Merge pull request #4260 from taosdata/feature/wal

[TD-2051]<fix>: taosd will crash if there a hostame not exist  #4249

src/inc/taoserror.h

@@ -201,7 +201,7 @@ TAOS_DEFINE_ERROR(TSDB_CODE_VND_NO_SUCH_FILE_OR_DIR,      0, 0x0507, "Missing da
 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_IS_FULL,                  0, 0x050B, "Vnode memory is full for commit is failed")
+TAOS_DEFINE_ERROR(TSDB_CODE_VND_IS_FULL,                  0, 0x050B, "Vnode memory is full because commit failed")
 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/sync/inc/syncInt.h

@@ -36,6 +36,7 @@ extern "C" {
 #define TAOS_SMSG_STATUS       7
 
 #define SYNC_MAX_SIZE (TSDB_MAX_WAL_SIZE + sizeof(SWalHead) + sizeof(SSyncHead) + 16)
+#define SYNC_RECV_BUFFER_SIZE (5*1024*1024)
 
 #define nodeRole    pNode->peerInfo[pNode->selfIndex]->role
 #define nodeVersion pNode->peerInfo[pNode->selfIndex]->version

src/sync/src/syncMain.c

@@ -179,6 +179,13 @@ int64_t syncStart(const SSyncInfo *pInfo) {
   for (int32_t i = 0; i < pCfg->replica; ++i) {
     const SNodeInfo *pNodeInfo = pCfg->nodeInfo + i;
     pNode->peerInfo[i] = syncAddPeer(pNode, pNodeInfo);
+    if (pNode->peerInfo[i] == NULL) {
+      sError("vgId:%d, node:%d fqdn:%s port:%u is not configured, stop taosd", pNode->vgId, pNodeInfo->nodeId, pNodeInfo->nodeFqdn,
+             pNodeInfo->nodePort);
+      syncStop(pNode->rid);
+      exit(1);
+    }
+
     if ((strcmp(pNodeInfo->nodeFqdn, tsNodeFqdn) == 0) && (pNodeInfo->nodePort == tsSyncPort)) {
       pNode->selfIndex = i;
     }
@@ -476,7 +483,11 @@ static void syncRemovePeer(SSyncPeer *pPeer) {
 
 static SSyncPeer *syncAddPeer(SSyncNode *pNode, const SNodeInfo *pInfo) {
   uint32_t ip = taosGetIpFromFqdn(pInfo->nodeFqdn);
-  if (ip == -1) return NULL;
+  if (ip == 0xFFFFFFFF) {
+    sError("failed to add peer, can resolve fqdn:%s since %s", pInfo->nodeFqdn, strerror(errno));
+    terrno = TSDB_CODE_RPC_FQDN_ERROR;
+    return NULL;
+  }
 
   SSyncPeer *pPeer = calloc(1, sizeof(SSyncPeer));
   if (pPeer == NULL) return NULL;

src/sync/src/syncRestore.c

@@ -136,7 +136,7 @@ static int32_t syncRestoreWal(SSyncPeer *pPeer) {
   SSyncNode *pNode = pPeer->pSyncNode;
   int32_t    ret, code = -1;
 
-  void *buffer = calloc(1024000, 1);  // size for one record
+  void *buffer = calloc(SYNC_MAX_SIZE, 1);  // size for one record
   if (buffer == NULL) return -1;
 
   SWalHead *pHead = (SWalHead *)buffer;
@@ -237,7 +237,7 @@ static int32_t syncOpenRecvBuffer(SSyncNode *pNode) {
   SRecvBuffer *pRecv = calloc(sizeof(SRecvBuffer), 1);
   if (pRecv == NULL) return -1;
 
-  pRecv->bufferSize = 5000000;
+  pRecv->bufferSize = SYNC_RECV_BUFFER_SIZE;
   pRecv->buffer = malloc(pRecv->bufferSize);
   if (pRecv->buffer == NULL) {
     free(pRecv);

src/util/src/tsocket.c

@@ -58,13 +58,13 @@ uint32_t taosGetIpFromFqdn(const char *fqdn) {
   } else {
 #ifdef EAI_SYSTEM
     if (ret == EAI_SYSTEM) {
-      uError("failed to get the ip address, fqdn:%s, code:%d, reason:%s", fqdn, ret, strerror(errno));
+      uError("failed to get the ip address, fqdn:%s, since:%s", fqdn, strerror(errno));
       terrno = TAOS_SYSTEM_ERROR(errno);
     } else {
-      uError("failed to get the ip address, fqdn:%s, code:%d, reason:%s", fqdn, ret, gai_strerror(ret));
+      uError("failed to get the ip address, fqdn:%s, since:%s", fqdn, gai_strerror(ret));
     }
 #else
-    uError("failed to get the ip address, fqdn:%s, code:%d, reason:%s", fqdn, ret, gai_strerror(ret));
+    uError("failed to get the ip address, fqdn:%s, since:%s", fqdn, gai_strerror(ret));
 #endif
     return 0xFFFFFFFF;
   }

src/vnode/src/vnodeMain.c

@@ -419,7 +419,11 @@ void vnodeRelease(void *pVnodeRaw) {
   }
 
   if (pVnode->wal) {
-    if (code == 0) walRemoveAllOldFiles(pVnode->wal);
+    if (code != 0) {
+      vError("vgId:%d, failed to commit while close tsdb repo, keep wal", pVnode->vgId);
+    } else {
+      walRemoveAllOldFiles(pVnode->wal);
+    }
     walClose(pVnode->wal);
     pVnode->wal = NULL;
   }