihao's avatar
dengyihao 已提交
4900 4901
      pOptr = createGroupOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pAggNode->node.pConditions,
                                      pScalarExprInfo, numOfScalarExpr, pTaskInfo, NULL);
H
Haojun Liao 已提交
4902
    } else {
dengyihao's avatar
dengyihao 已提交
4903 4904
      pOptr = createAggregateOperatorInfo(ops[0], pExprInfo, num, pResBlock, pScalarExprInfo, numOfScalarExpr,
                                          pTaskInfo, pTableGroupInfo);
H
Haojun Liao 已提交
4905
    }
X
Xiaoyu Wang 已提交
4906
  } else if (QUERY_NODE_PHYSICAL_PLAN_INTERVAL == type || QUERY_NODE_PHYSICAL_PLAN_STREAM_INTERVAL == type) {
H
Haojun Liao 已提交
4907
    SIntervalPhysiNode* pIntervalPhyNode = (SIntervalPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4908

H
Haojun Liao 已提交
4909
    SExprInfo*   pExprInfo = createExprInfo(pIntervalPhyNode->window.pFuncs, NULL, &num);
4910
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
H
Haojun Liao 已提交
4911

dengyihao's avatar
dengyihao 已提交
4912 4913 4914 4915 4916 4917
    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
4918

dengyihao's avatar
dengyihao 已提交
4919 4920
    STimeWindowAggSupp as = {.waterMark = pIntervalPhyNode->window.watermark,
                             .calTrigger = pIntervalPhyNode->window.triggerType};
4921

4922 4923 4924 4925
    int32_t tsSlotId = ((SColumnNode*)pIntervalPhyNode->window.pTspk)->slotId;
    pOptr = createIntervalOperatorInfo(ops[0], pExprInfo, num, pResBlock, &interval, tsSlotId, &as, pTableGroupInfo,
                                       pTaskInfo);

H
Haojun Liao 已提交
4926
  } else if (QUERY_NODE_PHYSICAL_PLAN_SORT == type) {
H
Haojun Liao 已提交
4927
    SSortPhysiNode* pSortPhyNode = (SSortPhysiNode*)pPhyNode;
H
Haojun Liao 已提交
4928

4929 4930 4931
    SDataBlockDescNode* pDescNode = pPhyNode->pOutputDataBlockDesc;

    SSDataBlock* pResBlock = createResDataBlock(pDescNode);
4932
    SArray*      info = createSortInfo(pSortPhyNode->pSortKeys);
4933 4934

    int32_t    numOfCols = 0;
4935 4936 4937
    SExprInfo* pExprInfo = createExprInfo(pSortPhyNode->pExprs, NULL, &numOfCols);

    int32_t numOfOutputCols = 0;
4938
    SArray* pColList = extractColMatchInfo(pSortPhyNode->pTargets, pDescNode, &numOfOutputCols, COL_MATCH_FROM_SLOT_ID);
4939

4940
    pOptr = createSortOperatorInfo(ops[0], pResBlock, info, pExprInfo, numOfCols, pColList, pTaskInfo);
H
Haojun Liao 已提交
4941
  } else if (QUERY_NODE_PHYSICAL_PLAN_SESSION_WINDOW == type) {
H
Haojun Liao 已提交
4942 4943
    SSessionWinodwPhysiNode* pSessionNode = (SSessionWinodwPhysiNode*)pPhyNode;

X
Xiaoyu Wang 已提交
4944 4945
    STimeWindowAggSupp as = {.waterMark = pSessionNode->window.watermark,
                             .calTrigger = pSessionNode->window.triggerType};
4946

H
Haojun Liao 已提交
4947
    SExprInfo*   pExprInfo = createExprInfo(pSessionNode->window.pFuncs, NULL, &num);
4948
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4949 4950
    int32_t      tsSlotId = ((SColumnNode*)pSessionNode->window.pTspk)->slotId;

X
Xiaoyu Wang 已提交
4951 4952
    pOptr =
        createSessionAggOperatorInfo(ops[0], pExprInfo, num, pResBlock, pSessionNode->gap, tsSlotId, &as, pTaskInfo);
H
Haojun Liao 已提交
4953
  } else if (QUERY_NODE_PHYSICAL_PLAN_PARTITION == type) {
dengyihao's avatar
dengyihao 已提交
4954 4955 4956
    SPartitionPhysiNode* pPartNode = (SPartitionPhysiNode*)pPhyNode;
    SArray*              pColList = extractPartitionColInfo(pPartNode->pPartitionKeys);
    SSDataBlock*         pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4957 4958

    SExprInfo* pExprInfo = createExprInfo(pPartNode->pTargets, NULL, &num);
4959
    pOptr = createPartitionOperatorInfo(ops[0], pExprInfo, num, pResBlock, pColList, pTaskInfo, NULL);
4960
  } else if (QUERY_NODE_PHYSICAL_PLAN_STATE_WINDOW == type) {
dengyihao's avatar
dengyihao 已提交
4961
    SStateWinodwPhysiNode* pStateNode = (SStateWinodwPhysiNode*)pPhyNode;
4962

4963 4964
    STimeWindowAggSupp as = {.waterMark = pStateNode->window.watermark, .calTrigger = pStateNode->window.triggerType};

dengyihao's avatar
dengyihao 已提交
4965
    SExprInfo*   pExprInfo = createExprInfo(pStateNode->window.pFuncs, NULL, &num);
4966
    SSDataBlock* pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4967 4968
    int32_t      tsSlotId = ((SColumnNode*)pStateNode->window.pTspk)->slotId;

4969 4970 4971
    SColumnNode* pColNode = (SColumnNode*)((STargetNode*)pStateNode->pStateKey)->pExpr;
    SColumn col = extractColumnFromColumnNode(pColNode);
    pOptr = createStatewindowOperatorInfo(ops[0], pExprInfo, num, pResBlock, &as, tsSlotId, &col, pTaskInfo);
4972
  } else if (QUERY_NODE_PHYSICAL_PLAN_JOIN == type) {
dengyihao's avatar
dengyihao 已提交
4973 4974
    SJoinPhysiNode* pJoinNode = (SJoinPhysiNode*)pPhyNode;
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
4975 4976

    SExprInfo* pExprInfo = createExprInfo(pJoinNode->pTargets, NULL, &num);
H
Haojun Liao 已提交
4977
    pOptr = createMergeJoinOperatorInfo(ops, size, pExprInfo, num, pResBlock, pJoinNode->pOnConditions, pTaskInfo);
4978 4979
  } else if (QUERY_NODE_PHYSICAL_PLAN_FILL == type) {
    SFillPhysiNode* pFillNode = (SFillPhysiNode*)pPhyNode;
X
Xiaoyu Wang 已提交
4980 4981
    SSDataBlock*    pResBlock = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
    SExprInfo*      pExprInfo = createExprInfo(pFillNode->pTargets, NULL, &num);
4982

4983
    SInterval* pInterval = &((SIntervalAggOperatorInfo*)ops[0]->info)->interval;
X
Xiaoyu Wang 已提交
4984 4985
    pOptr = createFillOperatorInfo(ops[0], pExprInfo, num, pInterval, &pFillNode->timeRange, pResBlock, pFillNode->mode,
                                   (SNodeListNode*)pFillNode->pValues, false, pTaskInfo);
H
Haojun Liao 已提交
4986 4987
  } else {
    ASSERT(0);
H
Haojun Liao 已提交
4988
  }
4989 4990 4991

  taosMemoryFree(ops);
  return pOptr;
4992
}
H
Haojun Liao 已提交
4993

4994
int32_t initQueryTableDataCond(SQueryTableDataCond* pCond, const STableScanPhysiNode* pTableScanNode) {
4995 4996 4997 4998 4999 5000
  pCond->loadExternalRows = false;

  pCond->order = pTableScanNode->scanSeq[0] > 0 ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
  pCond->numOfCols = LIST_LENGTH(pTableScanNode->scan.pScanCols);
  pCond->colList = taosMemoryCalloc(pCond->numOfCols, sizeof(SColumnInfo));
  if (pCond->colList == NULL) {
H
Haojun Liao 已提交
5001
    terrno = TSDB_CODE_QRY_OUT_OF_MEMORY;
5002 5003 5004 5005 5006 5007
    return terrno;
  }

  pCond->twindow = pTableScanNode->scanRange;

#if 1
L
Liu Jicong 已提交
5008
  // todo work around a problem, remove it later
5009 5010
  if ((pCond->order == TSDB_ORDER_ASC && pCond->twindow.skey > pCond->twindow.ekey) ||
      (pCond->order == TSDB_ORDER_DESC && pCond->twindow.skey < pCond->twindow.ekey)) {
wafwerar's avatar
wafwerar 已提交
5011
    TSWAP(pCond->twindow.skey, pCond->twindow.ekey);
H
Haojun Liao 已提交
5012
  }
5013
#endif
H
Haojun Liao 已提交
5014

5015 5016
  pCond->type = BLOCK_LOAD_OFFSET_SEQ_ORDER;
  //  pCond->type = pTableScanNode->scanFlag;
H
Haojun Liao 已提交
5017

H
Haojun Liao 已提交
5018
  int32_t j = 0;
5019
  for (int32_t i = 0; i < pCond->numOfCols; ++i) {
H
Haojun Liao 已提交
5020 5021
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pTableScanNode->scan.pScanCols, i);
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;
H
Haojun Liao 已提交
5022 5023 5024
    if (pColNode->colType == COLUMN_TYPE_TAG) {
      continue;
    }
H
Haojun Liao 已提交
5025

5026 5027 5028
    pCond->colList[j].type = pColNode->node.resType.type;
    pCond->colList[j].bytes = pColNode->node.resType.bytes;
    pCond->colList[j].colId = pColNode->colId;
H
Haojun Liao 已提交
5029
    j += 1;
H
Haojun Liao 已提交
5030 5031
  }

5032 5033
  pCond->numOfCols = j;
  return TSDB_CODE_SUCCESS;
H
Haojun Liao 已提交
5034
}
H
Haojun Liao 已提交
5035

5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046
SColumn extractColumnFromColumnNode(SColumnNode* pColNode) {
  SColumn c = {0};
  c.slotId    = pColNode->slotId;
  c.colId     = pColNode->colId;
  c.type      = pColNode->node.resType.type;
  c.bytes     = pColNode->node.resType.bytes;
  c.scale     = pColNode->node.resType.scale;
  c.precision = pColNode->node.resType.precision;
  return c;
}

H
Haojun Liao 已提交
5047
SArray* extractColumnInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
5048
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5049 5050 5051 5052 5053 5054
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
5055 5056
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
H
Haojun Liao 已提交
5057

5058 5059 5060
    if (nodeType(pNode->pExpr) == QUERY_NODE_COLUMN) {
      SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;

5061
      SColumn c = extractColumnFromColumnNode(pColNode);
5062 5063
      taosArrayPush(pList, &c);
    } else if (nodeType(pNode->pExpr) == QUERY_NODE_VALUE) {
L
Liu Jicong 已提交
5064 5065
      SValueNode* pValNode = (SValueNode*)pNode->pExpr;
      SColumn     c = {0};
5066
      c.slotId = pNode->slotId;
L
Liu Jicong 已提交
5067 5068 5069 5070
      c.colId = pNode->slotId;
      c.type = pValNode->node.type;
      c.bytes = pValNode->node.resType.bytes;
      c.scale = pValNode->node.resType.scale;
5071 5072 5073 5074
      c.precision = pValNode->node.resType.precision;

      taosArrayPush(pList, &c);
    }
H
Haojun Liao 已提交
5075 5076 5077 5078 5079
  }

  return pList;
}

5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093
SArray* extractPartitionColInfo(SNodeList* pNodeList) {
  size_t  numOfCols = LIST_LENGTH(pNodeList);
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColumn));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

  for (int32_t i = 0; i < numOfCols; ++i) {
    SColumnNode* pColNode = (SColumnNode*)nodesListGetNode(pNodeList, i);

    // todo extract method
    SColumn c = {0};
    c.slotId = pColNode->slotId;
dengyihao's avatar
dengyihao 已提交
5094 5095 5096
    c.colId = pColNode->colId;
    c.type = pColNode->node.resType.type;
    c.bytes = pColNode->node.resType.bytes;
5097 5098 5099 5100 5101 5102 5103 5104 5105
    c.precision = pColNode->node.resType.precision;
    c.scale = pColNode->node.resType.scale;

    taosArrayPush(pList, &c);
  }

  return pList;
}

5106
SArray* createSortInfo(SNodeList* pNodeList) {
L
Liu Jicong 已提交
5107
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5108 5109 5110 5111 5112 5113
  SArray* pList = taosArrayInit(numOfCols, sizeof(SBlockOrderInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return pList;
  }

L
Liu Jicong 已提交
5114
  for (int32_t i = 0; i < numOfCols; ++i) {
5115
    SOrderByExprNode* pSortKey = (SOrderByExprNode*)nodesListGetNode(pNodeList, i);
L
Liu Jicong 已提交
5116 5117
    SBlockOrderInfo   bi = {0};
    bi.order = (pSortKey->order == ORDER_ASC) ? TSDB_ORDER_ASC : TSDB_ORDER_DESC;
H
Haojun Liao 已提交
5118 5119 5120
    bi.nullFirst = (pSortKey->nullOrder == NULL_ORDER_FIRST);

    SColumnNode* pColNode = (SColumnNode*)pSortKey->pExpr;
5121
    bi.slotId = pColNode->slotId;
H
Haojun Liao 已提交
5122 5123 5124 5125 5126 5127
    taosArrayPush(pList, &bi);
  }

  return pList;
}

X
Xiaoyu Wang 已提交
5128 5129
SArray* extractColMatchInfo(SNodeList* pNodeList, SDataBlockDescNode* pOutputNodeList, int32_t* numOfOutputCols,
                            int32_t type) {
L
Liu Jicong 已提交
5130
  size_t  numOfCols = LIST_LENGTH(pNodeList);
H
Haojun Liao 已提交
5131 5132 5133 5134 5135 5136
  SArray* pList = taosArrayInit(numOfCols, sizeof(SColMatchInfo));
  if (pList == NULL) {
    terrno = TSDB_CODE_OUT_OF_MEMORY;
    return NULL;
  }

L
Liu Jicong 已提交
5137 5138 5139
  for (int32_t i = 0; i < numOfCols; ++i) {
    STargetNode* pNode = (STargetNode*)nodesListGetNode(pNodeList, i);
    SColumnNode* pColNode = (SColumnNode*)pNode->pExpr;
H
Haojun Liao 已提交
5140 5141

    SColMatchInfo c = {0};
X
Xiaoyu Wang 已提交
5142 5143 5144 5145
    c.output = true;
    c.colId = pColNode->colId;
    c.srcSlotId = pColNode->slotId;
    c.matchType = type;
H
Haojun Liao 已提交
5146 5147 5148 5149
    c.targetSlotId = pNode->slotId;
    taosArrayPush(pList, &c);
  }

H
Haojun Liao 已提交
5150 5151
  *numOfOutputCols = 0;
  int32_t num = LIST_LENGTH(pOutputNodeList->pSlots);
L
Liu Jicong 已提交
5152 5153
  for (int32_t i = 0; i < num; ++i) {
    SSlotDescNode* pNode = (SSlotDescNode*)nodesListGetNode(pOutputNodeList->pSlots, i);
5154

5155
    // todo: add reserve flag check
5156 5157
    // it is a column reserved for the arithmetic expression calculation
    if (pNode->slotId >= numOfCols) {
5158 5159 5160 5161
      (*numOfOutputCols) += 1;
      continue;
    }

H
Haojun Liao 已提交
5162
    SColMatchInfo* info = taosArrayGet(pList, pNode->slotId);
H
Haojun Liao 已提交
5163 5164 5165 5166 5167
    if (pNode->output) {
      (*numOfOutputCols) += 1;
    } else {
      info->output = false;
    }
H
Haojun Liao 已提交
5168 5169
  }

H
Haojun Liao 已提交
5170 5171 5172
  return pList;
}

dengyihao's avatar
dengyihao 已提交
5173 5174
int32_t doCreateTableGroup(void* metaHandle, int32_t tableType, uint64_t tableUid, STableGroupInfo* pGroupInfo,
                           uint64_t queryId, uint64_t taskId) {
5175
  int32_t code = 0;
H
Haojun Liao 已提交
5176
  if (tableType == TSDB_SUPER_TABLE) {
H
Haojun Liao 已提交
5177
    code = tsdbQuerySTableByTagCond(metaHandle, tableUid, 0, NULL, 0, 0, NULL, pGroupInfo, NULL, 0, queryId, taskId);
H
Haojun Liao 已提交
5178
  } else {  // Create one table group.
H
Haojun Liao 已提交
5179
    code = tsdbGetOneTableGroup(metaHandle, tableUid, 0, pGroupInfo);
5180 5181 5182 5183
  }

  return code;
}
H
Haojun Liao 已提交
5184

5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202
SArray* extractTableIdList(const STableGroupInfo* pTableGroupInfo) {
  SArray* tableIdList = taosArrayInit(4, sizeof(uint64_t));

  if (pTableGroupInfo->numOfTables > 0) {
    SArray* pa = taosArrayGetP(pTableGroupInfo->pGroupList, 0);
    ASSERT(taosArrayGetSize(pTableGroupInfo->pGroupList) == 1);

    // Transfer the Array of STableKeyInfo into uid list.
    size_t numOfTables = taosArrayGetSize(pa);
    for (int32_t i = 0; i < numOfTables; ++i) {
      STableKeyInfo* pkeyInfo = taosArrayGet(pa, i);
      taosArrayPush(tableIdList, &pkeyInfo->uid);
    }
  }

  return tableIdList;
}

L
Liu Jicong 已提交
5203 5204
tsdbReaderT doCreateDataReader(STableScanPhysiNode* pTableScanNode, SReadHandle* pHandle,
                               STableGroupInfo* pTableGroupInfo, uint64_t queryId, uint64_t taskId) {
5205
  uint64_t uid = pTableScanNode->scan.uid;
L
Liu Jicong 已提交
5206 5207
  int32_t  code =
      doCreateTableGroup(pHandle->meta, pTableScanNode->scan.tableType, uid, pTableGroupInfo, queryId, taskId);
5208 5209
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
H
Haojun Liao 已提交
5210
  }
H
Haojun Liao 已提交
5211

H
Haojun Liao 已提交
5212
  if (pTableGroupInfo->numOfTables == 0) {
H
Haojun Liao 已提交
5213
    code = 0;
L
Liu Jicong 已提交
5214
    qDebug("no table qualified for query, TID:0x%" PRIx64 ", QID:0x%" PRIx64, taskId, queryId);
H
Haojun Liao 已提交
5215 5216
    goto _error;
  }
H
Haojun Liao 已提交
5217

5218 5219 5220 5221
  SQueryTableDataCond cond = {0};
  code = initQueryTableDataCond(&cond, pTableScanNode);
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
X
Xiaoyu Wang 已提交
5222
  }
5223

5224
  return tsdbQueryTables(pHandle->vnode, &cond, pTableGroupInfo, queryId, taskId);
H
Haojun Liao 已提交
5225

L
Liu Jicong 已提交
5226
_error:
H
Haojun Liao 已提交
5227 5228 5229 5230
  terrno = code;
  return NULL;
}

dengyihao's avatar
dengyihao 已提交
5231 5232
int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,
                               EOPTR_EXEC_MODEL model) {
H
Haojun Liao 已提交
5233 5234
  uint64_t queryId = pPlan->id.queryId;

H
Haojun Liao 已提交
5235
  int32_t code = TSDB_CODE_SUCCESS;
5236
  *pTaskInfo = createExecTaskInfo(queryId, taskId, model);
H
Haojun Liao 已提交
5237 5238 5239 5240
  if (*pTaskInfo == NULL) {
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
    goto _complete;
  }
H
Haojun Liao 已提交
5241

L
Liu Jicong 已提交
5242 5243
  (*pTaskInfo)->pRoot =
      createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, queryId, taskId, &(*pTaskInfo)->tableqinfoGroupInfo);
D
dapan1121 已提交
5244
  if (NULL == (*pTaskInfo)->pRoot) {
L
Liu Jicong 已提交
5245
    code = terrno;
D
dapan1121 已提交
5246 5247
    goto _complete;
  }
H
Haojun Liao 已提交
5248

5249
  if ((*pTaskInfo)->pRoot == NULL) {
H
Haojun Liao 已提交
5250
    code = TSDB_CODE_QRY_OUT_OF_MEMORY;
H
Haojun Liao 已提交
5251
    goto _complete;
5252 5253
  }

H
Haojun Liao 已提交
5254 5255
  return code;

H
Haojun Liao 已提交
5256
_complete:
wafwerar's avatar
wafwerar 已提交
5257
  taosMemoryFreeClear(*pTaskInfo);
H
Haojun Liao 已提交
5258 5259 5260

  terrno = code;
  return code;
H
Haojun Liao 已提交
5261 5262
}

H
Haojun Liao 已提交
5263
void setResultBufSize(STaskAttr* pQueryAttr, SResultInfo* pResultInfo) {
5264 5265 5266 5267 5268 5269 5270 5271
  const int32_t DEFAULT_RESULT_MSG_SIZE = 1024 * (1024 + 512);

  // the minimum number of rows for projection query
  const int32_t MIN_ROWS_FOR_PRJ_QUERY = 8192;
  const int32_t DEFAULT_MIN_ROWS = 4096;

  const float THRESHOLD_RATIO = 0.85f;

dengyihao's avatar
dengyihao 已提交
5272 5273 5274 5275 5276 5277 5278 5279 5280 5281
  //  if (isProjQuery(pQueryAttr)) {
  //    int32_t numOfRes = DEFAULT_RESULT_MSG_SIZE / pQueryAttr->resultRowSize;
  //    if (numOfRes < MIN_ROWS_FOR_PRJ_QUERY) {
  //      numOfRes = MIN_ROWS_FOR_PRJ_QUERY;
  //    }
  //
  //    pResultInfo->capacity = numOfRes;
  //  } else {  // in case of non-prj query, a smaller output buffer will be used.
  //    pResultInfo->capacity = DEFAULT_MIN_ROWS;
  //  }
5282 5283

  pResultInfo->threshold = (int32_t)(pResultInfo->capacity * THRESHOLD_RATIO);
H
Haojun Liao 已提交
5284
  pResultInfo->totalRows = 0;
5285 5286
}

L
Liu Jicong 已提交
5287
// TODO refactor
5288
void freeColumnFilterInfo(SColumnFilterInfo* pFilter, int32_t numOfFilters) {
L
Liu Jicong 已提交
5289 5290 5291
  if (pFilter == NULL || numOfFilters == 0) {
    return;
  }
5292

L
Liu Jicong 已提交
5293 5294 5295
  for (int32_t i = 0; i < numOfFilters; i++) {
    if (pFilter[i].filterstr && pFilter[i].pz) {
      taosMemoryFree((void*)(pFilter[i].pz));
5296
    }
L
Liu Jicong 已提交
5297
  }
5298

L
Liu Jicong 已提交
5299
  taosMemoryFree(pFilter);
5300 5301 5302 5303
}

static void doDestroyTableQueryInfo(STableGroupInfo* pTableqinfoGroupInfo) {
  if (pTableqinfoGroupInfo->pGroupList != NULL) {
L
Liu Jicong 已提交
5304
    int32_t numOfGroups = (int32_t)taosArrayGetSize(pTableqinfoGroupInfo->pGroupList);
5305
    for (int32_t i = 0; i < numOfGroups; ++i) {
L
Liu Jicong 已提交
5306
      SArray* p = taosArrayGetP(pTableqinfoGroupInfo->pGroupList, i);
5307 5308

      size_t num = taosArrayGetSize(p);
L
Liu Jicong 已提交
5309
      for (int32_t j = 0; j < num; ++j) {
5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325
        STableQueryInfo* item = taosArrayGetP(p, j);
        destroyTableQueryInfoImpl(item);
      }

      taosArrayDestroy(p);
    }
  }

  taosArrayDestroy(pTableqinfoGroupInfo->pGroupList);
  taosHashCleanup(pTableqinfoGroupInfo->map);

  pTableqinfoGroupInfo->pGroupList = NULL;
  pTableqinfoGroupInfo->map = NULL;
  pTableqinfoGroupInfo->numOfTables = 0;
}

L
Liu Jicong 已提交
5326
void doDestroyTask(SExecTaskInfo* pTaskInfo) {
H
Haojun Liao 已提交
5327 5328
  qDebug("%s execTask is freed", GET_TASKID(pTaskInfo));

H
Haojun Liao 已提交
5329
  doDestroyTableQueryInfo(&pTaskInfo->tableqinfoGroupInfo);
H
Haojun Liao 已提交
5330
  destroyOperatorInfo(pTaskInfo->pRoot);
L
Liu Jicong 已提交
5331 5332
  //  taosArrayDestroy(pTaskInfo->summary.queryProfEvents);
  //  taosHashCleanup(pTaskInfo->summary.operatorProfResults);
5333

wafwerar's avatar
wafwerar 已提交
5334 5335 5336
  taosMemoryFreeClear(pTaskInfo->sql);
  taosMemoryFreeClear(pTaskInfo->id.str);
  taosMemoryFreeClear(pTaskInfo);
5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348
}

static void doSetTagValueToResultBuf(char* output, const char* val, int16_t type, int16_t bytes) {
  if (val == NULL) {
    setNull(output, type, bytes);
    return;
  }

  if (IS_VAR_DATA_TYPE(type)) {
    // Binary data overflows for sort of unknown reasons. Let trim the overflow data
    if (varDataTLen(val) > bytes) {
      int32_t maxLen = bytes - VARSTR_HEADER_SIZE;
L
Liu Jicong 已提交
5349
      int32_t len = (varDataLen(val) > maxLen) ? maxLen : varDataLen(val);
5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361
      memcpy(varDataVal(output), varDataVal(val), len);
      varDataSetLen(output, len);
    } else {
      varDataCopy(output, val);
    }
  } else {
    memcpy(output, val, bytes);
  }
}

static int64_t getQuerySupportBufSize(size_t numOfTables) {
  size_t s1 = sizeof(STableQueryInfo);
L
Liu Jicong 已提交
5362 5363
  //  size_t s3 = sizeof(STableCheckInfo);  buffer consumption in tsdb
  return (int64_t)(s1 * 1.5 * numOfTables);
5364 5365 5366 5367 5368 5369 5370
}

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
5371
    while (1) {
5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397
      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
5398

dengyihao's avatar
dengyihao 已提交
5399 5400
int32_t getOperatorExplainExecInfo(SOperatorInfo* operatorInfo, SExplainExecInfo** pRes, int32_t* capacity,
                                   int32_t* resNum) {
D
dapan1121 已提交
5401 5402
  if (*resNum >= *capacity) {
    *capacity += 10;
dengyihao's avatar
dengyihao 已提交
5403

D
dapan1121 已提交
5404 5405
    *pRes = taosMemoryRealloc(*pRes, (*capacity) * sizeof(SExplainExecInfo));
    if (NULL == *pRes) {
D
dapan1121 已提交
5406
      qError("malloc %d failed", (*capacity) * (int32_t)sizeof(SExplainExecInfo));
D
dapan1121 已提交
5407 5408 5409 5410
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

D
dapan1121 已提交
5411 5412 5413 5414
  (*pRes)[*resNum].numOfRows = operatorInfo->resultInfo.totalRows;
  (*pRes)[*resNum].startupCost = operatorInfo->cost.openCost;
  (*pRes)[*resNum].totalCost = operatorInfo->cost.totalCost;

5415 5416
  if (operatorInfo->fpSet.getExplainFn) {
    int32_t code = (*operatorInfo->fpSet.getExplainFn)(operatorInfo, &(*pRes)->verboseInfo);
D
dapan1121 已提交
5417 5418 5419 5420 5421
    if (code) {
      qError("operator getExplainFn failed, error:%s", tstrerror(code));
      return code;
    }
  }
dengyihao's avatar
dengyihao 已提交
5422

D
dapan1121 已提交
5423
  ++(*resNum);
dengyihao's avatar
dengyihao 已提交
5424

D
dapan1121 已提交
5425
  int32_t code = 0;
D
dapan1121 已提交
5426 5427
  for (int32_t i = 0; i < operatorInfo->numOfDownstream; ++i) {
    code = getOperatorExplainExecInfo(operatorInfo->pDownstream[i], pRes, capacity, resNum);
D
dapan1121 已提交
5428 5429 5430 5431 5432 5433 5434
    if (code) {
      taosMemoryFreeClear(*pRes);
      return TSDB_CODE_QRY_OUT_OF_MEMORY;
    }
  }

  return TSDB_CODE_SUCCESS;
D
dapan1121 已提交
5435 5436
}

5437
static SSDataBlock* doMergeJoin(struct SOperatorInfo* pOperator) {
5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449
  SJoinOperatorInfo* pJoinInfo = pOperator->info;

  SSDataBlock* pRes = pJoinInfo->pRes;
  blockDataCleanup(pRes);
  blockDataEnsureCapacity(pRes, 4096);

  int32_t nrows = 0;

  while (1) {
    if (pJoinInfo->pLeft == NULL || pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
      SOperatorInfo* ds1 = pOperator->pDownstream[0];
      publishOperatorProfEvent(ds1, QUERY_PROF_BEFORE_OPERATOR_EXEC);
5450
      pJoinInfo->pLeft = ds1->fpSet.getNextFn(ds1);
5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462
      publishOperatorProfEvent(ds1, QUERY_PROF_AFTER_OPERATOR_EXEC);

      pJoinInfo->leftPos = 0;
      if (pJoinInfo->pLeft == NULL) {
        setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
        break;
      }
    }

    if (pJoinInfo->pRight == NULL || pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
      SOperatorInfo* ds2 = pOperator->pDownstream[1];
      publishOperatorProfEvent(ds2, QUERY_PROF_BEFORE_OPERATOR_EXEC);
5463
      pJoinInfo->pRight = ds2->fpSet.getNextFn(ds2);
5464 5465 5466 5467 5468 5469 5470 5471 5472 5473
      publishOperatorProfEvent(ds2, QUERY_PROF_AFTER_OPERATOR_EXEC);

      pJoinInfo->rightPos = 0;
      if (pJoinInfo->pRight == NULL) {
        setTaskStatus(pOperator->pTaskInfo, TASK_COMPLETED);
        break;
      }
    }

    SColumnInfoData* pLeftCol = taosArrayGet(pJoinInfo->pLeft->pDataBlock, pJoinInfo->leftCol.slotId);
dengyihao's avatar
dengyihao 已提交
5474
    char*            pLeftVal = colDataGetData(pLeftCol, pJoinInfo->leftPos);
5475 5476

    SColumnInfoData* pRightCol = taosArrayGet(pJoinInfo->pRight->pDataBlock, pJoinInfo->rightCol.slotId);
dengyihao's avatar
dengyihao 已提交
5477
    char*            pRightVal = colDataGetData(pRightCol, pJoinInfo->rightPos);
5478 5479

    // only the timestamp match support for ordinary table
dengyihao's avatar
dengyihao 已提交
5480 5481
    ASSERT(pLeftCol->info.type == TSDB_DATA_TYPE_TIMESTAMP);
    if (*(int64_t*)pLeftVal == *(int64_t*)pRightVal) {
5482
      for (int32_t i = 0; i < pOperator->numOfExprs; ++i) {
dengyihao's avatar
dengyihao 已提交
5483
        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, i);
5484

dengyihao's avatar
dengyihao 已提交
5485
        SExprInfo* pExprInfo = &pOperator->pExpr[i];
5486

dengyihao's avatar
dengyihao 已提交
5487 5488
        int32_t blockId = pExprInfo->base.pParam[0].pCol->dataBlockId;
        int32_t slotId = pExprInfo->base.pParam[0].pCol->slotId;
5489

dengyihao's avatar
dengyihao 已提交
5490 5491 5492 5493 5494 5495
        SColumnInfoData* pSrc = NULL;
        if (pJoinInfo->pLeft->info.blockId == blockId) {
          pSrc = taosArrayGet(pJoinInfo->pLeft->pDataBlock, slotId);
        } else {
          pSrc = taosArrayGet(pJoinInfo->pRight->pDataBlock, slotId);
        }
5496

dengyihao's avatar
dengyihao 已提交
5497 5498 5499 5500 5501
        if (colDataIsNull_s(pSrc, pJoinInfo->leftPos)) {
          colDataAppendNULL(pDst, nrows);
        } else {
          char* p = colDataGetData(pSrc, pJoinInfo->leftPos);
          colDataAppend(pDst, nrows, p, false);
5502
        }
dengyihao's avatar
dengyihao 已提交
5503
      }
5504

dengyihao's avatar
dengyihao 已提交
5505 5506
      pJoinInfo->leftPos += 1;
      pJoinInfo->rightPos += 1;
5507

dengyihao's avatar
dengyihao 已提交
5508 5509 5510
      nrows += 1;
    } else if (*(int64_t*)pLeftVal < *(int64_t*)pRightVal) {
      pJoinInfo->leftPos += 1;
D
dapan1121 已提交
5511

dengyihao's avatar
dengyihao 已提交
5512 5513
      if (pJoinInfo->leftPos >= pJoinInfo->pLeft->info.rows) {
        continue;
5514
      }
dengyihao's avatar
dengyihao 已提交
5515 5516 5517 5518
    } else if (*(int64_t*)pLeftVal > *(int64_t*)pRightVal) {
      pJoinInfo->rightPos += 1;
      if (pJoinInfo->rightPos >= pJoinInfo->pRight->info.rows) {
        continue;
5519
      }
dengyihao's avatar
dengyihao 已提交
5520
    }
5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531

    // the pDataBlock are always the same one, no need to call this again
    pRes->info.rows = nrows;
    if (pRes->info.rows >= pOperator->resultInfo.threshold) {
      break;
    }
  }

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

H
Haojun Liao 已提交
5532
SOperatorInfo* createMergeJoinOperatorInfo(SOperatorInfo** pDownstream, int32_t numOfDownstream, SExprInfo* pExprInfo,
X
Xiaoyu Wang 已提交
5533 5534
                                           int32_t numOfCols, SSDataBlock* pResBlock, SNode* pOnCondition,
                                           SExecTaskInfo* pTaskInfo) {
5535
  SJoinOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SJoinOperatorInfo));
dengyihao's avatar
dengyihao 已提交
5536
  SOperatorInfo*     pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
5537 5538 5539 5540
  if (pOperator == NULL || pInfo == NULL) {
    goto _error;
  }

5541
  initResultSizeInfo(pOperator, 4096);
5542

X
Xiaoyu Wang 已提交
5543 5544
  pInfo->pRes = pResBlock;
  pOperator->name = "MergeJoinOperator";
5545
  pOperator->operatorType = QUERY_NODE_PHYSICAL_PLAN_JOIN;
X
Xiaoyu Wang 已提交
5546 5547 5548
  pOperator->blocking = false;
  pOperator->status = OP_NOT_OPENED;
  pOperator->pExpr = pExprInfo;
5549
  pOperator->numOfExprs = numOfCols;
X
Xiaoyu Wang 已提交
5550
  pOperator->info = pInfo;
dengyihao's avatar
dengyihao 已提交
5551
  pOperator->pTaskInfo = pTaskInfo;
5552

5553 5554 5555 5556
  SOperatorNode* pNode = (SOperatorNode*)pOnCondition;
  setJoinColumnInfo(&pInfo->leftCol, (SColumnNode*)pNode->pLeft);
  setJoinColumnInfo(&pInfo->rightCol, (SColumnNode*)pNode->pRight);

L
Liu Jicong 已提交
5557
  pOperator->fpSet =
H
Haojun Liao 已提交
5558
      createOperatorFpSet(operatorDummyOpenFn, doMergeJoin, NULL, NULL, destroyMergeJoinOperator, NULL, NULL, NULL);
5559
  int32_t code = appendDownstream(pOperator, pDownstream, numOfDownstream);
5560 5561 5562 5563
  if (code != TSDB_CODE_SUCCESS) {
    goto _error;
  }

5564 5565
  return pOperator;

dengyihao's avatar
dengyihao 已提交
5566
_error:
5567 5568 5569 5570
  taosMemoryFree(pInfo);
  taosMemoryFree(pOperator);
  pTaskInfo->code = TSDB_CODE_OUT_OF_MEMORY;
  return NULL;
dengyihao's avatar
dengyihao 已提交
5571
}
5572 5573

void setJoinColumnInfo(SColumnInfo* pColumn, const SColumnNode* pColumnNode) {
X
Xiaoyu Wang 已提交
5574 5575 5576
  pColumn->slotId = pColumnNode->slotId;
  pColumn->type = pColumnNode->node.resType.type;
  pColumn->bytes = pColumnNode->node.resType.bytes;
5577
  pColumn->precision = pColumnNode->node.resType.precision;
X
Xiaoyu Wang 已提交
5578
  pColumn->scale = pColumnNode->node.resType.scale;
5579
}