Merge pull request #21957 from taosdata/FIX/TD-24149-main

enh: terminate on IO error while appending wal logs or flushing data

source/dnode/vnode/src/vnd/vnodeCommit.c

@@ -360,7 +360,12 @@ static int32_t vnodeCommitTask(void *arg) {
 
   // commit
   code = vnodeCommitImpl(pInfo);
-  if (code) goto _exit;
+  if (code) {
+    vFatal("vgId:%d, failed to commit vnode since %s", TD_VID(pVnode), terrstr());
+    taosMsleep(100);
+    exit(EXIT_FAILURE);
+    goto _exit;
+  }
 
   vnodeReturnBufPool(pVnode);
 

source/libs/wal/src/walMeta.c

@@ -602,18 +602,18 @@ int walCheckAndRepairIdxFile(SWal* pWal, int32_t fileIdx) {
   // ftruncate idx file
   if (offset < fileSize) {
     if (taosFtruncateFile(pIdxFile, offset) < 0) {
-      wError("vgId:%d, failed to ftruncate file due to %s. offset:%" PRId64 ", file:%s", pWal->cfg.vgId,
-             strerror(errno), offset, fnameStr);
       terrno = TAOS_SYSTEM_ERROR(errno);
+      wError("vgId:%d, failed to ftruncate file since %s. offset:%" PRId64 ", file:%s", pWal->cfg.vgId, terrstr(),
+             offset, fnameStr);
       goto _err;
     }
   }
 
   // rebuild idx file
   if (taosLSeekFile(pIdxFile, 0, SEEK_END) < 0) {
-    wError("vgId:%d, failed to seek file due to %s. offset:%" PRId64 ", file:%s", pWal->cfg.vgId, strerror(errno),
-           offset, fnameStr);
     terrno = TAOS_SYSTEM_ERROR(errno);
+    wError("vgId:%d, failed to seek file since %s. offset:%" PRId64 ", file:%s", pWal->cfg.vgId, terrstr(), offset,
+           fnameStr);
     goto _err;
   }
 
@@ -625,11 +625,12 @@ int walCheckAndRepairIdxFile(SWal* pWal, int32_t fileIdx) {
     idxEntry.offset += sizeof(SWalCkHead) + ckHead.head.bodyLen;
 
     if (walReadLogHead(pLogFile, idxEntry.offset, &ckHead) < 0) {
-      wError("vgId:%d, failed to read wal log head since %s. offset:%" PRId64 ", file:%s", pWal->cfg.vgId, terrstr(),
+      wError("vgId:%d, failed to read wal log head since %s. index:%" PRId64 ", offset:%" PRId64 ", file:%s",
-             idxEntry.offset, fLogNameStr);
+             pWal->cfg.vgId, terrstr(), idxEntry.ver, idxEntry.offset, fLogNameStr);
       goto _err;
     }
     if (taosWriteFile(pIdxFile, &idxEntry, sizeof(SWalIdxEntry)) < 0) {
+      terrno = TAOS_SYSTEM_ERROR(errno);
       wError("vgId:%d, failed to append file since %s. file:%s", pWal->cfg.vgId, terrstr(), fnameStr);
       goto _err;
     }
@@ -637,6 +638,7 @@ int walCheckAndRepairIdxFile(SWal* pWal, int32_t fileIdx) {
   }
 
   if (taosFsyncFile(pIdxFile) < 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
     wError("vgId:%d, faild to fsync file since %s. file:%s", pWal->cfg.vgId, terrstr(), fnameStr);
     goto _err;
   }

source/libs/wal/src/walWrite.c

@@ -473,7 +473,10 @@ static int32_t walWriteIndex(SWal *pWal, int64_t ver, int64_t offset) {
   // check alignment of idx entries
   int64_t endOffset = taosLSeekFile(pWal->pIdxFile, 0, SEEK_END);
   if (endOffset < 0) {
-    wFatal("vgId:%d, failed to seek end of idxfile due to %s. ver:%" PRId64 "", pWal->cfg.vgId, strerror(errno), ver);
+    wFatal("vgId:%d, failed to seek end of WAL idxfile due to %s. ver:%" PRId64 "", pWal->cfg.vgId, strerror(errno),
+           ver);
+    taosMsleep(100);
+    exit(EXIT_FAILURE);
   }
   return 0;
 }
@@ -533,16 +536,20 @@ static FORCE_INLINE int32_t walWriteImpl(SWal *pWal, int64_t index, tmsg_t msgTy
 END:
   // recover in a reverse order
   if (taosFtruncateFile(pWal->pLogFile, offset) < 0) {
-    wFatal("vgId:%d, failed to ftruncate logfile to offset:%" PRId64 " during recovery due to %s", pWal->cfg.vgId,
-           offset, strerror(errno));
     terrno = TAOS_SYSTEM_ERROR(errno);
+    wFatal("vgId:%d, failed to recover WAL logfile from write error since %s, offset:%" PRId64, pWal->cfg.vgId,
+           terrstr(), offset);
+    taosMsleep(100);
+    exit(EXIT_FAILURE);
   }
 
   int64_t idxOffset = (index - pFileInfo->firstVer) * sizeof(SWalIdxEntry);
   if (taosFtruncateFile(pWal->pIdxFile, idxOffset) < 0) {
-    wFatal("vgId:%d, failed to ftruncate idxfile to offset:%" PRId64 "during recovery due to %s", pWal->cfg.vgId,
-           idxOffset, strerror(errno));
     terrno = TAOS_SYSTEM_ERROR(errno);
+    wFatal("vgId:%d, failed to recover WAL idxfile from write error since %s, offset:%" PRId64, pWal->cfg.vgId,
+           terrstr(), idxOffset);
+    taosMsleep(100);
+    exit(EXIT_FAILURE);
   }
   return -1;
 }

source/util/src/tlog.c

@@ -486,24 +486,11 @@ static inline int32_t taosBuildLogHead(char *buffer, const char *flags) {
 static inline void taosPrintLogImp(ELogLevel level, int32_t dflag, const char *buffer, int32_t len) {
   if ((dflag & DEBUG_FILE) && tsLogObj.logHandle && tsLogObj.logHandle->pFile != NULL && osLogSpaceAvailable()) {
     taosUpdateLogNums(level);
-#if 0 
-    // DEBUG_FATAL and DEBUG_ERROR are duplicated
-    // fsync will cause thread blocking and may also generate log misalignment in case of asyncLog
-    if (tsAsyncLog && level != DEBUG_FATAL) {
-      taosPushLogBuffer(tsLogObj.logHandle, buffer, len);
-    } else {
-      taosWriteFile(tsLogObj.logHandle->pFile, buffer, len);
-      if (level == DEBUG_FATAL) {
-        taosFsyncFile(tsLogObj.logHandle->pFile);
-      }
-    }
-#else
     if (tsAsyncLog) {
       taosPushLogBuffer(tsLogObj.logHandle, buffer, len);
     } else {
       taosWriteFile(tsLogObj.logHandle->pFile, buffer, len);
     }
-#endif
 
     if (tsLogObj.maxLines > 0) {
       atomic_add_fetch_32(&tsLogObj.lines, 1);