tsdb.h 27.1 KB
Newer Older
L
Liu Jicong 已提交
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
Hongze Cheng 已提交
15 16 17 18

#ifndef _TD_VNODE_TSDB_H_
#define _TD_VNODE_TSDB_H_

H
Hongze Cheng 已提交
19 20
#include "vnodeInt.h"

H
Hongze Cheng 已提交
21 22 23 24
#ifdef __cplusplus
extern "C" {
#endif

H
Hongze Cheng 已提交
25 26 27 28 29 30 31 32 33 34
// tsdbDebug ================
// clang-format off
#define tsdbFatal(...) do { if (tsdbDebugFlag & DEBUG_FATAL) { taosPrintLog("TSDB FATAL ", DEBUG_FATAL, 255, __VA_ARGS__); }}     while(0)
#define tsdbError(...) do { if (tsdbDebugFlag & DEBUG_ERROR) { taosPrintLog("TSDB ERROR ", DEBUG_ERROR, 255, __VA_ARGS__); }}     while(0)
#define tsdbWarn(...)  do { if (tsdbDebugFlag & DEBUG_WARN)  { taosPrintLog("TSDB WARN ", DEBUG_WARN, 255, __VA_ARGS__); }}       while(0)
#define tsdbInfo(...)  do { if (tsdbDebugFlag & DEBUG_INFO)  { taosPrintLog("TSDB ", DEBUG_INFO, 255, __VA_ARGS__); }}            while(0)
#define tsdbDebug(...) do { if (tsdbDebugFlag & DEBUG_DEBUG) { taosPrintLog("TSDB ", DEBUG_DEBUG, tsdbDebugFlag, __VA_ARGS__); }} while(0)
#define tsdbTrace(...) do { if (tsdbDebugFlag & DEBUG_TRACE) { taosPrintLog("TSDB ", DEBUG_TRACE, tsdbDebugFlag, __VA_ARGS__); }} while(0)
// clang-format on

H
Hongze Cheng 已提交
35 36 37 38 39 40 41 42
// tsdbMemTable ================
typedef struct STbData       STbData;
typedef struct STsdbMemTable STsdbMemTable;
typedef struct SMergeInfo    SMergeInfo;
typedef struct STable        STable;

int  tsdbMemTableCreate(STsdb *pTsdb, STsdbMemTable **ppMemTable);
void tsdbMemTableDestroy(STsdb *pTsdb, STsdbMemTable *pMemTable);
C
Cary Xu 已提交
43
int  tsdbInsertTableData(STsdb *pTsdb, SSubmitMsgIter *pMsgIter, SSubmitBlk *pBlock, int32_t *pAffectedRows);
H
Hongze Cheng 已提交
44 45 46 47 48
int  tsdbLoadDataFromCache(STable *pTable, SSkipListIterator *pIter, TSKEY maxKey, int maxRowsToRead, SDataCols *pCols,
                           TKEY *filterKeys, int nFilterKeys, bool keepDup, SMergeInfo *pMergeInfo);

// tsdbCommit ================

H
Hongze Cheng 已提交
49 50
// tsdbFS ================
typedef struct STsdbFS STsdbFS;
H
Hongze Cheng 已提交
51

H
Hongze Cheng 已提交
52
// tsdbSma ================
H
Hongze Cheng 已提交
53 54 55
typedef struct SSmaEnv  SSmaEnv;
typedef struct SSmaEnvs SSmaEnvs;

H
Hongze Cheng 已提交
56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
// structs
typedef struct {
  int   minFid;
  int   midFid;
  int   maxFid;
  TSKEY minKey;
} SRtn;

struct SSmaEnvs {
  int16_t  nTSma;
  int16_t  nRSma;
  SSmaEnv *pTSmaEnv;
  SSmaEnv *pRSmaEnv;
};

struct STsdb {
  char          *path;
  SVnode        *pVnode;
  bool           repoLocked;
C
Cary Xu 已提交
75
  int8_t         level;  // retention level
H
Hongze Cheng 已提交
76 77 78 79 80 81 82 83 84 85 86 87
  TdThreadMutex  mutex;
  STsdbMemTable *mem;
  STsdbMemTable *imem;
  SRtn           rtn;
  STsdbFS       *fs;
  SSmaEnvs       smaEnvs;
};

#if 1  // ======================================

typedef struct SSmaStat SSmaStat;

H
Hongze Cheng 已提交
88
struct STable {
H
Hongze Cheng 已提交
89 90 91
  uint64_t  tid;
  uint64_t  uid;
  STSchema *pSchema;
H
Hongze Cheng 已提交
92
};
H
Hongze Cheng 已提交
93 94 95 96 97 98 99 100 101 102

#define TABLE_TID(t) (t)->tid
#define TABLE_UID(t) (t)->uid

int     tsdbPrepareCommit(STsdb *pTsdb);
int32_t tsdbInitSma(STsdb *pTsdb);
int32_t tsdbDropTSma(STsdb *pTsdb, char *pMsg);
int32_t tsdbDropTSmaData(STsdb *pTsdb, int64_t indexUid);
int32_t tsdbInsertRSmaData(STsdb *pTsdb, char *msg);
void    tsdbCleanupReadHandle(tsdbReaderT queryHandle);
H
Hongze Cheng 已提交
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
typedef enum {
  TSDB_FILE_HEAD = 0,  // .head
  TSDB_FILE_DATA,      // .data
  TSDB_FILE_LAST,      // .last
  TSDB_FILE_SMAD,      // .smad(Block-wise SMA)
  TSDB_FILE_SMAL,      // .smal(Block-wise SMA)
  TSDB_FILE_MAX,       //
  TSDB_FILE_META,      // meta
  TSDB_FILE_TSMA,      // v2t100.${sma_index_name}, Time-range-wise SMA
  TSDB_FILE_RSMA,      // v2r100.${sma_index_name}, Time-range-wise Rollup SMA
} E_TSDB_FILE_T;

typedef struct {
  uint32_t magic;
  uint32_t fver;
  uint32_t len;
  uint32_t totalBlocks;
  uint32_t totalSubBlocks;
  uint32_t offset;
  uint64_t size;
  uint64_t tombSize;
} SDFInfo;

typedef struct {
  SDFInfo   info;
  STfsFile  f;
  TdFilePtr pFile;
  uint8_t   state;
} SDFile;

typedef struct {
  int      fid;
  int8_t   state;  // -128~127
  uint8_t  ver;    // 0~255, DFileSet version
  uint16_t reserve;
  SDFile   files[TSDB_FILE_MAX];
} SDFileSet;

H
Hongze Cheng 已提交
141
struct STbData {
H
Hongze Cheng 已提交
142 143 144
  tb_uid_t   uid;
  TSKEY      keyMin;
  TSKEY      keyMax;
H
Hongze Cheng 已提交
145 146
  int64_t    minVer;
  int64_t    maxVer;
H
Hongze Cheng 已提交
147 148
  int64_t    nrows;
  SSkipList *pData;
H
Hongze Cheng 已提交
149
};
H
Hongze Cheng 已提交
150

H
Hongze Cheng 已提交
151 152
struct STsdbMemTable {
  SVBufPool *pPool;
H
Hongze Cheng 已提交
153
  T_REF_DECLARE()
H
Hongze Cheng 已提交
154 155 156 157 158 159
  SRWLatch   latch;
  TSKEY      keyMin;
  TSKEY      keyMax;
  int64_t    minVer;
  int64_t    maxVer;
  int64_t    nRow;
H
Hongze Cheng 已提交
160
  SSkipList *pSlIdx;  // SSkiplist<STbData>
H
Hongze Cheng 已提交
161
  SHashObj  *pHashIdx;
H
Hongze Cheng 已提交
162
};
H
Hongze Cheng 已提交
163 164 165 166 167 168 169 170 171 172

typedef struct {
  uint32_t version;       // Commit version from 0 to increase
  int64_t  totalPoints;   // total points
  int64_t  totalStorage;  // Uncompressed total storage
} STsdbFSMeta;

// ==================
typedef struct {
  STsdbFSMeta meta;  // FS meta
H
Hongze Cheng 已提交
173 174
  SArray     *df;    // data file array
  SArray     *sf;    // sma data file array    v2f1900.index_name_1
H
Hongze Cheng 已提交
175 176
} SFSStatus;

H
Hongze Cheng 已提交
177
struct STsdbFS {
H
Hongze Cheng 已提交
178 179 180
  TdThreadRwlock lock;

  SFSStatus *cstatus;        // current status
H
Hongze Cheng 已提交
181 182
  SHashObj  *metaCache;      // meta cache
  SHashObj  *metaCacheComp;  // meta cache for compact
H
Hongze Cheng 已提交
183 184 185 186
  bool       intxn;
  SFSStatus *nstatus;  // new status
};

H
Hongze Cheng 已提交
187
#define REPO_ID(r)        TD_VID((r)->pVnode)
H
Hongze Cheng 已提交
188
#define REPO_CFG(r)       (&(r)->pVnode->config.tsdbCfg)
C
Cary Xu 已提交
189
#define REPO_LEVEL(r)     ((r)->level)
H
Hongze Cheng 已提交
190
#define REPO_FS(r)        ((r)->fs)
H
Hongze Cheng 已提交
191 192
#define REPO_META(r)      ((r)->pVnode->pMeta)
#define REPO_TFS(r)       ((r)->pVnode->pTfs)
H
Hongze Cheng 已提交
193 194 195 196 197 198 199 200 201 202 203 204 205 206
#define IS_REPO_LOCKED(r) ((r)->repoLocked)
#define REPO_TSMA_NUM(r)  ((r)->smaEnvs.nTSma)
#define REPO_RSMA_NUM(r)  ((r)->smaEnvs.nRSma)
#define REPO_TSMA_ENV(r)  ((r)->smaEnvs.pTSmaEnv)
#define REPO_RSMA_ENV(r)  ((r)->smaEnvs.pRSmaEnv)

int tsdbLockRepo(STsdb *pTsdb);
int tsdbUnlockRepo(STsdb *pTsdb);

static FORCE_INLINE STSchema *tsdbGetTableSchemaImpl(STable *pTable, bool lock, bool copy, int32_t version) {
  return pTable->pSchema;
}

// tsdbMemTable.h
H
Hongze Cheng 已提交
207
struct SMergeInfo {
H
Hongze Cheng 已提交
208 209 210 211 212 213 214
  int   rowsInserted;
  int   rowsUpdated;
  int   rowsDeleteSucceed;
  int   rowsDeleteFailed;
  int   nOperations;
  TSKEY keyFirst;
  TSKEY keyLast;
H
Hongze Cheng 已提交
215
};
H
Hongze Cheng 已提交
216

H
Hongze Cheng 已提交
217 218 219 220
static void  *taosTMalloc(size_t size);
static void  *taosTCalloc(size_t nmemb, size_t size);
static void  *taosTRealloc(void *ptr, size_t size);
static void  *taosTZfree(void *ptr);
H
Hongze Cheng 已提交
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
static size_t taosTSizeof(void *ptr);
static void   taosTMemset(void *ptr, int c);

static FORCE_INLINE STSRow *tsdbNextIterRow(SSkipListIterator *pIter) {
  if (pIter == NULL) return NULL;

  SSkipListNode *node = tSkipListIterGet(pIter);
  if (node == NULL) return NULL;

  return (STSRow *)SL_GET_NODE_DATA(node);
}

static FORCE_INLINE TSKEY tsdbNextIterKey(SSkipListIterator *pIter) {
  STSRow *row = tsdbNextIterRow(pIter);
  if (row == NULL) return TSDB_DATA_TIMESTAMP_NULL;

  return TD_ROW_KEY(row);
}

// tsdbReadImpl
typedef struct SReadH SReadH;

typedef struct {
  uint32_t len;
  uint32_t offset;
  uint32_t hasLast : 2;
  uint32_t numOfBlocks : 30;
  uint64_t uid;
  TSKEY    maxKey;
} SBlockIdx;

#ifdef TD_REFACTOR_3
typedef struct {
  int64_t last : 1;
  int64_t offset : 63;
  int32_t algorithm : 8;
  int32_t numOfRows : 24;
  int32_t len;
  int32_t keyLen;  // key column length, keyOffset = offset+sizeof(SBlockData)+sizeof(SBlockCol)*numOfCols
  int16_t numOfSubBlocks;
  int16_t numOfCols;  // not including timestamp column
  TSKEY   keyFirst;
  TSKEY   keyLast;
} SBlock;

#else

typedef enum {
  TSDB_SBLK_VER_0 = 0,
  TSDB_SBLK_VER_MAX,
} ESBlockVer;

#define SBlockVerLatest TSDB_SBLK_VER_0

typedef struct {
  uint8_t  last : 1;
  uint8_t  blkVer : 7;
  uint8_t  numOfSubBlocks;
  col_id_t numOfCols;    // not including timestamp column
  uint32_t len;          // data block length
  uint32_t keyLen : 20;  // key column length, keyOffset = offset+sizeof(SBlockData)+sizeof(SBlockCol)*numOfCols
  uint32_t algorithm : 4;
  uint32_t reserve : 8;
  col_id_t numOfBSma;
  uint16_t numOfRows;
  int64_t  offset;
  uint64_t aggrStat : 1;
  uint64_t aggrOffset : 63;
  TSKEY    keyFirst;
  TSKEY    keyLast;
} SBlockV0;

#define SBlock SBlockV0  // latest SBlock definition

C
Cary Xu 已提交
295 296 297
static FORCE_INLINE bool tsdbIsSupBlock(SBlock *pBlock) { return pBlock->numOfSubBlocks == 1; }
static FORCE_INLINE bool tsdbIsSubBlock(SBlock *pBlock) { return pBlock->numOfSubBlocks == 0; }

H
Hongze Cheng 已提交
298 299 300 301 302 303 304 305 306 307 308 309
#endif

typedef struct {
  int32_t  delimiter;  // For recovery usage
  int32_t  tid;
  uint64_t uid;
  SBlock   blocks[];
} SBlockInfo;

#ifdef TD_REFACTOR_3
typedef struct {
  int16_t  colId;
C
Cary Xu 已提交
310
  uint16_t bitmap : 1;  // 0: no bitmap if all rows are NORM, 1: has bitmap if has NULL/NORM rows
H
Hongze Cheng 已提交
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
  uint16_t reserve : 15;
  int32_t  len;
  uint32_t type : 8;
  uint32_t offset : 24;
  int64_t  sum;
  int64_t  max;
  int64_t  min;
  int16_t  maxIndex;
  int16_t  minIndex;
  int16_t  numOfNull;
  uint8_t  offsetH;
  char     padding[1];
} SBlockCol;
#else
typedef struct {
  int16_t  colId;
  uint16_t type : 6;
  uint16_t blen : 10;   // bitmap length(TODO: full UT for the bitmap compress of various data input)
C
Cary Xu 已提交
329
  uint32_t bitmap : 1;  // 0: no bitmap if all rows are NORM, 1: has bitmap if has NULL/NORM rows
H
Hongze Cheng 已提交
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
  uint32_t len : 31;    // data length + bitmap length
  uint32_t offset;
} SBlockColV0;

#define SBlockCol SBlockColV0  // latest SBlockCol definition

typedef struct {
  int16_t colId;
  int16_t maxIndex;
  int16_t minIndex;
  int16_t numOfNull;
  int64_t sum;
  int64_t max;
  int64_t min;
} SAggrBlkColV0;

#define SAggrBlkCol SAggrBlkColV0  // latest SAggrBlkCol definition

#endif

// Code here just for back-ward compatibility
static FORCE_INLINE void tsdbSetBlockColOffset(SBlockCol *pBlockCol, uint32_t offset) {
#ifdef TD_REFACTOR_3
  pBlockCol->offset = offset & ((((uint32_t)1) << 24) - 1);
  pBlockCol->offsetH = (uint8_t)(offset >> 24);
#else
  pBlockCol->offset = offset;
#endif
}

static FORCE_INLINE uint32_t tsdbGetBlockColOffset(SBlockCol *pBlockCol) {
#ifdef TD_REFACTOR_3
  uint32_t offset1 = pBlockCol->offset;
  uint32_t offset2 = pBlockCol->offsetH;
  return (offset1 | (offset2 << 24));
#else
  return pBlockCol->offset;
#endif
}

typedef struct {
  int32_t   delimiter;  // For recovery usage
  int32_t   numOfCols;  // For recovery usage
  uint64_t  uid;        // For recovery usage
  SBlockCol cols[];
} SBlockData;

typedef void SAggrBlkData;  // SBlockCol cols[];

struct SReadH {
  STsdb        *pRepo;
  SDFileSet     rSet;     // FSET to read
  SArray       *aBlkIdx;  // SBlockIdx array
  STable       *pTable;   // table to read
  SBlockIdx    *pBlkIdx;  // current reading table SBlockIdx
  int           cidx;
  SBlockInfo   *pBlkInfo;
  SBlockData   *pBlkData;      // Block info
  SAggrBlkData *pAggrBlkData;  // Aggregate Block info
  SDataCols    *pDCols[2];
  void         *pBuf;    // buffer
  void         *pCBuf;   // compression buffer
  void         *pExBuf;  // extra buffer
};

#define TSDB_READ_REPO(rh)      ((rh)->pRepo)
#define TSDB_READ_REPO_ID(rh)   REPO_ID(TSDB_READ_REPO(rh))
#define TSDB_READ_FSET(rh)      (&((rh)->rSet))
#define TSDB_READ_TABLE(rh)     ((rh)->pTable)
C
Cary Xu 已提交
399
#define TSDB_READ_TABLE_UID(rh) ((rh)->pTable->uid)
H
Hongze Cheng 已提交
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
#define TSDB_READ_HEAD_FILE(rh) TSDB_DFILE_IN_SET(TSDB_READ_FSET(rh), TSDB_FILE_HEAD)
#define TSDB_READ_DATA_FILE(rh) TSDB_DFILE_IN_SET(TSDB_READ_FSET(rh), TSDB_FILE_DATA)
#define TSDB_READ_LAST_FILE(rh) TSDB_DFILE_IN_SET(TSDB_READ_FSET(rh), TSDB_FILE_LAST)
#define TSDB_READ_SMAD_FILE(rh) TSDB_DFILE_IN_SET(TSDB_READ_FSET(rh), TSDB_FILE_SMAD)
#define TSDB_READ_SMAL_FILE(rh) TSDB_DFILE_IN_SET(TSDB_READ_FSET(rh), TSDB_FILE_SMAL)
#define TSDB_READ_BUF(rh)       ((rh)->pBuf)
#define TSDB_READ_COMP_BUF(rh)  ((rh)->pCBuf)
#define TSDB_READ_EXBUF(rh)     ((rh)->pExBuf)

#define TSDB_BLOCK_STATIS_SIZE(ncols, blkVer) \
  (sizeof(SBlockData) + sizeof(SBlockColV##blkVer) * (ncols) + sizeof(TSCKSUM))

static FORCE_INLINE size_t tsdbBlockStatisSize(int nCols, uint32_t blkVer) {
  switch (blkVer) {
    case TSDB_SBLK_VER_0:
    default:
      return TSDB_BLOCK_STATIS_SIZE(nCols, 0);
  }
}

#define TSDB_BLOCK_AGGR_SIZE(ncols, blkVer) (sizeof(SAggrBlkColV##blkVer) * (ncols) + sizeof(TSCKSUM))

static FORCE_INLINE size_t tsdbBlockAggrSize(int nCols, uint32_t blkVer) {
  switch (blkVer) {
    case TSDB_SBLK_VER_0:
    default:
      return TSDB_BLOCK_AGGR_SIZE(nCols, 0);
  }
}

H
Hongze Cheng 已提交
430 431 432 433 434 435 436 437 438 439 440 441
int  tsdbInitReadH(SReadH *pReadh, STsdb *pRepo);
void tsdbDestroyReadH(SReadH *pReadh);
int  tsdbSetAndOpenReadFSet(SReadH *pReadh, SDFileSet *pSet);
void tsdbCloseAndUnsetFSet(SReadH *pReadh);
int  tsdbLoadBlockIdx(SReadH *pReadh);
int  tsdbSetReadTable(SReadH *pReadh, STable *pTable);
int  tsdbLoadBlockInfo(SReadH *pReadh, void *pTarget);
int  tsdbLoadBlockData(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlockInfo);
int tsdbLoadBlockDataCols(SReadH *pReadh, SBlock *pBlock, SBlockInfo *pBlkInfo, const int16_t *colIds, int numOfColsIds,
                          bool mergeBitmap);
int tsdbLoadBlockStatis(SReadH *pReadh, SBlock *pBlock);
int tsdbEncodeSBlockIdx(void **buf, SBlockIdx *pIdx);
H
Hongze Cheng 已提交
442
void *tsdbDecodeSBlockIdx(void *buf, SBlockIdx *pIdx);
443
void  tsdbGetBlockStatis(SReadH *pReadh, SColumnDataAgg *pStatis, int numOfCols, SBlock *pBlock);
H
Hongze Cheng 已提交
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479

static FORCE_INLINE int tsdbMakeRoom(void **ppBuf, size_t size) {
  void  *pBuf = *ppBuf;
  size_t tsize = taosTSizeof(pBuf);

  if (tsize < size) {
    if (tsize == 0) tsize = 1024;

    while (tsize < size) {
      tsize *= 2;
    }

    *ppBuf = taosTRealloc(pBuf, tsize);
    if (*ppBuf == NULL) {
      terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
      return -1;
    }
  }

  return 0;
}

// tsdbMemory
static FORCE_INLINE void *taosTMalloc(size_t size) {
  if (size <= 0) return NULL;

  void *ret = taosMemoryMalloc(size + sizeof(size_t));
  if (ret == NULL) return NULL;

  *(size_t *)ret = size;

  return (void *)((char *)ret + sizeof(size_t));
}

static FORCE_INLINE void *taosTCalloc(size_t nmemb, size_t size) {
  size_t tsize = nmemb * size;
H
Hongze Cheng 已提交
480
  void  *ret = taosTMalloc(tsize);
H
Hongze Cheng 已提交
481 482 483 484 485 486 487 488 489 490
  if (ret == NULL) return NULL;

  taosTMemset(ret, 0);
  return ret;
}

static FORCE_INLINE size_t taosTSizeof(void *ptr) { return (ptr) ? (*(size_t *)((char *)ptr - sizeof(size_t))) : 0; }

static FORCE_INLINE void taosTMemset(void *ptr, int c) { memset(ptr, c, taosTSizeof(ptr)); }

H
Hongze Cheng 已提交
491
static FORCE_INLINE void *taosTRealloc(void *ptr, size_t size) {
H
Hongze Cheng 已提交
492 493 494 495
  if (ptr == NULL) return taosTMalloc(size);

  if (size <= taosTSizeof(ptr)) return ptr;

H
Hongze Cheng 已提交
496
  void  *tptr = (void *)((char *)ptr - sizeof(size_t));
H
Hongze Cheng 已提交
497
  size_t tsize = size + sizeof(size_t);
H
Hongze Cheng 已提交
498
  void  *tptr1 = taosMemoryRealloc(tptr, tsize);
H
Hongze Cheng 已提交
499 500 501 502 503 504 505 506
  if (tptr1 == NULL) return NULL;
  tptr = tptr1;

  *(size_t *)tptr = size;

  return (void *)((char *)tptr + sizeof(size_t));
}

H
Hongze Cheng 已提交
507
static FORCE_INLINE void *taosTZfree(void *ptr) {
H
Hongze Cheng 已提交
508
  if (ptr) {
H
Hongze Cheng 已提交
509
    taosMemoryFree((void *)((char *)ptr - sizeof(size_t)));
H
Hongze Cheng 已提交
510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
  }
  return NULL;
}

// tsdbCommit

void tsdbGetRtnSnap(STsdb *pRepo, SRtn *pRtn);

static FORCE_INLINE int TSDB_KEY_FID(TSKEY key, int32_t days, int8_t precision) {
  if (key < 0) {
    return (int)((key + 1) / tsTickPerDay[precision] / days - 1);
  } else {
    return (int)((key / tsTickPerDay[precision] / days));
  }
}

static FORCE_INLINE int tsdbGetFidLevel(int fid, SRtn *pRtn) {
  if (fid >= pRtn->maxFid) {
    return 0;
  } else if (fid >= pRtn->midFid) {
    return 1;
  } else if (fid >= pRtn->minFid) {
    return 2;
  } else {
    return -1;
  }
}

// tsdbFile
H
Hongze Cheng 已提交
539 540
#define TSDB_FILE_HEAD_SIZE  512
#define TSDB_FILE_DELIMITER  0xF00AFA0F
H
Hongze Cheng 已提交
541
#define TSDB_FILE_INIT_MAGIC 0xFFFFFFFF
H
Hongze Cheng 已提交
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559
#define TSDB_IVLD_FID        INT_MIN
#define TSDB_FILE_STATE_OK   0
#define TSDB_FILE_STATE_BAD  1

#define TSDB_FILE_INFO(tf)         (&((tf)->info))
#define TSDB_FILE_F(tf)            (&((tf)->f))
#define TSDB_FILE_PFILE(tf)        ((tf)->pFile)
#define TSDB_FILE_FULL_NAME(tf)    (TSDB_FILE_F(tf)->aname)
#define TSDB_FILE_OPENED(tf)       (TSDB_FILE_PFILE(tf) != NULL)
#define TSDB_FILE_CLOSED(tf)       (!TSDB_FILE_OPENED(tf))
#define TSDB_FILE_SET_CLOSED(f)    (TSDB_FILE_PFILE(f) = NULL)
#define TSDB_FILE_LEVEL(tf)        (TSDB_FILE_F(tf)->did.level)
#define TSDB_FILE_ID(tf)           (TSDB_FILE_F(tf)->did.id)
#define TSDB_FILE_DID(tf)          (TSDB_FILE_F(tf)->did)
#define TSDB_FILE_REL_NAME(tf)     (TSDB_FILE_F(tf)->rname)
#define TSDB_FILE_ABS_NAME(tf)     (TSDB_FILE_F(tf)->aname)
#define TSDB_FILE_FSYNC(tf)        taosFsyncFile(TSDB_FILE_PFILE(tf))
#define TSDB_FILE_STATE(tf)        ((tf)->state)
H
Hongze Cheng 已提交
560
#define TSDB_FILE_SET_STATE(tf, s) ((tf)->state = (s))
H
Hongze Cheng 已提交
561 562
#define TSDB_FILE_IS_OK(tf)        (TSDB_FILE_STATE(tf) == TSDB_FILE_STATE_OK)
#define TSDB_FILE_IS_BAD(tf)       (TSDB_FILE_STATE(tf) == TSDB_FILE_STATE_BAD)
H
Hongze Cheng 已提交
563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584

typedef int32_t TSDB_FILE_T;
typedef enum {
  TSDB_FS_VER_0 = 0,
  TSDB_FS_VER_MAX,
} ETsdbFsVer;

#define TSDB_LATEST_FVER    TSDB_FS_VER_0  // latest version for DFile
#define TSDB_LATEST_SFS_VER TSDB_FS_VER_0  // latest version for 'current' file

static FORCE_INLINE uint32_t tsdbGetDFSVersion(TSDB_FILE_T fType) {  // latest version for DFile
  switch (fType) {
    case TSDB_FILE_HEAD:  // .head
    case TSDB_FILE_DATA:  // .data
    case TSDB_FILE_LAST:  // .last
    case TSDB_FILE_SMAD:  // .smad(Block-wise SMA)
    case TSDB_FILE_SMAL:  // .smal(Block-wise SMA)
    default:
      return TSDB_LATEST_FVER;
  }
}

H
Hongze Cheng 已提交
585 586 587 588 589 590 591 592
void  tsdbInitDFile(STsdb *pRepo, SDFile *pDFile, SDiskID did, int fid, uint32_t ver, TSDB_FILE_T ftype);
void  tsdbInitDFileEx(SDFile *pDFile, SDFile *pODFile);
int   tsdbEncodeSDFile(void **buf, SDFile *pDFile);
void *tsdbDecodeSDFile(STsdb *pRepo, void *buf, SDFile *pDFile);
int   tsdbCreateDFile(STsdb *pRepo, SDFile *pDFile, bool updateHeader, TSDB_FILE_T fType);
int   tsdbUpdateDFileHeader(SDFile *pDFile);
int   tsdbLoadDFileHeader(SDFile *pDFile, SDFInfo *pInfo);
int   tsdbParseDFilename(const char *fname, int *vid, int *fid, TSDB_FILE_T *ftype, uint32_t *version);
H
Hongze Cheng 已提交
593

H
Hongze Cheng 已提交
594
static FORCE_INLINE void tsdbSetDFileInfo(SDFile *pDFile, SDFInfo *pInfo) { pDFile->info = *pInfo; }
H
Hongze Cheng 已提交
595

H
Hongze Cheng 已提交
596
static FORCE_INLINE int tsdbOpenDFile(SDFile *pDFile, int flags) {
H
Hongze Cheng 已提交
597 598 599 600 601 602 603 604 605 606 607
  ASSERT(!TSDB_FILE_OPENED(pDFile));

  pDFile->pFile = taosOpenFile(TSDB_FILE_FULL_NAME(pDFile), flags);
  if (pDFile->pFile == NULL) {
    terrno = TAOS_SYSTEM_ERROR(errno);
    return -1;
  }

  return 0;
}

H
Hongze Cheng 已提交
608
static FORCE_INLINE void tsdbCloseDFile(SDFile *pDFile) {
H
Hongze Cheng 已提交
609 610 611 612 613 614
  if (TSDB_FILE_OPENED(pDFile)) {
    taosCloseFile(&pDFile->pFile);
    TSDB_FILE_SET_CLOSED(pDFile);
  }
}

H
Hongze Cheng 已提交
615
static FORCE_INLINE int64_t tsdbSeekDFile(SDFile *pDFile, int64_t offset, int whence) {
H
Hongze Cheng 已提交
616 617 618 619 620 621 622 623 624 625 626
  // ASSERT(TSDB_FILE_OPENED(pDFile));

  int64_t loffset = taosLSeekFile(TSDB_FILE_PFILE(pDFile), offset, whence);
  if (loffset < 0) {
    terrno = TAOS_SYSTEM_ERROR(errno);
    return -1;
  }

  return loffset;
}

H
Hongze Cheng 已提交
627
static FORCE_INLINE int64_t tsdbWriteDFile(SDFile *pDFile, void *buf, int64_t nbyte) {
H
Hongze Cheng 已提交
628 629 630 631 632 633 634 635 636 637 638
  ASSERT(TSDB_FILE_OPENED(pDFile));

  int64_t nwrite = taosWriteFile(pDFile->pFile, buf, nbyte);
  if (nwrite < nbyte) {
    terrno = TAOS_SYSTEM_ERROR(errno);
    return -1;
  }

  return nwrite;
}

H
Hongze Cheng 已提交
639 640
static FORCE_INLINE void tsdbUpdateDFileMagic(SDFile *pDFile, void *pCksm) {
  pDFile->info.magic = taosCalcChecksum(pDFile->info.magic, (uint8_t *)(pCksm), sizeof(TSCKSUM));
H
Hongze Cheng 已提交
641 642
}

H
Hongze Cheng 已提交
643
static FORCE_INLINE int tsdbAppendDFile(SDFile *pDFile, void *buf, int64_t nbyte, int64_t *offset) {
H
Hongze Cheng 已提交
644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
  ASSERT(TSDB_FILE_OPENED(pDFile));

  int64_t toffset;

  if ((toffset = tsdbSeekDFile(pDFile, 0, SEEK_END)) < 0) {
    return -1;
  }

  ASSERT(pDFile->info.size == toffset);

  if (offset) {
    *offset = toffset;
  }

  if (tsdbWriteDFile(pDFile, buf, nbyte) < 0) {
    return -1;
  }

  pDFile->info.size += nbyte;

  return (int)nbyte;
}

H
Hongze Cheng 已提交
667
static FORCE_INLINE int tsdbRemoveDFile(SDFile *pDFile) { return tfsRemoveFile(TSDB_FILE_F(pDFile)); }
H
Hongze Cheng 已提交
668

H
Hongze Cheng 已提交
669
static FORCE_INLINE int64_t tsdbReadDFile(SDFile *pDFile, void *buf, int64_t nbyte) {
H
Hongze Cheng 已提交
670 671 672 673 674 675 676 677 678 679 680
  ASSERT(TSDB_FILE_OPENED(pDFile));

  int64_t nread = taosReadFile(pDFile->pFile, buf, nbyte);
  if (nread < 0) {
    terrno = TAOS_SYSTEM_ERROR(errno);
    return -1;
  }

  return nread;
}

H
Hongze Cheng 已提交
681
static FORCE_INLINE int tsdbCopyDFile(SDFile *pSrc, SDFile *pDest) {
H
Hongze Cheng 已提交
682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707
  if (tfsCopyFile(TSDB_FILE_F(pSrc), TSDB_FILE_F(pDest)) < 0) {
    terrno = TAOS_SYSTEM_ERROR(errno);
    return -1;
  }

  tsdbSetDFileInfo(pDest, TSDB_FILE_INFO(pSrc));
  return 0;
}

// =============== SDFileSet

typedef struct {
  int      fid;
  int8_t   state;
  uint8_t  ver;
  uint16_t reserve;
#if 0
  SDFInfo   info;
#endif
  STfsFile  f;
  TdFilePtr pFile;

} SSFile;  // files split by days with fid

#define TSDB_LATEST_FSET_VER 0

H
Hongze Cheng 已提交
708 709 710
#define TSDB_FSET_FID(s)        ((s)->fid)
#define TSDB_FSET_STATE(s)      ((s)->state)
#define TSDB_FSET_VER(s)        ((s)->ver)
H
Hongze Cheng 已提交
711
#define TSDB_DFILE_IN_SET(s, t) ((s)->files + (t))
H
Hongze Cheng 已提交
712 713
#define TSDB_FSET_LEVEL(s)      TSDB_FILE_LEVEL(TSDB_DFILE_IN_SET(s, 0))
#define TSDB_FSET_ID(s)         TSDB_FILE_ID(TSDB_DFILE_IN_SET(s, 0))
H
Hongze Cheng 已提交
714 715 716 717 718 719 720 721 722 723 724 725 726
#define TSDB_FSET_SET_CLOSED(s)                                                \
  do {                                                                         \
    for (TSDB_FILE_T ftype = TSDB_FILE_HEAD; ftype < TSDB_FILE_MAX; ftype++) { \
      TSDB_FILE_SET_CLOSED(TSDB_DFILE_IN_SET(s, ftype));                       \
    }                                                                          \
  } while (0);
#define TSDB_FSET_FSYNC(s)                                                     \
  do {                                                                         \
    for (TSDB_FILE_T ftype = TSDB_FILE_HEAD; ftype < TSDB_FILE_MAX; ftype++) { \
      TSDB_FILE_FSYNC(TSDB_DFILE_IN_SET(s, ftype));                            \
    }                                                                          \
  } while (0);

H
Hongze Cheng 已提交
727 728 729 730 731 732 733 734 735 736 737 738
void  tsdbInitDFileSet(STsdb *pRepo, SDFileSet *pSet, SDiskID did, int fid, uint32_t ver);
void  tsdbInitDFileSetEx(SDFileSet *pSet, SDFileSet *pOSet);
int   tsdbEncodeDFileSet(void **buf, SDFileSet *pSet);
void *tsdbDecodeDFileSet(STsdb *pRepo, void *buf, SDFileSet *pSet);
int   tsdbEncodeDFileSetEx(void **buf, SDFileSet *pSet);
void *tsdbDecodeDFileSetEx(void *buf, SDFileSet *pSet);
int   tsdbApplyDFileSetChange(SDFileSet *from, SDFileSet *to);
int   tsdbCreateDFileSet(STsdb *pRepo, SDFileSet *pSet, bool updateHeader);
int   tsdbUpdateDFileSetHeader(SDFileSet *pSet);
int   tsdbScanAndTryFixDFileSet(STsdb *pRepo, SDFileSet *pSet);

static FORCE_INLINE void tsdbCloseDFileSet(SDFileSet *pSet) {
H
Hongze Cheng 已提交
739 740 741 742 743
  for (TSDB_FILE_T ftype = 0; ftype < TSDB_FILE_MAX; ftype++) {
    tsdbCloseDFile(TSDB_DFILE_IN_SET(pSet, ftype));
  }
}

H
Hongze Cheng 已提交
744
static FORCE_INLINE int tsdbOpenDFileSet(SDFileSet *pSet, int flags) {
H
Hongze Cheng 已提交
745 746 747 748 749 750 751 752 753
  for (TSDB_FILE_T ftype = 0; ftype < TSDB_FILE_MAX; ftype++) {
    if (tsdbOpenDFile(TSDB_DFILE_IN_SET(pSet, ftype), flags) < 0) {
      tsdbCloseDFileSet(pSet);
      return -1;
    }
  }
  return 0;
}

H
Hongze Cheng 已提交
754
static FORCE_INLINE void tsdbRemoveDFileSet(SDFileSet *pSet) {
H
Hongze Cheng 已提交
755 756 757 758 759
  for (TSDB_FILE_T ftype = 0; ftype < TSDB_FILE_MAX; ftype++) {
    (void)tsdbRemoveDFile(TSDB_DFILE_IN_SET(pSet, ftype));
  }
}

H
Hongze Cheng 已提交
760
static FORCE_INLINE int tsdbCopyDFileSet(SDFileSet *pSrc, SDFileSet *pDest) {
H
Hongze Cheng 已提交
761 762 763 764 765 766 767 768 769 770
  for (TSDB_FILE_T ftype = 0; ftype < TSDB_FILE_MAX; ftype++) {
    if (tsdbCopyDFile(TSDB_DFILE_IN_SET(pSrc, ftype), TSDB_DFILE_IN_SET(pDest, ftype)) < 0) {
      tsdbRemoveDFileSet(pDest);
      return -1;
    }
  }

  return 0;
}

H
Hongze Cheng 已提交
771
static FORCE_INLINE void tsdbGetFidKeyRange(int days, int8_t precision, int fid, TSKEY *minKey, TSKEY *maxKey) {
H
Hongze Cheng 已提交
772 773 774 775
  *minKey = fid * days * tsTickPerDay[precision];
  *maxKey = *minKey + days * tsTickPerDay[precision] - 1;
}

H
Hongze Cheng 已提交
776
static FORCE_INLINE bool tsdbFSetIsOk(SDFileSet *pSet) {
H
Hongze Cheng 已提交
777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
  for (TSDB_FILE_T ftype = 0; ftype < TSDB_FILE_MAX; ftype++) {
    if (TSDB_FILE_IS_BAD(TSDB_DFILE_IN_SET(pSet, ftype))) {
      return false;
    }
  }

  return true;
}

// tsdbFS
// ================== TSDB global config
extern bool tsdbForceKeepFile;

// ================== CURRENT file header info
typedef struct {
  uint32_t version;  // Current file system version (relating to code)
  uint32_t len;      // Encode content length (including checksum)
} SFSHeader;

// ================== TSDB File System Meta

/**
 * @brief Directory structure of .tsma data files.
 *
 *  /vnode2/tsdb $ tree tsma/
 *   tsma/
 *   ├── v2f100.index_name_1
 *   ├── v2f101.index_name_1
 *   ├── v2f102.index_name_1
 *   ├── v2f1900.index_name_3
 *   ├── v2f1901.index_name_3
 *   ├── v2f1902.index_name_3
 *   ├── v2f200.index_name_2
 *   ├── v2f201.index_name_2
 *   └── v2f202.index_name_2
 *
 *   0 directories, 9 files
 */

#define FS_CURRENT_STATUS(pfs) ((pfs)->cstatus)
H
Hongze Cheng 已提交
817 818 819 820
#define FS_NEW_STATUS(pfs)     ((pfs)->nstatus)
#define FS_IN_TXN(pfs)         (pfs)->intxn
#define FS_VERSION(pfs)        ((pfs)->cstatus->meta.version)
#define FS_TXN_VERSION(pfs)    ((pfs)->nstatus->meta.version)
H
Hongze Cheng 已提交
821 822 823 824

typedef struct {
  int        direction;
  uint64_t   version;  // current FS version
H
Hongze Cheng 已提交
825
  STsdbFS   *pfs;
H
Hongze Cheng 已提交
826 827 828 829 830
  int        index;  // used to position next fset when version the same
  int        fid;    // used to seek when version is changed
  SDFileSet *pSet;
} SFSIter;

H
Hongze Cheng 已提交
831
#define TSDB_FS_ITER_FORWARD  TSDB_ORDER_ASC
H
Hongze Cheng 已提交
832 833 834
#define TSDB_FS_ITER_BACKWARD TSDB_ORDER_DESC

STsdbFS *tsdbNewFS(const STsdbCfg *pCfg);
H
Hongze Cheng 已提交
835
void    *tsdbFreeFS(STsdbFS *pfs);
H
Hongze Cheng 已提交
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876
int      tsdbOpenFS(STsdb *pRepo);
void     tsdbCloseFS(STsdb *pRepo);
void     tsdbStartFSTxn(STsdb *pRepo, int64_t pointsAdd, int64_t storageAdd);
int      tsdbEndFSTxn(STsdb *pRepo);
int      tsdbEndFSTxnWithError(STsdbFS *pfs);
void     tsdbUpdateFSTxnMeta(STsdbFS *pfs, STsdbFSMeta *pMeta);
// void     tsdbUpdateMFile(STsdbFS *pfs, const SMFile *pMFile);
int tsdbUpdateDFileSet(STsdbFS *pfs, const SDFileSet *pSet);

void       tsdbFSIterInit(SFSIter *pIter, STsdbFS *pfs, int direction);
void       tsdbFSIterSeek(SFSIter *pIter, int fid);
SDFileSet *tsdbFSIterNext(SFSIter *pIter);
int        tsdbLoadMetaCache(STsdb *pRepo, bool recoverMeta);

static FORCE_INLINE int tsdbRLockFS(STsdbFS *pFs) {
  int code = taosThreadRwlockRdlock(&(pFs->lock));
  if (code != 0) {
    terrno = TAOS_SYSTEM_ERROR(code);
    return -1;
  }
  return 0;
}

static FORCE_INLINE int tsdbWLockFS(STsdbFS *pFs) {
  int code = taosThreadRwlockWrlock(&(pFs->lock));
  if (code != 0) {
    terrno = TAOS_SYSTEM_ERROR(code);
    return -1;
  }
  return 0;
}

static FORCE_INLINE int tsdbUnLockFS(STsdbFS *pFs) {
  int code = taosThreadRwlockUnlock(&(pFs->lock));
  if (code != 0) {
    terrno = TAOS_SYSTEM_ERROR(code);
    return -1;
  }
  return 0;
}

H
refact  
Hongze Cheng 已提交
877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913
typedef struct SSmaKey SSmaKey;

struct SSmaKey {
  TSKEY   skey;
  int64_t groupId;
};

typedef struct SDBFile SDBFile;

struct SDBFile {
  int32_t fid;
  TDB    *pDB;
  char   *path;
};

int32_t tsdbOpenDBEnv(TENV **ppEnv, const char *path);
int32_t tsdbCloseDBEnv(TENV *pEnv);
int32_t tsdbOpenDBF(TENV *pEnv, SDBFile *pDBF);
int32_t tsdbCloseDBF(SDBFile *pDBF);
int32_t tsdbSaveSmaToDB(SDBFile *pDBF, void *pKey, int32_t keyLen, void *pVal, int32_t valLen, TXN *txn);
void   *tsdbGetSmaDataByKey(SDBFile *pDBF, const void *pKey, int32_t keyLen, int32_t *valLen);

void  tsdbDestroySmaEnv(SSmaEnv *pSmaEnv);
void *tsdbFreeSmaEnv(SSmaEnv *pSmaEnv);
#if 0
int32_t tsdbGetTSmaStatus(STsdb *pTsdb, STSma *param, void *result);
int32_t tsdbRemoveTSmaData(STsdb *pTsdb, STSma *param, STimeWindow *pWin);
#endif

// internal func
static FORCE_INLINE int32_t tsdbEncodeTSmaKey(int64_t groupId, TSKEY tsKey, void **pData) {
  int32_t len = 0;
  len += taosEncodeFixedI64(pData, tsKey);
  len += taosEncodeFixedI64(pData, groupId);
  return len;
}

H
Hongze Cheng 已提交
914 915
#endif

H
Hongze Cheng 已提交
916 917 918 919 920
#ifdef __cplusplus
}
#endif

#endif /*_TD_VNODE_TSDB_H_*/