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

taosdata / TDengine / #4143

24 May 2025 03:30AM UTC coverage: 32.868% (-29.4%) from 62.238%
#4143

push

travis-ci

web-flow
test: migrate stream cases (#31164)

76401 of 312956 branches covered (24.41%)

Branch coverage included in aggregate %.

128686 of 311012 relevant lines covered (41.38%)

579734.08 hits per line

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

0.0
/source/libs/executor/src/forecastoperator.c
1
/*
2
 * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
3
 *
4
 * This program is free software: you can use, redistribute, and/or modify
5
 * it under the terms of the GNU Affero General Public License, version 3
6
 * or later ("AGPL"), as published by the Free Software Foundation.
7
 *
8
 * This program is distributed in the hope that it will be useful, but WITHOUT
9
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10
 * FITNESS FOR A PARTICULAR PURPOSE.
11
 *
12
 * You should have received a copy of the GNU Affero General Public License
13
 * along with this program. If not, see <http://www.gnu.org/licenses/>.
14
 */
15

16
#include "executorInt.h"
17
#include "filter.h"
18
#include "functionMgt.h"
19
#include "operator.h"
20
#include "querytask.h"
21
#include "tanalytics.h"
22
#include "tcommon.h"
23
#include "tcompare.h"
24
#include "tdatablock.h"
25
#include "tfill.h"
26
#include "ttime.h"
27

28
#ifdef USE_ANALYTICS
29

30
#define ALGO_OPT_RETCONF_NAME      "return_conf"
31
#define ALGO_OPT_FORECASTROWS_NAME "rows"
32
#define ALGO_OPT_CONF_NAME         "conf"
33
#define ALGO_OPT_START_NAME        "start"
34
#define ALGO_OPT_EVERY_NAME        "every"
35

36
typedef struct {
37
  char         algoName[TSDB_ANALYTIC_ALGO_NAME_LEN];
38
  char         algoUrl[TSDB_ANALYTIC_ALGO_URL_LEN];
39
  char         algoOpt[TSDB_ANALYTIC_ALGO_OPTION_LEN];
40
  int64_t      maxTs;
41
  int64_t      minTs;
42
  int64_t      numOfRows;
43
  uint64_t     groupId;
44
  int64_t      forecastRows;
45
  int64_t      cachedRows;
46
  int32_t      numOfBlocks;
47
  int64_t      timeout;
48
  int16_t      resTsSlot;
49
  int16_t      resValSlot;
50
  int16_t      resLowSlot;
51
  int16_t      resHighSlot;
52
  int16_t      inputTsSlot;
53
  int16_t      targetValSlot;
54
  int8_t       targetValType;
55
  int8_t       inputPrecision;
56
  int8_t       wncheck;
57
  double       conf;
58
  int64_t      startTs;
59
  int64_t      every;
60
  int8_t       setStart;
61
  int8_t       setEvery;
62
  SArray*      pCovariateSlotList;   // covariate slot list
63
  int32_t      numOfInputCols;
64
  SAnalyticBuf analyBuf;
65
} SForecastSupp;
66

67
typedef struct SForecastOperatorInfo {
68
  SSDataBlock*  pRes;
69
  SExprSupp     scalarSup;  // scalar calculation
70
  SForecastSupp forecastSupp;
71
} SForecastOperatorInfo;
72

73
static void destroyForecastInfo(void* param);
74
static int32_t forecastParseOpt(SForecastSupp* pSupp, const char* id);
75

76
static FORCE_INLINE int32_t forecastEnsureBlockCapacity(SSDataBlock* pBlock, int32_t newRowsNum) {
77
  if (pBlock->info.rows < pBlock->info.capacity) {
×
78
    return TSDB_CODE_SUCCESS;
×
79
  }
80

81
  int32_t code = blockDataEnsureCapacity(pBlock, newRowsNum);
×
82
  if (code != TSDB_CODE_SUCCESS) {
×
83
    qError("%s failed at line %d since %s", __func__, __LINE__, tstrerror(code));
×
84
    return code;
×
85
  }
86

87
  return TSDB_CODE_SUCCESS;
×
88
}
89

90
static int32_t forecastCacheBlock(SForecastSupp* pSupp, SSDataBlock* pBlock, const char* id) {
×
91
  int32_t       code = TSDB_CODE_SUCCESS;
×
92
  int32_t       lino = 0;
×
93
  SAnalyticBuf* pBuf = &pSupp->analyBuf;
×
94

95
  if (pSupp->cachedRows > ANALY_FORECAST_MAX_ROWS) {
×
96
    code = TSDB_CODE_ANA_ANODE_TOO_MANY_ROWS;
×
97
    qError("%s rows:%" PRId64 " for forecast cache, error happens, code:%s, upper limit:%d", id, pSupp->cachedRows,
×
98
           tstrerror(code), ANALY_FORECAST_MAX_ROWS);
99
    return code;
×
100
  }
101

102
  pSupp->numOfBlocks++;
×
103
  qDebug("%s block:%d, %p rows:%" PRId64, id, pSupp->numOfBlocks, pBlock, pBlock->info.rows);
×
104

105
  for (int32_t j = 0; j < pBlock->info.rows; ++j) {
×
106
    SColumnInfoData* pValCol = taosArrayGet(pBlock->pDataBlock, pSupp->targetValSlot);
×
107
    SColumnInfoData* pTsCol = taosArrayGet(pBlock->pDataBlock, pSupp->inputTsSlot);
×
108
    if (pTsCol == NULL || pValCol == NULL) break;
×
109

110
    int32_t index = 0;
×
111
    int64_t ts = ((TSKEY*)pTsCol->pData)[j];
×
112
    char*   val = colDataGetData(pValCol, j);
×
113
    int16_t valType = pValCol->info.type;
×
114

115
    pSupp->minTs = MIN(pSupp->minTs, ts);
×
116
    pSupp->maxTs = MAX(pSupp->maxTs, ts);
×
117
    pSupp->numOfRows++;
×
118

119
    code = taosAnalyBufWriteColData(pBuf, index++, TSDB_DATA_TYPE_TIMESTAMP, &ts);
×
120
    if (TSDB_CODE_SUCCESS != code) {
×
121
      qError("%s failed to write ts in buf, code:%s", id, tstrerror(code));
×
122
      return code;
×
123
    }
124

125
    code = taosAnalyBufWriteColData(pBuf, index++, valType, val);
×
126
    if (TSDB_CODE_SUCCESS != code) {
×
127
      qError("%s failed to write val in buf, code:%s", id, tstrerror(code));
×
128
      return code;
×
129
    }
130

131
    for (int32_t i = 0; i < pSupp->numOfInputCols - 2; ++i) {
×
132
      SColumn*         pCol = taosArrayGet(pSupp->pCovariateSlotList, i);
×
133
      SColumnInfoData* pColData = taosArrayGet(pBlock->pDataBlock, pCol->slotId);
×
134

135
      char* pVal = colDataGetData(pColData, j);
×
136
      code = taosAnalyBufWriteColData(pBuf, index++, pCol->type, pVal);
×
137
      if (TSDB_CODE_SUCCESS != code) {
×
138
        qError("%s failed to write val in buf, code:%s", id, tstrerror(code));
×
139
        return code;
×
140
      }
141
    }
142
  }
143

144
  return 0;
×
145
}
146

147
static int32_t forecastCloseBuf(SForecastSupp* pSupp, const char* id) {
×
148
  SAnalyticBuf* pBuf = &pSupp->analyBuf;
×
149
  int32_t       code = 0;
×
150

151
  for (int32_t i = 0; i < pSupp->numOfInputCols; ++i) {
×
152
    code = taosAnalyBufWriteColEnd(pBuf, i);
×
153
    if (code != 0) return code;
×
154
  }
155

156
  code = taosAnalyBufWriteDataEnd(pBuf);
×
157
  if (code != 0) return code;
×
158

159
  code = taosAnalyBufWriteOptStr(pBuf, "option", pSupp->algoOpt);
×
160
  if (code != 0) return code;
×
161

162
  code = taosAnalyBufWriteOptStr(pBuf, "algo", pSupp->algoName);
×
163
  if (code != 0) return code;
×
164

165
  const char* prec = TSDB_TIME_PRECISION_MILLI_STR;
×
166
  if (pSupp->inputPrecision == TSDB_TIME_PRECISION_MICRO) prec = TSDB_TIME_PRECISION_MICRO_STR;
×
167
  if (pSupp->inputPrecision == TSDB_TIME_PRECISION_NANO) prec = TSDB_TIME_PRECISION_NANO_STR;
×
168
  code = taosAnalyBufWriteOptStr(pBuf, "prec", prec);
×
169
  if (code != 0) return code;
×
170

171
  code = taosAnalyBufWriteOptInt(pBuf, ALGO_OPT_WNCHECK_NAME, pSupp->wncheck);
×
172
  if (code != 0) return code;
×
173

174
  bool noConf = (pSupp->resHighSlot == -1 && pSupp->resLowSlot == -1);
×
175
  code = taosAnalyBufWriteOptInt(pBuf, ALGO_OPT_RETCONF_NAME, !noConf);
×
176
  if (code != 0) return code;
×
177

178
  if (pSupp->cachedRows < ANALY_FORECAST_MIN_ROWS) {
×
179
    qError("%s history rows for forecasting not enough, min required:%d, current:%" PRId64, id, ANALY_FORECAST_MIN_ROWS,
×
180
           pSupp->forecastRows);
181
    return TSDB_CODE_ANA_ANODE_NOT_ENOUGH_ROWS;
×
182
  }
183

184
  code = taosAnalyBufWriteOptInt(pBuf, "forecast_rows", pSupp->forecastRows);
×
185
  if (code != 0) return code;
×
186

187
  code = taosAnalyBufWriteOptFloat(pBuf, "conf", pSupp->conf);
×
188
  if (code != 0) return code;
×
189

190
  int32_t len = strlen(pSupp->algoOpt);
×
191
  int64_t every = (pSupp->setEvery != 0) ? pSupp->every : ((pSupp->maxTs - pSupp->minTs) / (pSupp->numOfRows - 1));
×
192
  code = taosAnalyBufWriteOptInt(pBuf, "every", every);
×
193
  if (code != 0) return code;
×
194

195
  int64_t start = (pSupp->setStart != 0) ? pSupp->startTs : pSupp->maxTs + every;
×
196
  code = taosAnalyBufWriteOptInt(pBuf, "start", start);
×
197
  if (code != 0) return code;
×
198

199
  if (taosArrayGetSize(pSupp->pCovariateSlotList) > 0) {
×
200
    code = taosAnalyBufWriteOptStr(pBuf, "type", "covariate");
×
201
    if (code != 0) return code;
×
202
  }
203

204
  code = taosAnalyBufClose(pBuf);
×
205
  return code;
×
206
}
207

208
static int32_t forecastAnalysis(SForecastSupp* pSupp, SSDataBlock* pBlock, const char* pId) {
×
209
  SAnalyticBuf* pBuf = &pSupp->analyBuf;
×
210
  int32_t       resCurRow = pBlock->info.rows;
×
211
  int8_t        tmpI8 = 0;
×
212
  int16_t       tmpI16 = 0;
×
213
  int32_t       tmpI32 = 0;
×
214
  int64_t       tmpI64 = 0;
×
215
  float         tmpFloat = 0;
×
216
  double        tmpDouble = 0;
×
217
  int32_t       code = 0;
×
218

219
  SColumnInfoData* pResValCol = taosArrayGet(pBlock->pDataBlock, pSupp->resValSlot);
×
220
  if (NULL == pResValCol) {
×
221
    return terrno;
×
222
  }
223

224
  SColumnInfoData* pResTsCol = ((pSupp->resTsSlot != -1) ? taosArrayGet(pBlock->pDataBlock, pSupp->resTsSlot) : NULL);
×
225
  SColumnInfoData* pResLowCol =
×
226
      ((pSupp->resLowSlot != -1) ? taosArrayGet(pBlock->pDataBlock, pSupp->resLowSlot) : NULL);
×
227
  SColumnInfoData* pResHighCol =
×
228
      (pSupp->resHighSlot != -1 ? taosArrayGet(pBlock->pDataBlock, pSupp->resHighSlot) : NULL);
×
229

230
  SJson* pJson = taosAnalySendReqRetJson(pSupp->algoUrl, ANALYTICS_HTTP_TYPE_POST, pBuf, pSupp->timeout * 1000);
×
231
  if (pJson == NULL) {
×
232
    return terrno;
×
233
  }
234

235
  int32_t rows = 0;
×
236
  tjsonGetInt32ValueFromDouble(pJson, "rows", rows, code);
×
237
  if (rows < 0 && code == 0) {
×
238
    char pMsg[1024] = {0};
×
239
    code = tjsonGetStringValue(pJson, "msg", pMsg);
×
240
    if (code != 0) {
×
241
      qError("%s failed to get msg from rsp, unknown error", pId);
×
242
    } else {
243
      qError("%s failed to exec forecast, msg:%s", pId, pMsg);
×
244
    }
245

246
    tjsonDelete(pJson);
×
247
    return TSDB_CODE_ANA_ANODE_RETURN_ERROR;
×
248
  }
249

250
  if (code < 0) {
×
251
    goto _OVER;
×
252
  }
253

254
  SJson* res = tjsonGetObjectItem(pJson, "res");
×
255
  if (res == NULL) goto _OVER;
×
256
  int32_t ressize = tjsonGetArraySize(res);
×
257
  bool    returnConf = (pSupp->resHighSlot != -1 || pSupp->resLowSlot != -1);
×
258

259
  if ((returnConf && (ressize != 4)) || ((!returnConf) && (ressize != 2))) {
×
260
    goto _OVER;
×
261
  }
262

263
  if (pResTsCol != NULL) {
×
264
    resCurRow = pBlock->info.rows;
×
265
    SJson* tsJsonArray = tjsonGetArrayItem(res, 0);
×
266
    if (tsJsonArray == NULL) goto _OVER;
×
267
    int32_t tsSize = tjsonGetArraySize(tsJsonArray);
×
268
    if (tsSize != rows) goto _OVER;
×
269
    for (int32_t i = 0; i < tsSize; ++i) {
×
270
      SJson* tsJson = tjsonGetArrayItem(tsJsonArray, i);
×
271
      tjsonGetObjectValueBigInt(tsJson, &tmpI64);
×
272
      colDataSetInt64(pResTsCol, resCurRow, &tmpI64);
×
273
      resCurRow++;
×
274
    }
275
  }
276

277
  if (pResLowCol != NULL) {
×
278
    resCurRow = pBlock->info.rows;
×
279
    SJson* lowJsonArray = tjsonGetArrayItem(res, 2);
×
280
    if (lowJsonArray == NULL) goto _OVER;
×
281
    int32_t lowSize = tjsonGetArraySize(lowJsonArray);
×
282
    if (lowSize != rows) goto _OVER;
×
283
    for (int32_t i = 0; i < lowSize; ++i) {
×
284
      SJson* lowJson = tjsonGetArrayItem(lowJsonArray, i);
×
285
      tjsonGetObjectValueDouble(lowJson, &tmpDouble);
×
286
      tmpFloat = (float)tmpDouble;
×
287
      colDataSetFloat(pResLowCol, resCurRow, &tmpFloat);
×
288
      resCurRow++;
×
289
    }
290
  }
291

292
  if (pResHighCol != NULL) {
×
293
    resCurRow = pBlock->info.rows;
×
294
    SJson* highJsonArray = tjsonGetArrayItem(res, 3);
×
295
    if (highJsonArray == NULL) goto _OVER;
×
296
    int32_t highSize = tjsonGetArraySize(highJsonArray);
×
297
    if (highSize != rows) goto _OVER;
×
298
    for (int32_t i = 0; i < highSize; ++i) {
×
299
      SJson* highJson = tjsonGetArrayItem(highJsonArray, i);
×
300
      tjsonGetObjectValueDouble(highJson, &tmpDouble);
×
301
      tmpFloat = (float)tmpDouble;
×
302
      colDataSetFloat(pResHighCol, resCurRow, &tmpFloat);
×
303
      resCurRow++;
×
304
    }
305
  }
306

307
  resCurRow = pBlock->info.rows;
×
308
  SJson* valJsonArray = tjsonGetArrayItem(res, 1);
×
309
  if (valJsonArray == NULL) goto _OVER;
×
310
  int32_t valSize = tjsonGetArraySize(valJsonArray);
×
311
  if (valSize != rows) goto _OVER;
×
312
  for (int32_t i = 0; i < valSize; ++i) {
×
313
    SJson* valJson = tjsonGetArrayItem(valJsonArray, i);
×
314
    tjsonGetObjectValueDouble(valJson, &tmpDouble);
×
315

316
    switch (pSupp->targetValType) {
×
317
      case TSDB_DATA_TYPE_BOOL:
×
318
      case TSDB_DATA_TYPE_UTINYINT:
319
      case TSDB_DATA_TYPE_TINYINT: {
320
        tmpI8 = (int8_t)tmpDouble;
×
321
        colDataSetInt8(pResValCol, resCurRow, &tmpI8);
×
322
        break;
×
323
      }
324
      case TSDB_DATA_TYPE_USMALLINT:
×
325
      case TSDB_DATA_TYPE_SMALLINT: {
326
        tmpI16 = (int16_t)tmpDouble;
×
327
        colDataSetInt16(pResValCol, resCurRow, &tmpI16);
×
328
        break;
×
329
      }
330
      case TSDB_DATA_TYPE_INT:
×
331
      case TSDB_DATA_TYPE_UINT: {
332
        tmpI32 = (int32_t)tmpDouble;
×
333
        colDataSetInt32(pResValCol, resCurRow, &tmpI32);
×
334
        break;
×
335
      }
336
      case TSDB_DATA_TYPE_TIMESTAMP:
×
337
      case TSDB_DATA_TYPE_UBIGINT:
338
      case TSDB_DATA_TYPE_BIGINT: {
339
        tmpI64 = (int64_t)tmpDouble;
×
340
        colDataSetInt64(pResValCol, resCurRow, &tmpI64);
×
341
        break;
×
342
      }
343
      case TSDB_DATA_TYPE_FLOAT: {
×
344
        tmpFloat = (float)tmpDouble;
×
345
        colDataSetFloat(pResValCol, resCurRow, &tmpFloat);
×
346
        break;
×
347
      }
348
      case TSDB_DATA_TYPE_DOUBLE: {
×
349
        colDataSetDouble(pResValCol, resCurRow, &tmpDouble);
×
350
        break;
×
351
      }
352
      default:
×
353
        code = TSDB_CODE_FUNC_FUNTION_PARA_TYPE;
×
354
        goto _OVER;
×
355
    }
356
    resCurRow++;
×
357
  }
358

359
  pBlock->info.rows += rows;
×
360

361
  if (pJson != NULL) tjsonDelete(pJson);
×
362
  return 0;
×
363

364
_OVER:
×
365
  tjsonDelete(pJson);
×
366
  if (code == 0) {
×
367
    code = TSDB_CODE_INVALID_JSON_FORMAT;
×
368
  }
369

370
  qError("%s failed to perform forecast finalize since %s", pId, tstrerror(code));
×
371
  return code;
×
372
}
373

374
static int32_t forecastAggregateBlocks(SForecastSupp* pSupp, SSDataBlock* pResBlock, const char* pId) {
×
375
  int32_t       code = TSDB_CODE_SUCCESS;
×
376
  int32_t       lino = 0;
×
377
  SAnalyticBuf* pBuf = &pSupp->analyBuf;
×
378

379
  code = forecastCloseBuf(pSupp, pId);
×
380
  QUERY_CHECK_CODE(code, lino, _end);
×
381

382
  code = forecastEnsureBlockCapacity(pResBlock, 1);
×
383
  QUERY_CHECK_CODE(code, lino, _end);
×
384

385
  code = forecastAnalysis(pSupp, pResBlock, pId);
×
386
  QUERY_CHECK_CODE(code, lino, _end);
×
387

388
  uInfo("%s block:%d, forecast finalize", pId, pSupp->numOfBlocks);
×
389

390
_end:
×
391
  pSupp->numOfBlocks = 0;
×
392
  taosAnalyBufDestroy(&pSupp->analyBuf);
×
393
  return code;
×
394
}
395

396
static int32_t forecastNext(SOperatorInfo* pOperator, SSDataBlock** ppRes) {
×
397
  int32_t                code = TSDB_CODE_SUCCESS;
×
398
  int32_t                lino = 0;
×
399
  SExecTaskInfo*         pTaskInfo = pOperator->pTaskInfo;
×
400
  SForecastOperatorInfo* pInfo = pOperator->info;
×
401
  SSDataBlock*           pResBlock = pInfo->pRes;
×
402
  SForecastSupp*         pSupp = &pInfo->forecastSupp;
×
403
  SAnalyticBuf*          pBuf = &pSupp->analyBuf;
×
404
  int64_t                st = taosGetTimestampUs();
×
405
  int32_t                numOfBlocks = pSupp->numOfBlocks;
×
406
  const char*            pId = GET_TASKID(pOperator->pTaskInfo);
×
407

408
  blockDataCleanup(pResBlock);
×
409

410
  while (1) {
×
411
    SSDataBlock* pBlock = getNextBlockFromDownstream(pOperator, 0);
×
412
    if (pBlock == NULL) {
×
413
      break;
×
414
    }
415

416
    if (pSupp->groupId == 0 || pSupp->groupId == pBlock->info.id.groupId) {
×
417
      pSupp->groupId = pBlock->info.id.groupId;
×
418
      numOfBlocks++;
×
419
      pSupp->cachedRows += pBlock->info.rows;
×
420
      qDebug("%s group:%" PRId64 ", blocks:%d, rows:%" PRId64 ", total rows:%" PRId64, pId, pSupp->groupId, numOfBlocks,
×
421
             pBlock->info.rows, pSupp->cachedRows);
422
      code = forecastCacheBlock(pSupp, pBlock, pId);
×
423
      QUERY_CHECK_CODE(code, lino, _end);
×
424
    } else {
425
      qDebug("%s group:%" PRId64 ", read finish for new group coming, blocks:%d", pId, pSupp->groupId, numOfBlocks);
×
426
      code = forecastAggregateBlocks(pSupp, pResBlock, pId);
×
427
      QUERY_CHECK_CODE(code, lino, _end);
×
428
      pSupp->groupId = pBlock->info.id.groupId;
×
429
      numOfBlocks = 1;
×
430
      pSupp->cachedRows = pBlock->info.rows;
×
431
      qDebug("%s group:%" PRId64 ", new group, rows:%" PRId64 ", total rows:%" PRId64, pId, pSupp->groupId,
×
432
             pBlock->info.rows, pSupp->cachedRows);
433
      code = forecastCacheBlock(pSupp, pBlock, pId);
×
434
      QUERY_CHECK_CODE(code, lino, _end);
×
435
    }
436

437
    if (pResBlock->info.rows > 0) {
×
438
      (*ppRes) = pResBlock;
×
439
      qDebug("%s group:%" PRId64 ", return to upstream, blocks:%d", pId, pResBlock->info.id.groupId, numOfBlocks);
×
440
      return code;
×
441
    }
442
  }
443

444
  if (numOfBlocks > 0) {
×
445
    qDebug("%s group:%" PRId64 ", read finish, blocks:%d", pId, pSupp->groupId, numOfBlocks);
×
446
    code = forecastAggregateBlocks(pSupp, pResBlock, pId);
×
447
    QUERY_CHECK_CODE(code, lino, _end);
×
448
  }
449

450
  int64_t cost = taosGetTimestampUs() - st;
×
451
  qDebug("%s all groups finished, cost:%" PRId64 "us", pId, cost);
×
452

453
_end:
×
454
  if (code != TSDB_CODE_SUCCESS) {
×
455
    qError("%s %s failed at line %d since %s", pId, __func__, lino, tstrerror(code));
×
456
    pTaskInfo->code = code;
×
457
    T_LONG_JMP(pTaskInfo->env, code);
×
458
  }
459

460
  (*ppRes) = (pResBlock->info.rows == 0) ? NULL : pResBlock;
×
461
  return code;
×
462
}
463

464
static int32_t forecastParseOutput(SForecastSupp* pSupp, SExprSupp* pExprSup) {
×
465
  pSupp->resLowSlot = -1;
×
466
  pSupp->resHighSlot = -1;
×
467
  pSupp->resTsSlot = -1;
×
468
  pSupp->resValSlot = -1;
×
469

470
  for (int32_t j = 0; j < pExprSup->numOfExprs; ++j) {
×
471
    SExprInfo* pExprInfo = &pExprSup->pExprInfo[j];
×
472
    int32_t    dstSlot = pExprInfo->base.resSchema.slotId;
×
473
    if (pExprInfo->pExpr->_function.functionType == FUNCTION_TYPE_FORECAST) {
×
474
      pSupp->resValSlot = dstSlot;
×
475
    } else if (pExprInfo->pExpr->_function.functionType == FUNCTION_TYPE_FORECAST_ROWTS) {
×
476
      pSupp->resTsSlot = dstSlot;
×
477
    } else if (pExprInfo->pExpr->_function.functionType == FUNCTION_TYPE_FORECAST_LOW) {
×
478
      pSupp->resLowSlot = dstSlot;
×
479
    } else if (pExprInfo->pExpr->_function.functionType == FUNCTION_TYPE_FORECAST_HIGH) {
×
480
      pSupp->resHighSlot = dstSlot;
×
481
    } else {
482
    }
483
  }
484

485
  return 0;
×
486
}
487

488
static int32_t validInputParams(SFunctionNode* pFunc, const char* id) {
×
489
  int32_t code = 0;
×
490
  int32_t lino = 0;
×
491
  int32_t num = LIST_LENGTH(pFunc->pParameterList);
×
492

493
  TSDB_CHECK_CONDITION(num > 1, code, lino, _end, TSDB_CODE_PLAN_INTERNAL_ERROR);
×
494

495
  for (int32_t i = 0; i < num; ++i) {
×
496
    SNode* p = nodesListGetNode(pFunc->pParameterList, i);
×
497
    TSDB_CHECK_NULL(p, code, lino, _end, TSDB_CODE_PLAN_INTERNAL_ERROR)
×
498
  }
499

500
  if (num == 2) {  // column_name, timestamp_column_name
×
501
    SNode* p1 = nodesListGetNode(pFunc->pParameterList, 0);
×
502
    SNode* p2 = nodesListGetNode(pFunc->pParameterList, 1);
×
503

504
    TSDB_CHECK_CONDITION(p1->type == QUERY_NODE_COLUMN, code, lino, _end, TSDB_CODE_PLAN_INTERNAL_ERROR);
×
505
    TSDB_CHECK_CONDITION(p2->type == QUERY_NODE_COLUMN, code, lino, _end, TSDB_CODE_PLAN_INTERNAL_ERROR);
×
506
  } else if (num >= 3) {  // column_name_#1, column_name_#2...., analytics_options, timestamp_column_name
×
507
    for (int32_t i = 0; i < num - 2; ++i) {
×
508
      SNode* p1 = nodesListGetNode(pFunc->pParameterList, i);
×
509
      if (p1->type != QUERY_NODE_COLUMN) {
×
510
        code = TSDB_CODE_PLAN_INTERNAL_ERROR;
×
511
        goto _end;
×
512
      }
513
    }
514

515
    SNode* p2 = nodesListGetNode(pFunc->pParameterList, num - 2);
×
516
    SNode* p3 = nodesListGetNode(pFunc->pParameterList, num - 1);
×
517

518
    TSDB_CHECK_CONDITION(p2->type == QUERY_NODE_VALUE, code, lino, _end, TSDB_CODE_PLAN_INTERNAL_ERROR);
×
519
    TSDB_CHECK_CONDITION(p3->type == QUERY_NODE_COLUMN, code, lino, _end, TSDB_CODE_PLAN_INTERNAL_ERROR);
×
520
  }
521

522
_end:
×
523
  if (code) {
×
524
    qError("%s valid the parameters failed, line:%d, code:%s", id, lino, tstrerror(code));
×
525
  }
526
  return code;
×
527
}
528

529
static int32_t forecastParseInput(SForecastSupp* pSupp, SNodeList* pFuncs, const char* id) {
×
530
  int32_t code = 0;
×
531
  SNode* pNode = NULL;
×
532
  pSupp->inputTsSlot = -1;
×
533
  pSupp->targetValSlot = -1;
×
534
  pSupp->targetValType = -1;
×
535
  pSupp->inputPrecision = -1;
×
536

537
  FOREACH(pNode, pFuncs) {
×
538
    if ((nodeType(pNode) == QUERY_NODE_TARGET) && (nodeType(((STargetNode*)pNode)->pExpr) == QUERY_NODE_FUNCTION)) {
×
539
      SFunctionNode* pFunc = (SFunctionNode*)((STargetNode*)pNode)->pExpr;
×
540
      int32_t        numOfParam = LIST_LENGTH(pFunc->pParameterList);
×
541

542
      if (pFunc->funcType == FUNCTION_TYPE_FORECAST) {
×
543
        code = validInputParams(pFunc, id);
×
544
        if (code) {
×
545
          return code;
×
546
        }
547

548
        pSupp->numOfInputCols = 2;
×
549

550
        if (numOfParam == 2) {
×
551
          // column, ts
552
          SColumnNode* pValNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, 0);
×
553
          SColumnNode* pTsNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, 1);
×
554

555
          pSupp->inputTsSlot = pTsNode->slotId;
×
556
          pSupp->inputPrecision = pTsNode->node.resType.precision;
×
557
          pSupp->targetValSlot = pValNode->slotId;
×
558
          pSupp->targetValType = pValNode->node.resType.type;
×
559

560
          // let's add the holtwinters as the default forecast algorithm
561
          tstrncpy(pSupp->algoOpt, "algo=holtwinters", TSDB_ANALYTIC_ALGO_OPTION_LEN);
×
562
        } else if (numOfParam == 3) {
×
563
          // column, options, ts
564
          SColumnNode* pValNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, 0);
×
565
          SValueNode*  pOptNode = (SValueNode*)nodesListGetNode(pFunc->pParameterList, 1);
×
566
          SColumnNode* pTsNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, 2);
×
567

568
          pSupp->inputTsSlot = pTsNode->slotId;
×
569
          pSupp->inputPrecision = pTsNode->node.resType.precision;
×
570
          pSupp->targetValSlot = pValNode->slotId;
×
571
          pSupp->targetValType = pValNode->node.resType.type;
×
572
          tstrncpy(pSupp->algoOpt, pOptNode->literal, sizeof(pSupp->algoOpt));
×
573
        } else {
574
          SColumnNode* pValNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, 0);
×
575
          SValueNode*  pOptNode = (SValueNode*)nodesListGetNode(pFunc->pParameterList, numOfParam - 2);
×
576
          SColumnNode* pTsNode = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, numOfParam - 1);
×
577

578
          pSupp->inputTsSlot = pTsNode->slotId;
×
579
          pSupp->inputPrecision = pTsNode->node.resType.precision;
×
580

581
          pSupp->targetValSlot = pValNode->slotId;
×
582
          pSupp->targetValType = pValNode->node.resType.type;
×
583
          tstrncpy(pSupp->algoOpt, pOptNode->literal, sizeof(pSupp->algoOpt));
×
584

585
          pSupp->pCovariateSlotList = taosArrayInit(4, sizeof(SColumn));
×
586
          for(int32_t i = 1; i < numOfParam - 2; ++i) {
×
587
            SColumnNode* p = (SColumnNode*)nodesListGetNode(pFunc->pParameterList, i);
×
588
            SColumn col = {.slotId = p->slotId, .colType = p->colType, .type = p->node.resType.type};
×
589
            taosArrayPush(pSupp->pCovariateSlotList, &col);
×
590
          }
591

592
          pSupp->numOfInputCols += (numOfParam - 3);
×
593
        }
594
      }
595
    }
596
  }
597

598
  return 0;
×
599
}
600

601
static void initForecastOpt(SForecastSupp* pSupp) {
×
602
  pSupp->maxTs = 0;
×
603
  pSupp->minTs = INT64_MAX;
×
604
  pSupp->numOfRows = 0;
×
605
  pSupp->wncheck = ANALY_FORECAST_DEFAULT_WNCHECK;
×
606
  pSupp->forecastRows = ANALY_FORECAST_DEFAULT_ROWS;
×
607
  pSupp->conf = ANALY_FORECAST_DEFAULT_CONF;
×
608
  pSupp->setEvery = 0;
×
609
  pSupp->setStart = 0;
×
610
}
×
611

612
static int32_t filterNotSupportForecast(SForecastSupp* pSupp) {
×
613
  if (taosArrayGetSize(pSupp->pCovariateSlotList) > 0) {
×
614
    if (taosStrcasecmp(pSupp->algoName, "holtwinters") == 0) {
×
615
      return TSDB_CODE_ANA_NOT_SUPPORT_FORECAST;
×
616
    } else if (taosStrcasecmp(pSupp->algoName, "arima") == 0) {
×
617
      return TSDB_CODE_ANA_NOT_SUPPORT_FORECAST;
×
618
    } else if (taosStrcasecmp(pSupp->algoName, "timemoe-fc") == 0) {
×
619
      return TSDB_CODE_ANA_NOT_SUPPORT_FORECAST;
×
620
    }
621
  }
622

623
  return TSDB_CODE_SUCCESS;
×
624
}
625

626

627
static int32_t forecastParseOpt(SForecastSupp* pSupp, const char* id) {
×
628
  int32_t   code = 0;
×
629
  int32_t   lino = 0;
×
630
  SHashObj* pHashMap = NULL;
×
631

632
  initForecastOpt(pSupp);
×
633

634
  code = taosAnalyGetOpts(pSupp->algoOpt, &pHashMap);
×
635
  if (code != TSDB_CODE_SUCCESS) {
×
636
    return code;
×
637
  }
638

639
  code = taosAnalysisParseAlgo(pSupp->algoOpt, pSupp->algoName, pSupp->algoUrl, ANALY_ALGO_TYPE_FORECAST,
×
640
                               tListLen(pSupp->algoUrl), pHashMap, id);
641
  TSDB_CHECK_CODE(code, lino, _end);
×
642

643
  code = filterNotSupportForecast(pSupp);
×
644
  if (code) {
×
645
    qError("%s not support forecast model, %s", id, pSupp->algoName);
×
646
    TSDB_CHECK_CODE(code, lino, _end);
×
647
  }
648

649
  // extract the timeout parameter
650
  pSupp->timeout = taosAnalysisParseTimout(pHashMap, id);
×
651
  pSupp->wncheck = taosAnalysisParseWncheck(pHashMap, id);
×
652

653
  char* pRows = taosHashGet(pHashMap, ALGO_OPT_FORECASTROWS_NAME, strlen(ALGO_OPT_FORECASTROWS_NAME));
×
654
  if (pRows != NULL) {
×
655
    int64_t v = 0;
×
656
    code = toInteger(pRows, taosHashGetValueSize(pRows), 10, &v);
×
657

658
    pSupp->forecastRows = v;
×
659
    qDebug("%s forecast rows:%"PRId64, id, pSupp->forecastRows);
×
660
  } else {
661
    qDebug("%s forecast rows not found:%s, use default:%" PRId64, id, pSupp->algoOpt, pSupp->forecastRows);
×
662
  }
663

664
  if (pSupp->forecastRows > ANALY_FORECAST_RES_MAX_ROWS) {
×
665
    qError("%s required too many forecast rows, max allowed:%d, required:%" PRId64, id, ANALY_FORECAST_RES_MAX_ROWS,
×
666
           pSupp->forecastRows);
667
    code = TSDB_CODE_ANA_ANODE_TOO_MANY_ROWS;
×
668
    goto _end;
×
669
  }
670

671
  char* pConf = taosHashGet(pHashMap, ALGO_OPT_CONF_NAME, strlen(ALGO_OPT_CONF_NAME));
×
672
  if (pConf != NULL) {
×
673
    char* endPtr = NULL;
×
674
    double v = taosStr2Double(pConf, &endPtr);
×
675
    pSupp->conf = v;
×
676

677
    if (v <= 0 || v > 1.0) {
×
678
      pSupp->conf = ANALY_FORECAST_DEFAULT_CONF;
×
679
      qWarn("%s valid conf range is (0, 1], user specified:%.2f out of range, set the default:%.2f", id, v,
×
680
             pSupp->conf);
681
    } else {
682
      qDebug("%s forecast conf:%.2f", id, pSupp->conf);
×
683
    }
684
  } else {
685
    qDebug("%s forecast conf not found:%s, use default:%.2f", id, pSupp->algoOpt, pSupp->conf);
×
686
  }
687

688
  char* pStart = taosHashGet(pHashMap, ALGO_OPT_START_NAME, strlen(ALGO_OPT_START_NAME));
×
689
  if (pStart != NULL) {
×
690
    int64_t v = 0;
×
691
    code = toInteger(pStart, taosHashGetValueSize(pStart), 10, &v);
×
692
    pSupp->startTs = v;
×
693
    pSupp->setStart = 1;
×
694
    qDebug("%s forecast set start ts:%"PRId64, id, pSupp->startTs);
×
695
  }
696

697
  char* pEvery = taosHashGet(pHashMap, ALGO_OPT_EVERY_NAME, strlen(ALGO_OPT_EVERY_NAME));
×
698
  if (pEvery != NULL) {
×
699
    int64_t v = 0;
×
700
    code = toInteger(pEvery, taosHashGetValueSize(pEvery), 10, &v);
×
701
    pSupp->every = v;
×
702
    pSupp->setEvery = 1;
×
703
    qDebug("%s forecast set every ts:%"PRId64, id, pSupp->every);
×
704
  }
705

706
_end:
×
707
  taosHashCleanup(pHashMap);
×
708
  return code;
×
709
}
710

711
static int32_t forecastCreateBuf(SForecastSupp* pSupp) {
×
712
  SAnalyticBuf* pBuf = &pSupp->analyBuf;
×
713
  int64_t       ts = 0;  // taosGetTimestampMs();
×
714
  int32_t       index = 0;
×
715

716
  pBuf->bufType = ANALYTICS_BUF_TYPE_JSON_COL;
×
717
  snprintf(pBuf->fileName, sizeof(pBuf->fileName), "%s/tdengine-forecast-%" PRId64, tsTempDir, ts);
×
718

719
  int32_t numOfCols = taosArrayGetSize(pSupp->pCovariateSlotList) + 2;
×
720

721
  int32_t code = tsosAnalyBufOpen(pBuf, numOfCols);
×
722
  if (code != 0) goto _OVER;
×
723

724
  code = taosAnalyBufWriteColMeta(pBuf, index++, TSDB_DATA_TYPE_TIMESTAMP, "ts");
×
725
  if (code != 0) goto _OVER;
×
726

727
  code = taosAnalyBufWriteColMeta(pBuf, index++, pSupp->targetValType, "val");
×
728
  if (code != 0) goto _OVER;
×
729

730
  for(int32_t i = 0; i < taosArrayGetSize(pSupp->pCovariateSlotList); ++i) {
×
731
    SColumn* pCol = taosArrayGet(pSupp->pCovariateSlotList, i);
×
732

733
    char name[128] = {0};
×
734
    (void) tsnprintf(name, tListLen(name), "past_dynamic_real_%d", i + 1);
×
735

736
    code = taosAnalyBufWriteColMeta(pBuf, index++, pCol->type, name);
×
737
    if (code) {
×
738
      goto _OVER;
×
739
    }
740
  }
741

742
  code = taosAnalyBufWriteDataBegin(pBuf);
×
743
  if (code != 0) goto _OVER;
×
744

745
  for (int32_t i = 0; i < pSupp->numOfInputCols; ++i) {
×
746
    code = taosAnalyBufWriteColBegin(pBuf, i);
×
747
    if (code != 0) goto _OVER;
×
748
  }
749

750
_OVER:
×
751
  if (code != 0) {
×
752
    (void)taosAnalyBufClose(pBuf);
×
753
    taosAnalyBufDestroy(pBuf);
×
754
  }
755
  return code;
×
756
}
757

758
int32_t createForecastOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo,
×
759
                                   SOperatorInfo** pOptrInfo) {
760
  QRY_PARAM_CHECK(pOptrInfo);
×
761

762
  int32_t                code = 0;
×
763
  int32_t                lino = 0;
×
764
  SForecastOperatorInfo* pInfo = taosMemoryCalloc(1, sizeof(SForecastOperatorInfo));
×
765
  SOperatorInfo*         pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
×
766
  if (pOperator == NULL || pInfo == NULL) {
×
767
    code = terrno;
×
768
    goto _error;
×
769
  }
770

771
  const char*             pId = pTaskInfo->id.str;
×
772
  SForecastSupp*          pSupp = &pInfo->forecastSupp;
×
773
  SForecastFuncPhysiNode* pForecastPhyNode = (SForecastFuncPhysiNode*)pPhyNode;
×
774
  SExprSupp*              pExprSup = &pOperator->exprSupp;
×
775
  int32_t                 numOfExprs = 0;
×
776
  SExprInfo*              pExprInfo = NULL;
×
777

778
  code = createExprInfo(pForecastPhyNode->pFuncs, NULL, &pExprInfo, &numOfExprs);
×
779
  QUERY_CHECK_CODE(code, lino, _error);
×
780

781
  code = initExprSupp(pExprSup, pExprInfo, numOfExprs, &pTaskInfo->storageAPI.functionStore);
×
782
  QUERY_CHECK_CODE(code, lino, _error);
×
783

784
  if (pForecastPhyNode->pExprs != NULL) {
×
785
    int32_t    num = 0;
×
786
    SExprInfo* pScalarExprInfo = NULL;
×
787
    code = createExprInfo(pForecastPhyNode->pExprs, NULL, &pScalarExprInfo, &num);
×
788
    QUERY_CHECK_CODE(code, lino, _error);
×
789

790
    code = initExprSupp(&pInfo->scalarSup, pScalarExprInfo, num, &pTaskInfo->storageAPI.functionStore);
×
791
    QUERY_CHECK_CODE(code, lino, _error);
×
792
  }
793

794
  code = filterInitFromNode((SNode*)pForecastPhyNode->node.pConditions, &pOperator->exprSupp.pFilterInfo, 0);
×
795
  QUERY_CHECK_CODE(code, lino, _error);
×
796

797
  code = forecastParseInput(pSupp, pForecastPhyNode->pFuncs, pId);
×
798
  QUERY_CHECK_CODE(code, lino, _error);
×
799

800
  code = forecastParseOutput(pSupp, pExprSup);
×
801
  QUERY_CHECK_CODE(code, lino, _error);
×
802

803
  code = forecastParseOpt(pSupp, pId);
×
804
  QUERY_CHECK_CODE(code, lino, _error);
×
805

806
  code = forecastCreateBuf(pSupp);
×
807
  QUERY_CHECK_CODE(code, lino, _error);
×
808

809
  initResultSizeInfo(&pOperator->resultInfo, 4096);
×
810

811
  pInfo->pRes = createDataBlockFromDescNode(pPhyNode->pOutputDataBlockDesc);
×
812
  QUERY_CHECK_NULL(pInfo->pRes, code, lino, _error, terrno);
×
813

814
  setOperatorInfo(pOperator, "ForecastOperator", QUERY_NODE_PHYSICAL_PLAN_FORECAST_FUNC, false, OP_NOT_OPENED, pInfo,
×
815
                  pTaskInfo);
816
  pOperator->fpSet = createOperatorFpSet(optrDummyOpenFn, forecastNext, NULL, destroyForecastInfo, optrDefaultBufFn,
×
817
                                         NULL, optrDefaultGetNextExtFn, NULL);
818

819
  code = blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
×
820
  QUERY_CHECK_CODE(code, lino, _error);
×
821

822
  code = appendDownstream(pOperator, &downstream, 1);
×
823
  QUERY_CHECK_CODE(code, lino, _error);
×
824

825
  *pOptrInfo = pOperator;
×
826

827
  qDebug("%s forecast env is initialized, option:%s", pId, pSupp->algoOpt);
×
828
  return TSDB_CODE_SUCCESS;
×
829

830
_error:
×
831
  if (code != TSDB_CODE_SUCCESS) {
×
832
    qError("%s %s failed at line %d since %s", pId, __func__, lino, tstrerror(code));
×
833
  }
834
  if (pInfo != NULL) destroyForecastInfo(pInfo);
×
835
  destroyOperatorAndDownstreams(pOperator, &downstream, 1);
×
836
  pTaskInfo->code = code;
×
837
  return code;
×
838
}
839

840
static void destroyForecastInfo(void* param) {
×
841
  SForecastOperatorInfo* pInfo = (SForecastOperatorInfo*)param;
×
842

843
  blockDataDestroy(pInfo->pRes);
×
844
  pInfo->pRes = NULL;
×
845

846
  taosArrayDestroy(pInfo->forecastSupp.pCovariateSlotList);
×
847
  pInfo->forecastSupp.pCovariateSlotList = NULL;
×
848

849
  cleanupExprSupp(&pInfo->scalarSup);
×
850
  taosAnalyBufDestroy(&pInfo->forecastSupp.analyBuf);
×
851
  taosMemoryFreeClear(param);
×
852
}
×
853

854
#else
855

856
int32_t createForecastOperatorInfo(SOperatorInfo* downstream, SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo,
857
                                   SOperatorInfo** pOptrInfo) {
858
  return TSDB_CODE_OPS_NOT_SUPPORT;
859
}
860

861
#endif
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc