force repair

source/libs/wal/src/walMeta.c

@@ -65,6 +65,7 @@ static inline int64_t walScanLogGetLastVer(SWal* pWal) {
   struct stat statbuf;
   stat(fnameStr, &statbuf);
   int readSize = MIN(WAL_MAX_SIZE + 2, statbuf.st_size);
+  pLastFileInfo->fileSize = statbuf.st_size;
 
   FileFd fd = taosOpenFileRead(fnameStr);
   if (fd < 0) {
@@ -91,7 +92,7 @@ static inline int64_t walScanLogGetLastVer(SWal* pWal) {
   
   char* haystack = buf;
   char* found = NULL;
-  char *candidate = NULL;
+  char *candidate;
   while((candidate = tmemmem(haystack, readSize - (haystack - buf), (char*)&magic, sizeof(uint64_t))) != NULL) {
     // read and validate
     SWalHead *logContent = (SWalHead*)candidate;
@@ -142,7 +143,6 @@ int walCheckAndRepairMeta(SWal* pWal) {
       SWalFileInfo fileInfo;
       memset(&fileInfo, -1, sizeof(SWalFileInfo));
       sscanf(name, "%" PRId64 ".log", &fileInfo.firstVer);
-      FileFd fd = taosOpenFileRead(ent->d_name);
       //get lastVer
       //get size
       taosArrayPush(pLogInfoArray, &fileInfo);
@@ -159,28 +159,25 @@ int walCheckAndRepairMeta(SWal* pWal) {
   }
   int newSz = taosArrayGetSize(pLogInfoArray);
   // case 1. meta file not exist / cannot be parsed
-  if (pWal->fileInfoSet == NULL && newSz != 0) {
-    // recover fileInfo set
-    pWal->fileInfoSet = pLogInfoArray;
-    if (newSz != 0) {
-    // recover meta version
-      pWal->vers.firstVer = ((SWalFileInfo*)taosArrayGet(pLogInfoArray, 0))->firstVer;
-      pWal->writeCur = newSz - 1;
-    }
-    // recover file size
-  } else if (oldSz < newSz) {
+  if (oldSz < newSz) {
     for (int i = oldSz; i < newSz; i++) {
       SWalFileInfo *pFileInfo = taosArrayGet(pLogInfoArray, i);
       taosArrayPush(pWal->fileInfoSet, pFileInfo);
     }
+
     pWal->writeCur = newSz - 1;
+    pWal->vers.firstVer = ((SWalFileInfo*)taosArrayGet(pLogInfoArray, 0))->firstVer;
+    pWal->vers.lastVer = walScanLogGetLastVer(pWal);
+    ((SWalFileInfo*)taosArrayGetLast(pWal->fileInfoSet))->lastVer = pWal->vers.lastVer;
+    ASSERT(pWal->vers.lastVer != -1);
+
+    int code = walSaveMeta(pWal);
+    if (code < 0) {
+      taosArrayDestroy(pLogInfoArray);
+      return -1;
+    }
   }
   
-  if (pWal->fileInfoSet && taosArrayGetSize(pWal->fileInfoSet) != 0) {
-    pWal->vers.lastVer = walScanLogGetLastVer(pWal);
-    ASSERT(pWal->vers.lastVer != -1);
-  }
-
   // case 2. versions in meta not match log 
   //         or some log not included in meta
   // (e.g. program killed)
@@ -204,14 +201,11 @@ int walCheckAndRepairMeta(SWal* pWal) {
   }
 #endif
 
-  int code = walSaveMeta(pWal);
-  if (code < 0) {
-    return -1;
-  }
 
   // get last version of this file
   //
   // rebuild meta
+  taosArrayDestroy(pLogInfoArray);
   return 0;
 }
 
@@ -419,6 +413,10 @@ int walLoadMeta(SWal* pWal) {
   }
   memset(buf, 0, size + 5);
   FileFd fd = taosOpenFileRead(fnameStr);
+  if (fd < 0) {
+    terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
+    return -1;
+  }
   if (taosReadFile(fd, buf, size) != size) {
     terrno = TAOS_SYSTEM_ERROR(errno);
     taosCloseFile(fd);

source/libs/wal/src/walMgmt.c

@@ -122,7 +122,9 @@ SWal *walOpen(const char *path, SWalCfg *pCfg) {
     return NULL;
   }
 
-  if (walLoadMeta(pWal) < 0 && walCheckAndRepairMeta(pWal) < 0) {
+  walLoadMeta(pWal);
+
+  if (walCheckAndRepairMeta(pWal) < 0) {
     taosRemoveRef(tsWal.refSetId, pWal->refId);
     pthread_mutex_destroy(&pWal->mutex);
     taosArrayDestroy(pWal->fileInfoSet);
@@ -131,6 +133,7 @@ SWal *walOpen(const char *path, SWalCfg *pCfg) {
   }
 
   if (walCheckAndRepairIdx(pWal) < 0) {
+
   }
 
   wDebug("vgId:%d, wal:%p is opened, level:%d fsyncPeriod:%d", pWal->cfg.vgId, pWal, pWal->cfg.level,

source/libs/wal/test/walMetaTest.cpp

@@ -380,4 +380,26 @@ TEST_F(WalRetentionEnv, repairMeta1) {
     ASSERT_EQ(code, 0);
   }
 
+  for (int i = 0; i < 1000; i++) {
+    int ver = rand() % 200;
+    code = walReadWithHandle(pRead, ver);
+    ASSERT_EQ(code, 0);
+
+    // printf("rrbody: \n");
+    // for(int i = 0; i < pRead->pHead->head.len; i++) {
+    // printf("%d ", pRead->pHead->head.body[i]);
+    //}
+    // printf("\n");
+
+    ASSERT_EQ(pRead->pHead->head.version, ver);
+    ASSERT_EQ(pRead->curVersion, ver + 1);
+    char newStr[100];
+    sprintf(newStr, "%s-%d", ranStr, ver);
+    int len = strlen(newStr);
+    ASSERT_EQ(pRead->pHead->head.len, len);
+    for (int j = 0; j < len; j++) {
+      EXPECT_EQ(newStr[j], pRead->pHead->head.body[j]);
+    }
+  }
+
 }

tests/script/sim/db/basic1.sim

@@ -85,7 +85,6 @@ if $data02 != 2 then
   return -1
 endi
 
-return
 system sh/exec.sh -n dnode1 -s stop -x SIGKILL
 system sh/exec.sh -n dnode1 -s start
 
@@ -104,4 +103,4 @@ if $rows != 2 then
   return -1
 endi
 
-system sh/exec.sh -n dnode1 -s stop -x SIGINT
+system sh/exec.sh -n dnode1 -s stop -x SIGINT