Merge pull request #769 from taosdata/feature/slguan

Solve the bug of incomplete vnode deletion and vpeer synchronization

src/os/linux/inc/os.h

@@ -224,6 +224,7 @@ int64_t str2int64(char *str);
 
 void taosSetCoreDump();
 
+void taosBlockSIGPIPE();
 
 #define BUILDIN_CLZL(val) __builtin_clzl(val)
 #define BUILDIN_CLZ(val) __builtin_clz(val)

src/os/linux/src/tlinux.c

@@ -341,3 +341,13 @@ bool taosSkipSocketCheck() {
 
   return false;
 }
+
+void taosBlockSIGPIPE() {
+  sigset_t signal_mask;
+  sigemptyset(&signal_mask);
+  sigaddset(&signal_mask, SIGPIPE);
+  int rc = pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
+  if (rc != 0) {
+    pError("failed to block SIGPIPE");
+  }
+}

src/system/detail/src/dnodeMgmt.c

@@ -350,8 +350,9 @@ int vnodeProcessVPeerCfg(char *msg, int msgLen, SMgmtObj *pMgmtObj) {
   pCfg->rowsInFileBlock = htonl(pCfg->rowsInFileBlock);
 
   if (pCfg->replications > 0) {
-    dPrint("vid:%d, vpeer cfg received, replica:%d session:%d, vnodeList replica:%d session:%d",
-        vnode, pCfg->replications, pCfg->maxSessions, vnodeList[vnode].cfg.replications, vnodeList[vnode].cfg.maxSessions);
+    dPrint("vid:%d, vpeer cfg received, replica:%d session:%d, vnodeList replica:%d session:%d, acct:%s db:%s",
+        vnode, pCfg->replications, pCfg->maxSessions, vnodeList[vnode].cfg.replications, vnodeList[vnode].cfg.maxSessions,
+        pCfg->acct, pCfg->db);
     for (i = 0; i < pCfg->replications; ++i) {
       pMsg->vpeerDesc[i].vnode = htonl(pMsg->vpeerDesc[i].vnode);
       pMsg->vpeerDesc[i].ip = htonl(pMsg->vpeerDesc[i].ip);

src/system/detail/src/mgmtDnodeInt.c

@@ -142,7 +142,7 @@ int mgmtProcessVPeersRsp(char *msg, int msgLen, SDnodeObj *pObj) {
 
   SDbObj *pDb = mgmtGetDb(pRsp->more);
   if (!pDb) {
-    mError("dnode:%s, db not find, code:%d", taosIpStr(pObj->privateIp), pRsp->code);
+    mError("dnode:%s, db:%s not find, code:%d", taosIpStr(pObj->privateIp), pRsp->more, pRsp->code);
     return 0;
   }
 

src/system/detail/src/vnodeMeter.c

@@ -24,6 +24,7 @@
 #include "vnodeMgmt.h"
 #include "vnodeShell.h"
 #include "vnodeUtil.h"
+#include "tstatus.h"
 
 #pragma GCC diagnostic ignored "-Wpointer-sign"
 
@@ -717,7 +718,8 @@ void vnodeUpdateMeter(void *param, void *tmrId) {
   SVnodeObj* pVnode = &vnodeList[pNew->vnode];
 
   if (pVnode->meterList == NULL) {
-    dTrace("vid:%d sid:%d id:%s, vnode is deleted, abort update schema", pNew->vnode, pNew->sid, pNew->meterId);
+    dTrace("vid:%d sid:%d id:%s, vnode is deleted, status:%s, abort update schema",
+            pNew->vnode, pNew->sid, pNew->meterId, taosGetVnodeStatusStr(vnodeList[pNew->vnode].vnodeStatus));
     free(pNew->schema);
     free(pNew);
     return;

src/system/detail/src/vnodeShell.c

@@ -184,13 +184,15 @@ int vnodeOpenShellVnode(int vnode) {
 
 static void vnodeDelayedFreeResource(void *param, void *tmrId) {
   int32_t vnode = *(int32_t*) param;
-  dTrace("vid:%d, start to free resources", vnode);
+  dTrace("vid:%d, start to free resources for 500ms arrived", vnode);
 
   taosCloseRpcChann(pShellServer, vnode); // close connection
   tfree(shellList[vnode]);  //free SShellObj
   tfree(param);
 
   memset(vnodeList + vnode, 0, sizeof(SVnodeObj));
+  dTrace("vid:%d, status set to %s", vnode, taosGetVnodeStatusStr(vnodeList[vnode].vnodeStatus));
+
   vnodeCalcOpenVnodes();
 }
 

src/system/detail/src/vnodeStore.c

@@ -250,7 +250,7 @@ static void vnodeRemoveDataFiles(int vnode) {
 
   sprintf(vnodeDir, "%s/vnode%d", tsDirectory, vnode);
   rmdir(vnodeDir);
-  dPrint("vid:%d, vnode is removed!", vnode);
+  dPrint("vid:%d, vnode is removed, status:%s", vnode, taosGetVnodeStatusStr(vnodeList[vnode].vnodeStatus));
 }
 
 int vnodeRemoveVnode(int vnode) {
@@ -261,7 +261,7 @@ int vnodeRemoveVnode(int vnode) {
     if (pVnode->vnodeStatus == TSDB_VN_STATUS_CREATING
         || pVnode->vnodeStatus == TSDB_VN_STATUS_OFFLINE
         || pVnode->vnodeStatus == TSDB_VN_STATUS_DELETING) {
-      dError("vid:%d, status:%s, cannot enter close/delete operation", vnode, taosGetVnodeStatusStr(pVnode->vnodeStatus));
+      dTrace("vid:%d, status:%s, cannot enter close/delete operation", vnode, taosGetVnodeStatusStr(pVnode->vnodeStatus));
       return TSDB_CODE_ACTION_IN_PROGRESS;
     } else {
       int32_t ret = vnodeCloseVnode(vnode);
@@ -269,6 +269,7 @@ int vnodeRemoveVnode(int vnode) {
         return ret;
       }
 
+      dTrace("vid:%d, status:%s, do delete operation", vnode, taosGetVnodeStatusStr(pVnode->vnodeStatus));
       vnodeRemoveDataFiles(vnode);
     }