more code

CMakeLists.txt

@@ -16,7 +16,6 @@ SET(TD_GRANT FALSE)
 SET(TD_SYNC TRUE)
 SET(TD_MQTT TRUE)
 SET(TD_TSDB_PLUGINS FALSE)
-SET(TD_DNODE_PLUGINS FALSE)
 
 SET(TD_COVER FALSE)
 SET(TD_MEM_CHECK FALSE)

cmake/define.inc

@@ -25,10 +25,6 @@ IF (TD_TSDB_PLUGINS)
   ADD_DEFINITIONS(-D_TSDB_PLUGINS)
 ENDIF ()
 
-IF (TD_DNODE_PLUGINS)
-  ADD_DEFINITIONS(-D_DNODE_PLUGINS)
-ENDIF ()
-
 IF (TD_GODLL)
   ADD_DEFINITIONS(-D_TD_GO_DLL_)
 ENDIF ()

src/common/inc/tpath.h

@@ -22,28 +22,36 @@
 extern "C" {
 #endif
 
-static FORCE_INLINE void tdGetMnodeRootDir(char *rootDir, char *dirName) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "%s/mnode", rootDir);
+static FORCE_INLINE void tdGetMnodeRootDir(char *baseDir, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/mnode", baseDir);
 }
 
-static FORCE_INLINE void tdGetDnodeRootDir(char *rootDir, char *dirName) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "%s/dnode", rootDir);
+static FORCE_INLINE void tdGetDnodeRootDir(char *baseDir, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/dnode", baseDir);
 }
 
-static FORCE_INLINE void tdGetVnodeRootDir(char *rootDir, char *dirName) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode", rootDir);
+static FORCE_INLINE void tdGetVnodeRootDir(char *baseDir, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode", baseDir);
 }
 
-static FORCE_INLINE void tdGetVnodeBackRootDir(char *rootDir, char *dirName) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode_bak", rootDir);
+static FORCE_INLINE void tdGetVnodeBackRootDir(char *baseDir, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode_bak", baseDir);
 }
 
-static FORCE_INLINE void tdGetVnodeDir(char *rootDir, int vid, char *dirName) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode/vnode%d", rootDir, vid);
+static FORCE_INLINE void tdGetVnodeDir(char *baseDir, int vid, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode/vnode%d", baseDir, vid);
 }
 
-static FORCE_INLINE void tdGetVnodeBackDir(char *rootDir, int vid, char *dirName) {
-  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode_bak/vnode%d", rootDir, vid);
+static FORCE_INLINE void tdGetVnodeBackDir(char *baseDir, int vid, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode_bak/vnode%d", baseDir, vid);
+}
+
+static FORCE_INLINE void tdGetTsdbRootDir(char *baseDir, int vid, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode/vnode%d/tsdb", baseDir, vid);
+}
+
+static FORCE_INLINE void tdGetTsdbDataDir(char *baseDir, int vid, char *dirName) {
+  snprintf(dirName, TSDB_FILENAME_LEN, "%s/vnode/vnode%d/tsdb/data", baseDir, vid);
 }
 
 #ifdef __cplusplus

src/dnode/CMakeLists.txt

@@ -36,10 +36,6 @@ IF (TD_LINUX)
     TARGET_LINK_LIBRARIES(taosd balance sync)
   ENDIF ()
 
-  IF (TD_DNODE_PLUGINS)
-    TARGET_LINK_LIBRARIES(taosd dnodePlugins)
-  ENDIF()
-
   SET(PREPARE_ENV_CMD "prepare_env_cmd")
   SET(PREPARE_ENV_TARGET "prepare_env_target")
   ADD_CUSTOM_COMMAND(OUTPUT ${PREPARE_ENV_CMD}

src/dnode/src/dnodeMain.c

@@ -34,8 +34,6 @@
 #include "dnodeTelemetry.h"
 #include "tpath.h"
 
-struct SDnodeTier *pDnodeTier = NULL;
-
 static int32_t dnodeInitStorage();
 static void dnodeCleanupStorage();
 static void dnodeSetRunStatus(SDnodeRunStatus status);

src/dnode/src/dnodeTier.c

@@ -12,18 +12,293 @@
  * 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/>.
  */
-
 #include "os.h"
+
 #include "dnode.h"
+#include "dnodeInt.h"
+#include "taosdef.h"
+
+#define DISK_MIN_FREE_SPACE 30 * 1024 * 1024  // disk free space less than 100M will not create new file again
+#define DNODE_DISK_AVAIL(pDisk) ((pDisk)->dmeta.free > DISK_MIN_FREE_SPACE)
+
+static int dnodeFormatDir(char *idir, char *odir);
+static int dnodeCheckDisk(char *dirName, int level, int primary);
+static int dnodeUpdateDiskMeta(SDisk *pDisk);
+static int dnodeAddDisk(SDnodeTier *pDnodeTier, char *dir, int level, int primary);
+
+SDnodeTier *dnodeNewTier() {
+  SDnodeTier *pDnodeTier = (SDnodeTier *)calloc(1, sizeof(*pDnodeTier));
+  if (pDnodeTier == NULL) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return NULL;
+  }
+
+  int ret = pthread_rwlock_init(&(pDnodeTier->rwlock), NULL);
+  if (ret != 0) {
+    terrno = TAOS_SYSTEM_ERROR(ret);
+    dnodeCloseTier(pDnodeTier);
+    return NULL;
+  }
+
+  pDnodeTier->map = taosHashInit(DNODE_MAX_TIERS * DNODE_MAX_DISKS_PER_TIER * 2,
+                                 taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false, HASH_NO_LOCK);
+  if (pDnodeTier->map == NULL) {
+    terrno = TSDB_CODE_COM_OUT_OF_MEMORY;
+    dnodeCloseTier(pDnodeTier);
+    return NULL;
+  }
+
+  return pDnodeTier;
+}
+
+void *dnodeCloseTier(SDnodeTier *pDnodeTier) {
+  if (pDnodeTier) {
+    if (pDnodeTier->map) {
+      taosHashCleanup(pDnodeTier->map);
+      pDnodeTier->map = NULL;
+    }
+
+    pthread_rwlock_destroy(&(pDnodeTier->rwlock));
+
+    for (int i = 0; i < pDnodeTier->nTiers; i++) {
+      STier *pTier = pDnodeTier->tiers + i;
+      for (int j = 0; j < pTier->nDisks; j++) {
+        if (pTier->disks[j]) {
+          free(pTier->disks[j]);
+          pTier->disks[j] = NULL;
+        }
+      }
+    }
+    free(pDnodeTier);
+  }
+  return NULL;
+}
+
+int dnodeAddDisks(SDnodeTier *pDnodeTier, SDiskCfg *pDiskCfgs, int ndisks) {
+  ASSERT(ndisks > 0);
+
+  for (int i = 0; i < ndisks; i++) {
+    SDiskCfg *pCfg = pDiskCfgs + i;
+    dnodeAddDisk(pDnodeTier, pCfg->dir, pCfg->level, pCfg->primary);
+  }
+
+  if (dnodeCheckTiers(pDnodeTier) < 0) return -1;
+  
+  return 0;
+}
+
+int dnodeUpdateTiersInfo(SDnodeTier *pDnodeTier) {
+  for (int i = 0; i < pDnodeTier->nTiers; i++) {
+    STier *pTier = pDnodeTier->tiers + i;
+
+    for (int j = 0; j < pTier->nDisks; j++) {
+      SDisk *pDisk = pTier->disks[j];
+      if (dnodeUpdateDiskMeta(pDisk) < 0) return -1;
+    }
+  }
+  return 0;
+}
+
+int dnodeCheckTiers(SDnodeTier *pDnodeTier) {
+  ASSERT(pDnodeTier->nTiers > 0);
+  if (DNODE_PRIMARY_DISK(pDnodeTier) == NULL) {
+    terrno = TSDB_CODE_DND_LACK_PRIMARY_DISK;
+    return -1;
+  }
+
+  for (int i = 0; i < pDnodeTier->nTiers; i++) {
+    if (pDnodeTier->tiers[i].nDisks == 0) {
+      terrno = TSDB_CODE_DND_NO_DISK_AT_TIER;
+      return -1;
+    }
+  }
+
+  return 0;
+}
+
+SDisk *dnodeAssignDisk(SDnodeTier *pDnodeTier, int level) {
+  ASSERT(level < pDnodeTier->nTiers);
+
+  STier *pTier = pDnodeTier->tiers + level;
+  SDisk *pDisk = NULL;
+
+  ASSERT(pTier->nDisks > 0);
+
+  for (int i = 0; i < pTier->nDisks; i++) {
+    SDisk *iDisk = pTier->disks[i];
+    if (dnodeUpdateDiskMeta(iDisk) < 0) return NULL;
+    if (DNODE_DISK_AVAIL(iDisk)) {
+      if (pDisk == NULL || pDisk->dmeta.nfiles > iDisk->dmeta.nfiles) {
+        pDisk = iDisk;
+      }
+    }
+  }
+
+  if (pDisk == NULL) {
+    terrno = TSDB_CODE_DND_NO_DISK_SPACE;
+  }
+
+  return NULL;
+}
+
+SDisk *dnodeGetDiskByName(SDnodeTier *pDnodeTier, char *dirName) {
+  char     fdirName[TSDB_FILENAME_LEN] = "\0";
+  SDiskID *pDiskID = NULL;
+
+  if (dnodeFormatDir(dirName, fdirName) < 0) {
+    return NULL;
+  }
+
+  void *ptr = taosHashGet(pDnodeTier->map, (void *)fdirName, strnlen(fdirName, TSDB_FILENAME_LEN));
+  if (ptr == NULL) return NULL;
+  pDiskID = (SDiskID *)ptr;
+
+  return dnodeGetDisk(pDnodeTier, pDiskID->level, pDiskID->did);
+}
+
+static int dnodeFormatDir(char *idir, char *odir) {
+  wordexp_t wep;
+
+  int code = wordexp(idir, &wep, 0);
+  if (code != 0) {
+    dError("failed to format dir %s since %s", idir, strerror(code));
+    terrno = TAOS_SYSTEM_ERROR(code);
+    return -1;
+  }
+
+  if (realpath(wep.we_wordv[0], odir) == NULL) {
+    dError("failed to format dir %s since %s", idir, strerror(errno));
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    wordfree(&wep);
+    return -1;
+  }
+
+  wordfree(&wep);
+  return 0;
+}
+
+static int dnodeCheckDisk(char *dirName, int level, int primary) {
+  if (access(dirName, W_OK | R_OK | F_OK) != 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  struct stat pstat;
+  if (stat(dirName, &pstat) < 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  if (S_ISDIR(pstat.st_mode)) {
+    return 0;
+  } else {
+    terrno = TSDB_CODE_DND_DISK_NOT_DIRECTORY;
+    return -1;
+  }
+}
+
+static int dnodeUpdateDiskMeta(SDisk *pDisk) {
+  struct statvfs dstat;
+  if (statvfs(pDisk->dir, &dstat) < 0) {
+    dError("failed to get dir %s information since %s", pDisk->dir, strerror(errno));
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    return -1;
+  }
+
+  pDisk->dmeta.size = dstat.f_bsize * dstat.f_blocks;
+  pDisk->dmeta.free = dstat.f_bsize * dstat.f_bavail;
+
+  return 0;
+}
+
+static int dnodeAddDisk(SDnodeTier *pDnodeTier, char *dir, int level, int primary) {
+  char    dirName[TSDB_FILENAME_LEN] = "\0";
+  STier * pTier = NULL;
+  SDiskID diskid = {0};
+  SDisk * pDisk = NULL;
+
+  if (level < 0 || level >= DNODE_MAX_TIERS) {
+    terrno = TSDB_CODE_DND_INVALID_DISK_TIER;
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
+
+  if (dnodeFormatDir(dir, dirName) < 0) {
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
+
+  pTier = pDnodeTier->tiers + level;
+  diskid.level = level;
+
+  if (pTier->nDisks >= DNODE_MAX_DISKS_PER_TIER) {
+    terrno = TSDB_CODE_DND_TOO_MANY_DISKS;
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
+
+  if (dnodeGetDiskByName(pDnodeTier, dirName) != NULL) {
+    terrno = TSDB_CODE_DND_DISK_ALREADY_EXISTS;
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
+
+  if (dnodeCheckDisk(dirName, level, primary) < 0) {
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
+
+  if (primary) {
+    if (level != 0) {
+      terrno = TSDB_CODE_DND_INVALID_DISK_TIER;
+      dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+      return -1;
+    }
+
+    if (DNODE_PRIMARY_DISK(pDnodeTier) != NULL) {
+      terrno = TSDB_CODE_DND_DUPLICATE_PRIMARY_DISK;
+      dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+      return -1;
+    }
+
+    diskid.did = 0;
+  } else {
+    if (level == 0) {
+      if (DNODE_PRIMARY_DISK(pDnodeTier) != NULL) {
+        diskid.did = pTier->nDisks;
+      } else {
+        diskid.did = pTier->nDisks + 1;
+        if (diskid.did >= DNODE_MAX_DISKS_PER_TIER) {
+          terrno = TSDB_CODE_DND_TOO_MANY_DISKS;
+          dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+          return -1;
+        }
+      }
+    } else {
+      diskid.did = pTier->nDisks;
+    }
+  }
+
+  pDisk = (SDisk *)calloc(1, sizeof(SDisk));
+  if (pDisk == NULL) {
+    terrno = TSDB_CODE_DND_OUT_OF_MEMORY;
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
+
+  strncpy(pDisk->dir, dirName, TSDB_FILENAME_LEN);
 
-#ifndef _DNODE_PLUGINS
+  if (taosHashPut(pDnodeTier->map, (void *)dirName, strnlen(dirName, TSDB_FILENAME_LEN), (void *)(&diskid),
+                  sizeof(diskid)) < 0) {
+    free(pDisk);
+    terrno = TSDB_CODE_DND_OUT_OF_MEMORY;
+    dError("failed to add disk %s to tier %d level since %s", dir, level, tstrerror(terrno));
+    return -1;
+  }
 
-SDnodeTier *dnodeNewTier() { return NULL; }
-void *      dnodeCloseTier(SDnodeTier *pDnodeTier) { return NULL; }
-int         dnodeAddDisks(SDnodeTier *pDnodeTier, SDiskCfg *pDiskCfgs, int ndisks) { return 0; }
-int         dnodeUpdateTiersInfo(SDnodeTier *pDnodeTier) { return 0; }
-int         dnodeCheckTiers(SDnodeTier *pDnodeTier) { return 0; }
-SDisk *     dnodeAssignDisk(SDnodeTier *pDnodeTier, int level) { return NULL; }
-SDisk *     dnodeGetDiskByName(SDnodeTier *pDnodeTier, char *dirName) { return NULL; }
+  pTier->nDisks++;
+  pTier->disks[diskid.did] = pDisk;
+  pDnodeTier->nTiers = MAX(pDnodeTier->nTiers, level);
 
-#endif
\ No newline at end of file
+  return 0;
+}
\ No newline at end of file

src/tsdb/src/tsdbFile.c

@@ -18,21 +18,25 @@
 #define TAOS_RANDOM_FILE_FAIL_TEST
 
 #include "os.h"
+#include "tglobal.h"
 #include "talgo.h"
 #include "tchecksum.h"
 #include "tsdbMain.h"
 #include "tutil.h"
+#include "dnode.h"
+#include "tpath.h"
 
+struct SDnodeTier *pDnodeTier = NULL;
+const char *       tsdbFileSuffix[] = {".head", ".data", ".last", ".stat", ".h", ".d", ".l", ".s"};
 
-const char *tsdbFileSuffix[] = {".head", ".data", ".last", ".stat", ".h", ".d", ".l", ".s"};
-
-static int   tsdbInitFile(SFile *pFile, STsdbRepo *pRepo, int fid, int type);
-static void  tsdbDestroyFile(SFile *pFile);
-static int   compFGroup(const void *arg1, const void *arg2);
-static int   keyFGroupCompFunc(const void *key, const void *fgroup);
-static void  tsdbInitFileGroup(SFileGroup *pFGroup, STsdbRepo *pRepo);
-static TSKEY tsdbGetCurrMinKey(int8_t precision, int32_t keep);
-static int   tsdbGetCurrMinFid(int8_t precision, int32_t keep, int32_t days);
+static void      tsdbDestroyFile(SFile *pFile);
+static int       compFGroup(const void *arg1, const void *arg2);
+static int       keyFGroupCompFunc(const void *key, const void *fgroup);
+static TSKEY     tsdbGetCurrMinKey(int8_t precision, int32_t keep);
+static int       tsdbGetCurrMinFid(int8_t precision, int32_t keep, int32_t days);
+static int       tsdbLoadFilesFromDisk(STsdbRepo *pRepo, SDisk *pDisk);
+static SHashObj *tsdbGetAllFids(STsdbRepo *pRepo, char *dirName);
+static int       tsdbRestoreFileGroup(STsdbRepo *pRepo, SDisk *pDisk, int fid, SFileGroup *pFileGroup);
 
 // ---------------- INTERNAL FUNCTIONS ----------------
 STsdbFileH *tsdbNewFileH(STsdbCfg *pCfg) {
@@ -74,129 +78,25 @@ void tsdbFreeFileH(STsdbFileH *pFileH) {
 
 int tsdbOpenFileH(STsdbRepo *pRepo) {
   ASSERT(pRepo != NULL && pRepo->tsdbFileH != NULL);
+  char dataDir[TSDB_FILENAME_LEN] = "\0";
 
-  char *  tDataDir = NULL;
-  DIR *   dir = NULL;
-  int     fid = 0;
-  int     vid = 0;
-  regex_t regex1, regex2;
-  int     code = 0;
-  char    fname[TSDB_FILENAME_LEN] = "\0";
-
-  SFileGroup  fileGroup = {0};
-  STsdbFileH *pFileH = pRepo->tsdbFileH;
-  STsdbCfg *  pCfg = &(pRepo->config);
-
-  tDataDir = tsdbGetDataDirName(pRepo->rootDir);
-  if (tDataDir == NULL) {
-    terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
-    goto _err;
-  }
-
-  dir = opendir(tDataDir);
-  if (dir == NULL) {
-    tsdbError("vgId:%d failed to open directory %s since %s", REPO_ID(pRepo), tDataDir, strerror(errno));
-    terrno = TAOS_SYSTEM_ERROR(errno);
-    goto _err;
-  }
-
-  code = regcomp(&regex1, "^v[0-9]+f[0-9]+\\.(head|data|last|stat)$", REG_EXTENDED);
-  if (code != 0) {
-    terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
-    goto _err;
-  }
-
-  code = regcomp(&regex2, "^v[0-9]+f[0-9]+\\.(h|d|l|s)$", REG_EXTENDED);
-  if (code != 0) {
-    terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
-    goto _err;
-  }
-
-  int mfid = tsdbGetCurrMinFid(pCfg->precision, pCfg->keep, pCfg->daysPerFile);
-
-  struct dirent *dp = NULL;
-  while ((dp = readdir(dir)) != NULL) {
-    if (strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0) continue;
+  for (int level = 0; level < pDnodeTier->nTiers; level++) {
+    STier *pTier = pDnodeTier->tiers + level;
+    for (int did = 0; did < pTier->nDisks; did++) {
+      SDisk *pDisk = pTier->disks[did];
 
-    code = regexec(&regex1, dp->d_name, 0, NULL, 0);
-    if (code == 0) {
-      sscanf(dp->d_name, "v%df%d", &vid, &fid);
-      if (vid != REPO_ID(pRepo)) {
-        tsdbError("vgId:%d invalid file %s exists, ignore it", REPO_ID(pRepo), dp->d_name);
+      tdGetTsdbDataDir(pDisk->dir, REPO_ID(pRepo), dataDir);
+      
+      if (access(dataDir, F_OK) != 0) {
+        // Skip those disks without data
         continue;
       }
 
-      if (fid < mfid) {
-        for (int type = 0; type < TSDB_FILE_TYPE_MAX; type++) {
-          tsdbGetDataFileName(pRepo->rootDir, pCfg->tsdbId, fid, type, fname);
-          (void)remove(fname);
-        }
-        continue;
-      }
-
-      if (tsdbSearchFGroup(pFileH, fid, TD_EQ) != NULL) continue;
-      memset((void *)(&fileGroup), 0, sizeof(SFileGroup));
-      fileGroup.fileId = fid;
-
-      tsdbInitFileGroup(&fileGroup, pRepo);
-    } else if (code == REG_NOMATCH) {
-      code = regexec(&regex2, dp->d_name, 0, NULL, 0);
-      if (code == 0) {
-        size_t tsize = strlen(tDataDir) + strlen(dp->d_name) + 2;
-        char * fname1 = malloc(tsize);
-        if (fname1 == NULL) {
-          terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
-          goto _err;
-        }
-        sprintf(fname1, "%s/%s", tDataDir, dp->d_name);
-
-        tsize = tsize + 64;
-        char *fname2 = malloc(tsize);
-        if (fname2 == NULL) {
-          free(fname1);
-          terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
-          goto _err;
-        }
-        sprintf(fname2, "%s/%s_back_%" PRId64, tDataDir, dp->d_name, taosGetTimestamp(TSDB_TIME_PRECISION_MILLI));
-
-        (void)rename(fname1, fname2);
-
-        tsdbDebug("vgId:%d file %s exists, backup it as %s", REPO_ID(pRepo), fname1, fname2);
-
-        free(fname1);
-        free(fname2);
-        continue;
-      } else if (code == REG_NOMATCH) {
-        tsdbError("vgId:%d invalid file %s exists, ignore it", REPO_ID(pRepo), dp->d_name);
-        continue;
-      } else {
-        goto _err;
-      }
-    } else {
-      goto _err;
+      tsdbLoadFilesFromDisk(pRepo, pDisk);
     }
-
-    pFileH->pFGroup[pFileH->nFGroups++] = fileGroup;
-    qsort((void *)(pFileH->pFGroup), pFileH->nFGroups, sizeof(SFileGroup), compFGroup);
-    tsdbDebug("vgId:%d file group %d is restored, nFGroups %d", REPO_ID(pRepo), fileGroup.fileId, pFileH->nFGroups);
   }
 
-  regfree(&regex1);
-  regfree(&regex2);
-  taosTFree(tDataDir);
-  closedir(dir);
   return 0;
-
-_err:
-  for (int type = 0; type < TSDB_FILE_TYPE_MAX; type++) tsdbDestroyFile(&fileGroup.files[type]);
-
-  regfree(&regex1);
-  regfree(&regex2);
-
-  taosTFree(tDataDir);
-  if (dir != NULL) closedir(dir);
-  tsdbCloseFileH(pRepo);
-  return -1;
 }
 
 void tsdbCloseFileH(STsdbRepo *pRepo) {
@@ -522,37 +422,6 @@ _err:
 }
 
 // ---------------- LOCAL FUNCTIONS ----------------
-static int tsdbInitFile(SFile *pFile, STsdbRepo *pRepo, int fid, int type) {
-  uint32_t version;
-
-  tsdbGetDataFileName(pRepo->rootDir, REPO_ID(pRepo), fid, type, pFile->fname);
-
-  pFile->fd = -1;
-  if (tsdbOpenFile(pFile, O_RDONLY) < 0) goto _err;
-
-  if (tsdbLoadFileHeader(pFile, &version) < 0) {
-    tsdbError("vgId:%d failed to load file %s header part since %s", REPO_ID(pRepo), pFile->fname, tstrerror(terrno));
-    goto _err;
-  }
-
-  if (pFile->info.size == TSDB_FILE_HEAD_SIZE) {
-    pFile->info.size = lseek(pFile->fd, 0, SEEK_END);
-  }
-
-  if (version != TSDB_FILE_VERSION) {
-    // TODO: deal with error
-    tsdbError("vgId:%d file %s version %u is not the same as program version %u which may cause problem",
-              REPO_ID(pRepo), pFile->fname, version, TSDB_FILE_VERSION);
-  }
-
-  tsdbCloseFile(pFile);
-
-  return 0;
-_err:
-  tsdbDestroyFile(pFile);
-  return -1;
-}
-
 static void tsdbDestroyFile(SFile *pFile) { tsdbCloseFile(pFile); }
 
 static int compFGroup(const void *arg1, const void *arg2) {
@@ -578,22 +447,219 @@ static int keyFGroupCompFunc(const void *key, const void *fgroup) {
   }
 }
 
-static void tsdbInitFileGroup(SFileGroup *pFGroup, STsdbRepo *pRepo) {
+static TSKEY tsdbGetCurrMinKey(int8_t precision, int32_t keep) {
+  return (TSKEY)(taosGetTimestamp(precision) - keep * tsMsPerDay[precision]);
+}
+
+static int tsdbGetCurrMinFid(int8_t precision, int32_t keep, int32_t days) {
+  return (int)(TSDB_KEY_FILEID(tsdbGetCurrMinKey(precision, keep), days, precision));
+}
+
+static int tsdbLoadFilesFromDisk(STsdbRepo *pRepo, SDisk *pDisk) {
+  char                  tsdbDataDir[TSDB_FILENAME_LEN] = "\0";
+  char                  tsdbRootDir[TSDB_FILENAME_LEN] = "\0";
+  char                  fname[TSDB_FILENAME_LEN] = "\0";
+  SHashObj *            pFids = NULL;
+  SHashMutableIterator *pIter = NULL;
+  STsdbFileH *          pFileH = pRepo->tsdbFileH;
+  SFileGroup            fgroup = {0};
+  STsdbCfg *            pCfg = &(pRepo->config);
+  int                   mfid = 0;
+
+  tdGetTsdbRootDir(pDisk->dir, REPO_ID(pRepo), tsdbRootDir);
+  tdGetTsdbDataDir(pDisk->dir, REPO_ID(pRepo), tsdbDataDir);
+
+  pFids = tsdbGetAllFids(pRepo, tsdbDataDir);
+  if (pFids == NULL) {
+    goto _err;
+  }
+
+  pIter = taosHashCreateIter(pFids);
+  if (pIter == NULL) {
+    goto _err;
+  }
+
+  mfid = tsdbGetCurrMinFid(pCfg->precision, pCfg->keep, pCfg->daysPerFile);
+
+  while (taosHashIterNext(pIter)) {
+    int32_t fid = *(int32_t *)taosHashIterGet(pIter);
+
+    if (fid < mfid) {
+      for (int type = 0; type < TSDB_FILE_TYPE_MAX; type++) {
+        tsdbGetDataFileName(tsdbRootDir, REPO_ID(pRepo), fid, type, fname);
+        (void)remove(fname);
+      }
+
+      tsdbGetDataFileName(tsdbRootDir, REPO_ID(pRepo), fid, TSDB_FILE_TYPE_NHEAD, fname);
+      (void)remove(fname);
+
+      tsdbGetDataFileName(tsdbRootDir, REPO_ID(pRepo), fid, TSDB_FILE_TYPE_NLAST, fname);
+      (void)remove(fname);
+
+      continue;
+    }
+
+    tsdbRestoreFileGroup(pRepo, pDisk, fid, &fgroup);
+    pFileH->pFGroup[pFileH->nFGroups++] = fgroup;
+    qsort((void *)(pFileH->pFGroup), pFileH->nFGroups, sizeof(fgroup), compFGroup);
+
+    // TODO
+    pDisk->dmeta.nfiles++;
+  }
+
+  taosHashDestroyIter(pIter);
+  taosHashCleanup(pFids);
+  return 0;
+
+_err:
+  taosHashDestroyIter(pIter);
+  taosHashCleanup(pFids);
+  return -1;
+}
+
+static int tsdbRestoreFileGroup(STsdbRepo *pRepo, SDisk *pDisk, int fid, SFileGroup *pFileGroup) {
+  char tsdbRootDir[TSDB_FILENAME_LEN] = "\0";
+  char nheadF[TSDB_FILENAME_LEN] = "\0";
+  char nlastF[TSDB_FILENAME_LEN] = "\0";
+  bool newHeadExists = false;
+  bool newLastExists = false;
+
+  uint32_t version = 0;
+
+  terrno = TSDB_CODE_SUCCESS;
+
+  memset((void *)pFileGroup, 0, sizeof(*pFileGroup));
+  pFileGroup->fileId = fid;
   for (int type = 0; type < TSDB_FILE_TYPE_MAX; type++) {
-    if (tsdbInitFile(&pFGroup->files[type], pRepo, pFGroup->fileId, type) < 0) {
-      memset(&pFGroup->files[type].info, 0, sizeof(STsdbFileInfo));
-      pFGroup->files[type].info.magic = TSDB_FILE_INIT_MAGIC;
-      pFGroup->state = 1;
-      pRepo->state = TSDB_STATE_BAD_FILE;
+    SFile *pFile = pFileGroup->files + type;
+    pFile->fd = -1;
+  }
+
+  tdGetTsdbRootDir(pDisk->dir, REPO_ID(pRepo), tsdbRootDir);
+  for (int type = 0; type < TSDB_FILE_TYPE_MAX; type++) {
+    SFile *pFile = pFileGroup->files + type;
+    tsdbGetDataFileName(tsdbRootDir, REPO_ID(pRepo), fid, TSDB_FILE_TYPE_HEAD, pFile->fname);
+    if (access(pFile->fname, F_OK) != 0) {
+      memset(&(pFile->info), 0, sizeof(pFile->info));
+      pFile->info.magic = TSDB_FILE_INIT_MAGIC;
+      pFileGroup->state = 1;
       terrno = TSDB_CODE_TDB_FILE_CORRUPTED;
     }
   }
-}
 
-static TSKEY tsdbGetCurrMinKey(int8_t precision, int32_t keep) {
-  return (TSKEY)(taosGetTimestamp(precision) - keep * tsMsPerDay[precision]);
+  tsdbGetDataFileName(tsdbRootDir, REPO_ID(pRepo), fid, TSDB_FILE_TYPE_NHEAD, nheadF);
+  tsdbGetDataFileName(tsdbRootDir, REPO_ID(pRepo), fid, TSDB_FILE_TYPE_NLAST, nlastF);
+
+  if (access(nheadF, F_OK) == 0) {
+    newHeadExists = true;
+  }
+
+  if (access(nlastF, F_OK) == 0) {
+    newLastExists = true;
+  }
+
+  if (newHeadExists) {
+    (void)remove(nheadF);
+    (void)remove(nlastF);
+  } else {
+    if (newLastExists) {
+      (void)rename(nlastF, pFileGroup->files[TSDB_FILE_TYPE_LAST].fname);
+    }
+  }
+
+  if (terrno != TSDB_CODE_SUCCESS) {
+    return -1;
+  }
+
+  for (int type = 0; type < TSDB_FILE_TYPE_MAX; type++) {
+    SFile *pFile = pFileGroup->files + type;
+    if (tsdbOpenFile(pFile, O_RDONLY) < 0) {
+      memset(&(pFile->info), 0, sizeof(pFile->info));
+      pFile->info.magic = TSDB_FILE_INIT_MAGIC;
+      pFileGroup->state = 1;
+      terrno = TSDB_CODE_TDB_FILE_CORRUPTED;
+      continue;
+    }
+
+    if (tsdbLoadFileHeader(pFile, &version) < 0) {
+      memset(&(pFile->info), 0, sizeof(pFile->info));
+      pFile->info.magic = TSDB_FILE_INIT_MAGIC;
+      pFileGroup->state = 1;
+      terrno = TSDB_CODE_TDB_FILE_CORRUPTED;
+      tsdbCloseFile(pFile);
+      continue;
+    }
+
+    if (version != TSDB_FILE_VERSION) {
+      tsdbError("vgId:%d file %s version %u is not the same as program version %u which may cause problem",
+                REPO_ID(pRepo), pFile->fname, version, TSDB_FILE_VERSION);
+    }
+
+    tsdbCloseFile(pFile);
+  }
+
+  if (terrno != TSDB_CODE_SUCCESS) {
+    return -1;
+  } else {
+    return 0;
+  }
 }
 
-static int tsdbGetCurrMinFid(int8_t precision, int32_t keep, int32_t days) {
-  return (int)(TSDB_KEY_FILEID(tsdbGetCurrMinKey(precision, keep), days, precision));
+static SHashObj *tsdbGetAllFids(STsdbRepo *pRepo, char *dirName) {
+  DIR *     dir = NULL;
+  regex_t   regex = {0};
+  int       code = 0;
+  int32_t   vid, fid;
+  SHashObj *pHash = NULL;
+
+  code = regcomp(&regex, "^v[0-9]+f[0-9]+\\.(head|data|last|h|d|l)$", REG_EXTENDED);
+  if (code != 0) {
+    terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
+    goto _err;
+  }
+
+  dir = opendir(dirName);
+  if (dir == NULL) {
+    tsdbError("vgId:%d failed to open directory %s since %s", REPO_ID(pRepo), dirName, strerror(errno));
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    goto _err;
+  }
+
+  pHash = taosHashInit(1024, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false, HASH_NO_LOCK);
+  if (pHash == NULL) {
+    terrno = TSDB_CODE_TDB_OUT_OF_MEMORY;
+    goto _err;
+  }
+
+  struct dirent *dp = NULL;
+  while ((dp = readdir(dir)) != NULL) {
+    if (strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0) continue;
+
+    code = regexec(&regex, dp->d_name, 0, NULL, 0);
+    if (code == 0) {
+      sscanf(dp->d_name, "v%df%d", &vid, &fid);
+
+      if (vid != REPO_ID(pRepo)) {
+        tsdbError("vgId:%d invalid file %s exists, ignore it", REPO_ID(pRepo), dp->d_name);
+        continue;
+      }
+
+      taosHashPut(pHash, (void *)(&fid), sizeof(fid), (void *)(&fid), sizeof(fid));
+    } else if (code == REG_NOMATCH) {
+      tsdbError("vgId:%d invalid file %s exists, ignore it", REPO_ID(pRepo), dp->d_name);
+      continue;
+    } else {
+      goto _err;
+    }
+  }
+
+  closedir(dir);
+  regfree(&regex);
+  return pHash;
+
+_err:
+  taosHashCleanup(pHash);
+  if (dir != NULL) closedir(dir);
+  regfree(&regex);
+  return NULL;
 }
\ No newline at end of file

src/vnode/src/vnodeMain.c

@@ -105,7 +105,7 @@ int32_t vnodeCreate(SMDCreateVnodeMsg *pVnodeCfg) {
   }
 
   char rootDir[TSDB_FILENAME_LEN] = {0};
-  sprintf(rootDir, "%s/vnode%d", tsVnodeDir, pVnodeCfg->cfg.vgId);
+  tdGetVnodeDir(tsDataDir, pVnodeCfg->cfg.vgId, rootDir);
   if (mkdir(rootDir, 0755) != 0 && errno != EEXIST) {
     vError("vgId:%d, failed to create vnode, reason:%s dir:%s", pVnodeCfg->cfg.vgId, strerror(errno), rootDir);
     if (errno == EACCES) {
@@ -138,7 +138,7 @@ int32_t vnodeCreate(SMDCreateVnodeMsg *pVnodeCfg) {
   tsdbCfg.compression         = pVnodeCfg->cfg.compression;
 
   char tsdbDir[TSDB_FILENAME_LEN] = {0};
-  sprintf(tsdbDir, "%s/vnode%d/tsdb", tsVnodeDir, pVnodeCfg->cfg.vgId);
+  tdGetTsdbRootDir(tsDataDir, pVnodeCfg->cfg.vgId, tsdbDir);
   if (tsdbCreateRepo(tsdbDir, &tsdbCfg) < 0) {
     vError("vgId:%d, failed to create tsdb in vnode, reason:%s", pVnodeCfg->cfg.vgId, tstrerror(terrno));
     return TSDB_CODE_VND_INIT_FAILED;