Merge pull request #3865 from taosdata/feature/crash

[TD-857]<fix>: Allow vnode start even when read version files fails, #3703 

src/inc/taoserror.h

@@ -193,6 +193,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_VND_NO_DISK_PERMISSIONS,      0, 0x0506, "No write p
 TAOS_DEFINE_ERROR(TSDB_CODE_VND_NO_SUCH_FILE_OR_DIR,      0, 0x0507, "Missing data file")
 TAOS_DEFINE_ERROR(TSDB_CODE_VND_OUT_OF_MEMORY,            0, 0x0508, "Out of memory")
 TAOS_DEFINE_ERROR(TSDB_CODE_VND_APP_ERROR,                0, 0x0509, "Unexpected generic error in vnode")
+TAOS_DEFINE_ERROR(TSDB_CODE_VND_INVALID_VRESION_FILE,     0, 0x050A, "Invalid version file")
 TAOS_DEFINE_ERROR(TSDB_CODE_VND_NOT_SYNCED,               0, 0x0511, "Database suspended")
 TAOS_DEFINE_ERROR(TSDB_CODE_VND_NO_WRITE_AUTH,            0, 0x0512, "Write operation denied")
 

src/inc/twal.h

@@ -51,6 +51,7 @@ int     walWrite(twalh, SWalHead *);
 void    walFsync(twalh);
 int     walRestore(twalh, void *pVnode, FWalWrite writeFp);
 int     walGetWalFile(twalh, char *name, uint32_t *index);
+int64_t walGetVersion(twalh);
 
 extern int wDebugFlag;
 

src/vnode/src/vnodeMain.c

@@ -239,8 +239,10 @@ int32_t vnodeOpen(int32_t vnode, char *rootDir) {
 
   code = vnodeReadVersion(pVnode);
   if (code != TSDB_CODE_SUCCESS) {
-    vnodeCleanUp(pVnode);
-    return code;
+    vError("vgId:%d, failed to read version, generate it from data file", pVnode->vgId);
+    // Allow vnode start even when read version fails, set version as walVersion or zero
+    // vnodeCleanUp(pVnode);
+    // return code;
   }
 
   pVnode->fversion = pVnode->version;
@@ -292,6 +294,9 @@ int32_t vnodeOpen(int32_t vnode, char *rootDir) {
   }
 
   walRestore(pVnode->wal, pVnode, vnodeWriteToQueue);
+  if (pVnode->version == 0) {
+    pVnode->version = walGetVersion(pVnode->wal);
+  }
 
   SSyncInfo syncInfo;
   syncInfo.vgId = pVnode->vgId;
@@ -947,6 +952,7 @@ static int32_t vnodeSaveVersion(SVnodeObj *pVnode) {
   len += snprintf(content + len, maxLen - len, "}\n");
 
   fwrite(content, 1, len, fp);
+  fflush(fp);
   fclose(fp);
 
   vInfo("vgId:%d, save vnode version:%" PRId64 " succeed", pVnode->vgId, pVnode->fversion);
@@ -960,7 +966,7 @@ static int32_t vnodeReadVersion(SVnodeObj *pVnode) {
   cJSON  *root = NULL;
   int     maxLen = 100;
 
-  terrno = TSDB_CODE_VND_APP_ERROR;
+  terrno = TSDB_CODE_VND_INVALID_VRESION_FILE;
   sprintf(versionFile, "%s/vnode%d/version.json", tsVnodeDir, pVnode->vgId);
   FILE *fp = fopen(versionFile, "r");
   if (!fp) {
@@ -974,7 +980,7 @@ static int32_t vnodeReadVersion(SVnodeObj *pVnode) {
   }
 
   content = calloc(1, maxLen + 1);
-  int   len = fread(content, 1, maxLen, fp);
+  int len = fread(content, 1, maxLen, fp);
   if (len <= 0) {
     vError("vgId:%d, failed to read vnode version, content is null", pVnode->vgId);
     goto PARSE_OVER;
@@ -999,6 +1005,6 @@ static int32_t vnodeReadVersion(SVnodeObj *pVnode) {
 PARSE_OVER:
   taosTFree(content);
   cJSON_Delete(root);
-  if(fp) fclose(fp);
+  if (fp) fclose(fp);
   return terrno;
 }

src/wal/src/walMain.c

@@ -240,10 +240,7 @@ int walWrite(void *handle, SWalHead *pHead) {
 
   // no wal
   if (pWal->level == TAOS_WAL_NOLOG) return 0;
-  if (pHead->version <= pWal->version) {
-    wError("wal:%s, failed to write ver:%" PRIu64 ", last ver:%" PRIu64, pWal->name, pHead->version, pWal->version);
-    return 0;
-  }
+  if (pHead->version <= pWal->version) return 0;
 
   pHead->signature = walSignature;
   taosCalcChecksumAppend(0, (uint8_t *)pHead, sizeof(SWalHead));
@@ -560,3 +557,10 @@ static void walProcessFsyncTimer(void *param, void *tmrId) {
     pWal->timer = NULL;
   }
 }
+
+int64_t walGetVersion(twalh param) {
+  SWal *pWal = param;
+  if (pWal == 0) return 0;
+
+  return pWal->version;
+}
\ No newline at end of file