qExecutor.h 14.9 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/>.
 */
H
Haojun Liao 已提交
15 16
#ifndef TDENGINE_QEXECUTOR_H
#define TDENGINE_QEXECUTOR_H
17 18 19 20

#include "os.h"

#include "hash.h"
H
Haojun Liao 已提交
21
#include "qAggMain.h"
H
Haojun Liao 已提交
22 23
#include "qFill.h"
#include "qResultbuf.h"
24
#include "qSqlparser.h"
H
Haojun Liao 已提交
25
#include "qTsbuf.h"
26
#include "query.h"
27
#include "taosdef.h"
H
Haojun Liao 已提交
28
#include "tarray.h"
B
Bomin Zhang 已提交
29
#include "tlockfree.h"
H
Haojun Liao 已提交
30
#include "tsdb.h"
31 32

struct SColumnFilterElem;
33
typedef bool (*__filter_func_t)(struct SColumnFilterElem* pFilter, const char* val1, const char* val2, int16_t type);
H
hjxilinx 已提交
34
typedef int32_t (*__block_search_fn_t)(char* data, int32_t num, int64_t key, int32_t order);
35

H
Haojun Liao 已提交
36 37 38 39 40 41
#define IS_QUERY_KILLED(_q) ((_q)->code == TSDB_CODE_TSC_QUERY_CANCELLED)
#define Q_STATUS_EQUAL(p, s)  (((p) & (s)) != 0u)
#define QUERY_IS_ASC_QUERY(q) (GET_FORWARD_DIRECTION_FACTOR((q)->order.order) == QUERY_ASC_FORWARD_STEP)

#define GET_TABLEGROUP(q, _index)   ((SArray*) taosArrayGetP((q)->tableqinfoGroupInfo.pGroupList, (_index)))

H
Haojun Liao 已提交
42 43
#define GET_NUM_OF_RESULTS(_r) (((_r)->outputBuf) == NULL? 0:((_r)->outputBuf)->info.rows)

H
Haojun Liao 已提交
44 45 46 47 48 49 50 51
enum {
  // when query starts to execute, this status will set
      QUERY_NOT_COMPLETED = 0x1u,

  /* query is over
   * 1. this status is used in one row result query process, e.g., count/sum/first/last/ avg...etc.
   * 2. when all data within queried time window, it is also denoted as query_completed
   */
H
Haojun Liao 已提交
52
      QUERY_COMPLETED = 0x2u,
H
Haojun Liao 已提交
53 54 55 56

  /* when the result is not completed return to client, this status will be
   * usually used in case of interval query with interpolation option
   */
H
Haojun Liao 已提交
57
      QUERY_OVER = 0x4u,
H
Haojun Liao 已提交
58 59
};

H
Haojun Liao 已提交
60
typedef struct SResultRowPool {
61 62 63 64 65 66 67 68 69 70
  int32_t elemSize;
  int32_t blockSize;
  int32_t numOfElemPerBlock;

  struct {
    int32_t blockIndex;
    int32_t pos;
  } position;

  SArray* pData;    // SArray<void*>
H
Haojun Liao 已提交
71
} SResultRowPool;
72

73
typedef struct SSqlGroupbyExpr {
H
Haojun Liao 已提交
74 75 76 77 78
  int16_t tableIndex;
  SArray* columnInfo;  // SArray<SColIndex>, group by columns information
  int16_t numOfGroupCols;
  int16_t orderIndex;  // order by column index
  int16_t orderType;   // order by type: asc/desc
79 80
} SSqlGroupbyExpr;

H
Haojun Liao 已提交
81
typedef struct SResultRow {
82
  int32_t       pageId;      // pageId & rowId is the position of current result in disk-based output buffer
H
Haojun Liao 已提交
83
  int32_t       offset:29;    // row index in buffer page
84 85 86 87
  bool          startInterp; // the time window start timestamp has done the interpolation already.
  bool          endInterp;   // the time window end timestamp has done the interpolation already.
  bool          closed;      // this result status: closed or opened
  uint32_t      numOfRows;   // number of rows of current time window
H
Haojun Liao 已提交
88
  SResultRowCellInfo*  pCellInfo;  // For each result column, there is a resultInfo
H
Haojun Liao 已提交
89
  union {STimeWindow win; char* key;};  // start key of current result row
H
Haojun Liao 已提交
90
} SResultRow;
91

92
typedef struct SGroupResInfo {
H
Haojun Liao 已提交
93 94
  int32_t totalGroup;
  int32_t currentGroup;
95 96 97 98
  int32_t index;
  SArray* pRows;      // SArray<SResultRow*>
} SGroupResInfo;

H
Haojun Liao 已提交
99 100 101 102
/**
 * If the number of generated results is greater than this value,
 * query query will be halt and return results to client immediate.
 */
H
Haojun Liao 已提交
103
typedef struct SRspResultInfo {
H
Haojun Liao 已提交
104
  int64_t total;      // total generated result size in rows
H
Haojun Liao 已提交
105
  int32_t capacity;   // capacity of current result output buffer
H
Haojun Liao 已提交
106
  int32_t threshold;  // result size threshold in rows.
H
Haojun Liao 已提交
107
} SRspResultInfo;
108

H
Haojun Liao 已提交
109
typedef struct SResultRowInfo {
H
Haojun Liao 已提交
110 111 112 113 114 115
  SResultRow** pResult;    // result list
  int16_t      type:8;     // data type for hash key
  int32_t      size:24;    // number of result set
  int32_t      capacity;   // max capacity
  int32_t      curIndex;   // current start active index
  int64_t      prevSKey;   // previous (not completed) sliding window start key
H
Haojun Liao 已提交
116
} SResultRowInfo;
117 118 119 120 121 122 123 124

typedef struct SColumnFilterElem {
  int16_t           bytes;  // column length
  __filter_func_t   fp;
  SColumnFilterInfo filterInfo;
} SColumnFilterElem;

typedef struct SSingleColumnFilterInfo {
H
Haojun Liao 已提交
125
  void*              pData;
126
  int32_t            numOfFilters;
H
Haojun Liao 已提交
127
  SColumnInfo        info;
128 129 130
  SColumnFilterElem* pFilters;
} SSingleColumnFilterInfo;

H
Haojun Liao 已提交
131
typedef struct STableQueryInfo {
H
hjxilinx 已提交
132
  TSKEY       lastKey;
H
Haojun Liao 已提交
133
  int32_t     groupIndex;     // group id in table list
134
  tVariant    tag;
H
hjxilinx 已提交
135
  STimeWindow win;
136
  STSCursor   cur;
H
Haojun Liao 已提交
137
  void*       pTable;         // for retrieve the page id list
H
Haojun Liao 已提交
138
  SResultRowInfo resInfo;
139 140
} STableQueryInfo;

H
Haojun Liao 已提交
141 142 143 144 145 146 147 148 149
typedef struct SQueryCostInfo {
  uint64_t loadStatisTime;
  uint64_t loadFileBlockTime;
  uint64_t loadDataInCacheTime;
  uint64_t loadStatisSize;
  uint64_t loadFileBlockSize;
  uint64_t loadDataInCacheSize;
  
  uint64_t loadDataTime;
150 151 152 153
  uint64_t totalRows;
  uint64_t totalCheckedRows;
  uint32_t totalBlocks;
  uint32_t loadBlocks;
H
Haojun Liao 已提交
154 155
  uint32_t loadBlockStatis;
  uint32_t discardBlocks;
156
  uint64_t elapsedTime;
H
Haojun Liao 已提交
157
  uint64_t firstStageMergeTime;
H
Haojun Liao 已提交
158 159
  uint64_t winInfoSize;
  uint64_t tableInfoSize;
H
Haojun Liao 已提交
160
  uint64_t hashSize;
161
  uint64_t numOfTimeWindows;
H
Haojun Liao 已提交
162
} SQueryCostInfo;
H
hjxilinx 已提交
163

H
Haojun Liao 已提交
164 165 166 167 168
typedef struct {
  int64_t vgroupLimit;
  int64_t ts;
} SOrderedPrjQueryInfo;

169 170 171 172 173
typedef struct {
  char*   tags;
  SArray* pResult;  // SArray<SStddevInterResult>
} SInterResult;

H
Haojun Liao 已提交
174 175 176 177 178 179
typedef struct SSDataBlock {
  SDataStatis *pBlockStatis;
  SArray      *pDataBlock;
  SDataBlockInfo info;
} SSDataBlock;

180
typedef struct SQuery {
H
Haojun Liao 已提交
181 182
  SLimitVal        limit;

H
Haojun Liao 已提交
183 184 185 186 187 188 189 190 191
  bool             stableQuery;      // super table query or not
  bool             topBotQuery;      // TODO used bitwise flag
  bool             groupbyColumn;    // denote if this is a groupby normal column query
  bool             hasTagResults;    // if there are tag values in final result or not
  bool             timeWindowInterpo;// if the time window start/end required interpolation
  bool             queryWindowIdentical; // all query time windows are identical for all tables in one group
  bool             queryBlockDist;    // if query data block distribution
  bool             stabledev;        // super table stddev query
  int32_t          interBufSize;     // intermediate buffer sizse
H
Haojun Liao 已提交
192 193

  SOrderVal        order;
H
Haojun Liao 已提交
194 195
  int16_t          numOfCols;
  int16_t          numOfTags;
H
Haojun Liao 已提交
196

H
Haojun Liao 已提交
197
  STimeWindow      window;
198
  SInterval        interval;
199
  int16_t          precision;
H
Haojun Liao 已提交
200 201
  int16_t          numOfOutput;
  int16_t          fillType;
202 203 204 205
  int16_t          checkResultBuf;   // check if the buffer is full during scan each block

  int32_t          srcRowSize;       // todo extract struct
  int32_t          resultRowSize;
H
Haojun Liao 已提交
206
  int32_t          intermediateResultRowSize; // intermediate result row size, in case of top-k query.
207 208
  int32_t          maxSrcColumnSize;
  int32_t          tagLen;           // tag value length of current query
H
Haojun Liao 已提交
209
  SSqlGroupbyExpr* pGroupbyExpr;
H
Haojun Liao 已提交
210
  SExprInfo*       pExpr1;
H
Haojun Liao 已提交
211 212
  SExprInfo*       pExpr2;
  int32_t          numOfExpr2;
H
Haojun Liao 已提交
213 214 215 216
  SColumnInfo*     colList;
  SColumnInfo*     tagColList;
  int32_t          numOfFilterCols;
  int64_t*         fillVal;
H
Haojun Liao 已提交
217 218 219
  SOrderedPrjQueryInfo prjInfo;        // limit value for each vgroup, only available in global order projection query.
  SSingleColumnFilterInfo* pFilterInfo;

H
Haojun Liao 已提交
220
  STableQueryInfo* current;
H
Haojun Liao 已提交
221 222
  void*            tsdb;
  SMemRef          memRef;
H
Haojun Liao 已提交
223
  STableGroupInfo  tableGroupInfo;       // table <tid, last_key> list  SArray<STableKeyInfo>
H
Haojun Liao 已提交
224
  int32_t          vgId;
225 226
} SQuery;

H
Haojun Liao 已提交
227
typedef SSDataBlock* (*__operator_fn_t)(void* param);
H
Haojun Liao 已提交
228
typedef void (*__optr_cleanup_fn_t)(void* param, int32_t num);
H
Haojun Liao 已提交
229

H
Haojun Liao 已提交
230
struct SOperatorInfo;
H
Haojun Liao 已提交
231

232
typedef struct SQueryRuntimeEnv {
H
Haojun Liao 已提交
233 234
  jmp_buf               env;
  SQuery*               pQuery;
H
Haojun Liao 已提交
235
  uint32_t              status;           // query status
H
Haojun Liao 已提交
236
  void*                 qinfo;
H
Haojun Liao 已提交
237
  uint8_t               scanFlag;         // denotes reversed scan of data or not
H
Haojun Liao 已提交
238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
  void*                 pQueryHandle;

  int32_t               prevGroupId;      // previous executed group id
  SDiskbasedResultBuf*  pResultBuf;       // query result buffer based on blocked-wised disk file
  SHashObj*             pResultRowHashTable; // quick locate the window object for each result
  char*                 keyBuf;           // window key buffer
  SResultRowPool*       pool;             // window result object pool
  char**                prevRow;

  SArray*               prevResult;       // intermediate result, SArray<SInterResult>
  STSBuf*               pTsBuf;           // timestamp filter list
  STSCursor             cur;

  char*                 tagVal;           // tag value of current data block
  SArithmeticSupport   *sasArray;

  SSDataBlock          *outputBuf;
  STableGroupInfo       tableqinfoGroupInfo;  // this is a group array list, including SArray<STableQueryInfo*> structure
H
Haojun Liao 已提交
256 257 258 259
  struct SOperatorInfo *proot;
  struct SOperatorInfo *pTableScanner;   // table scan operator
  SGroupResInfo         groupResInfo;
  int64_t               currentOffset;   // dynamic offset value
H
Haojun Liao 已提交
260

H
Haojun Liao 已提交
261
  SRspResultInfo        resultInfo;
H
Haojun Liao 已提交
262
  SHashObj             *pTableRetrieveTsMap;
263 264
} SQueryRuntimeEnv;

H
Haojun Liao 已提交
265 266 267 268 269 270
enum {
  OP_IN_EXECUTING   = 1,
  OP_RES_TO_RETURN  = 2,
  OP_EXEC_DONE      = 3,
};

H
Haojun Liao 已提交
271 272 273 274 275
enum OPERATOR_TYPE_E {
  OP_TableScan         = 1,
  OP_DataBlocksOptScan = 2,
  OP_TableSeqScan      = 3,
  OP_TagScan           = 4,
H
Haojun Liao 已提交
276 277 278 279 280 281 282 283 284 285
  OP_TableBlockInfoScan= 5,
  OP_Aggregate         = 6,
  OP_Arithmetic        = 7,
  OP_Groupby           = 8,
  OP_Limit             = 9,
  OP_Offset            = 10,
  OP_TimeInterval      = 11,
  OP_Fill              = 12,
  OP_MultiTableAggregate     = 13,
  OP_MultiTableTimeInterval  = 14,
H
Haojun Liao 已提交
286 287
};

H
Haojun Liao 已提交
288
typedef struct SOperatorInfo {
H
Haojun Liao 已提交
289 290 291 292 293 294 295 296
  uint8_t               operatorType;
  bool                  blockingOptr;  // block operator or not
  uint8_t               status;        // denote if current operator is completed
  int32_t               numOfOutput;   // number of columns of the current operator results
  char                 *name;          // name, used to show the query execution plan
  void                 *info;          // extension attribution
  SExprInfo            *pExpr;
  SQueryRuntimeEnv     *pRuntimeEnv;
H
Haojun Liao 已提交
297

H
Haojun Liao 已提交
298
  struct SOperatorInfo *upstream;
H
Haojun Liao 已提交
299
  __operator_fn_t       exec;
H
Haojun Liao 已提交
300
  __optr_cleanup_fn_t   cleanup;
H
Haojun Liao 已提交
301 302
} SOperatorInfo;

303 304 305 306 307
enum {
  QUERY_RESULT_NOT_READY = 1,
  QUERY_RESULT_READY     = 2,
};

308
typedef struct SQInfo {
H
Haojun Liao 已提交
309
  void*            signature;
H
Haojun Liao 已提交
310 311
  int32_t          code;        // error code to returned to client
  int64_t          owner;       // if it is in execution
H
Haojun Liao 已提交
312

313
  SQueryRuntimeEnv runtimeEnv;
H
Haojun Liao 已提交
314
  SQuery           query;
315
  void*            pBuf;        // allocated buffer for STableQueryInfo, sizeof(STableQueryInfo)*numOfTables;
H
Haojun Liao 已提交
316

317
  pthread_mutex_t  lock;        // used to synchronize the rsp/query threads
H
Haojun Liao 已提交
318
  tsem_t           ready;
319 320
  int32_t          dataReady;   // denote if query result is ready or not
  void*            rspContext;  // response context
321
  int64_t          startExecTs; // start to exec timestamp
322
  char*            sql;         // query sql string
H
Haojun Liao 已提交
323
  SQueryCostInfo   summary;
324 325
} SQInfo;

H
Haojun Liao 已提交
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341
typedef struct SQueryParam {
  char            *sql;
  char            *tagCond;
  char            *tbnameCond;
  char            *prevResult;
  SArray          *pTableIdList;
  SSqlFuncMsg    **pExprMsg;
  SSqlFuncMsg    **pSecExprMsg;
  SExprInfo       *pExprs;
  SExprInfo       *pSecExprs;

  SColIndex       *pGroupColIndex;
  SColumnInfo     *pTagColumnInfo;
  SSqlGroupbyExpr *pGroupbyExpr;
} SQueryParam;

H
Haojun Liao 已提交
342
typedef struct STableScanInfo {
H
Haojun Liao 已提交
343 344 345 346 347 348 349 350 351 352 353 354
  void           *pQueryHandle;
  int32_t         numOfBlocks;
  int32_t         numOfSkipped;
  int32_t         numOfBlockStatis;
  int64_t         numOfRows;
                 
  int32_t         order;        // scan order
  int32_t         times;        // repeat counts
  int32_t         current;
  int32_t         reverseTimes; // 0 by default

  SQLFunctionCtx *pCtx;         // next operator query context
H
Haojun Liao 已提交
355
  SResultRowInfo *pResultRowInfo;
H
Haojun Liao 已提交
356 357
  int32_t        *rowCellInfoOffset;
  SExprInfo      *pExpr;
H
Haojun Liao 已提交
358 359
  SSDataBlock     block;
  bool            loadExternalRows; // load external rows (prev & next rows)
H
Haojun Liao 已提交
360
  int32_t         numOfOutput;
H
Haojun Liao 已提交
361
  int64_t         elapsedTime;
H
Haojun Liao 已提交
362

H
Haojun Liao 已提交
363
  int32_t         tableIndex;
H
Haojun Liao 已提交
364 365
} STableScanInfo;

366 367 368
typedef struct STagScanInfo {
  SColumnInfo* pCols;
  SSDataBlock* pRes;
H
Haojun Liao 已提交
369 370
  int32_t      totalTables;
  int32_t      currentIndex;
371 372
} STagScanInfo;

H
Haojun Liao 已提交
373
typedef struct SOptrBasicInfo {
H
Haojun Liao 已提交
374
  SResultRowInfo    resultRowInfo;
H
Haojun Liao 已提交
375
  int32_t          *rowCellInfoOffset;  // offset value for each row result cell info
H
Haojun Liao 已提交
376
  SQLFunctionCtx   *pCtx;
H
Haojun Liao 已提交
377
  SSDataBlock      *pRes;
H
Haojun Liao 已提交
378 379
} SOptrBasicInfo;

H
Haojun Liao 已提交
380 381 382 383 384 385
typedef struct SOptrBasicInfo STableIntervalOperatorInfo;

typedef struct SAggOperatorInfo {
  SOptrBasicInfo binfo;
  uint32_t       seed;
} SAggOperatorInfo;
H
Haojun Liao 已提交
386

H
Haojun Liao 已提交
387
typedef struct SArithOperatorInfo {
H
Haojun Liao 已提交
388 389
  SOptrBasicInfo binfo;
  int32_t        bufCapacity;
H
Haojun Liao 已提交
390
  uint32_t       seed;
H
Haojun Liao 已提交
391 392
} SArithOperatorInfo;

H
Haojun Liao 已提交
393 394 395 396 397 398 399 400 401
typedef struct SLimitOperatorInfo {
  int64_t limit;
  int64_t total;
} SLimitOperatorInfo;

typedef struct SOffsetOperatorInfo {
  int64_t offset;
} SOffsetOperatorInfo;

H
Haojun Liao 已提交
402
typedef struct SFillOperatorInfo {
H
Haojun Liao 已提交
403
  SFillInfo   *pFillInfo;
H
Haojun Liao 已提交
404 405
  SSDataBlock *pRes;
  int64_t      totalInputRows;
H
Haojun Liao 已提交
406 407
} SFillOperatorInfo;

H
Haojun Liao 已提交
408 409 410
typedef struct SGroupbyOperatorInfo {
  SOptrBasicInfo binfo;
  int32_t        colIndex;
H
Haojun Liao 已提交
411
  char          *prevData;   // previous group by value
H
Haojun Liao 已提交
412
} SGroupbyOperatorInfo;
H
Haojun Liao 已提交
413

H
Haojun Liao 已提交
414 415 416 417
void freeParam(SQueryParam *param);
int32_t convertQueryMsg(SQueryTableMsg *pQueryMsg, SQueryParam* param);
int32_t createQueryFuncExprFromMsg(SQueryTableMsg *pQueryMsg, int32_t numOfOutput, SExprInfo **pExprInfo, SSqlFuncMsg **pExprMsg,
                                   SColumnInfo* pTagCols);
H
Haojun Liao 已提交
418 419 420
int32_t createIndirectQueryFuncExprFromMsg(SQueryTableMsg *pQueryMsg, int32_t numOfOutput, SExprInfo **pExprInfo,
                                           SSqlFuncMsg **pExprMsg, SExprInfo *prevExpr);

H
Haojun Liao 已提交
421 422 423 424 425 426 427 428 429
SSqlGroupbyExpr *createGroupbyExprFromMsg(SQueryTableMsg *pQueryMsg, SColIndex *pColIndex, int32_t *code);
SQInfo *createQInfoImpl(SQueryTableMsg *pQueryMsg, SSqlGroupbyExpr *pGroupbyExpr, SExprInfo *pExprs,
                        SExprInfo *pSecExprs, STableGroupInfo *pTableGroupInfo, SColumnInfo* pTagCols, bool stableQuery, char* sql);
int32_t initQInfo(SQueryTableMsg *pQueryMsg, void *tsdb, int32_t vgId, SQInfo *pQInfo, SQueryParam* param, bool isSTable);
void freeColumnFilterInfo(SColumnFilterInfo* pFilter, int32_t numOfFilters);

bool isQueryKilled(SQInfo *pQInfo);
int32_t checkForQueryBuf(size_t numOfTables);
bool doBuildResCheck(SQInfo* pQInfo);
H
Haojun Liao 已提交
430
void setQueryStatus(SQueryRuntimeEnv *pRuntimeEnv, int8_t status);
H
Haojun Liao 已提交
431 432 433 434 435 436 437 438 439 440 441 442 443

bool onlyQueryTags(SQuery* pQuery);
bool isValidQInfo(void *param);

int32_t doDumpQueryResult(SQInfo *pQInfo, char *data);

size_t getResultSize(SQInfo *pQInfo, int64_t *numOfRows);
void setQueryKilled(SQInfo *pQInfo);
void queryCostStatis(SQInfo *pQInfo);
void freeQInfo(SQInfo *pQInfo);

int32_t getMaximumIdleDurationSec();

H
Haojun Liao 已提交
444
#endif  // TDENGINE_QEXECUTOR_H