refact(cluster): modify node management code

source/dnode/mgmt/bm/bmHandle.c

@@ -58,7 +58,7 @@ int32_t bmProcessCreateReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     dError("failed to create bnode since %s, input:%d cur:%d", terrstr(), createReq.dnodeId, pDnode->dnodeId);
     return -1;
   } else {
-    return bmOpen(pWrapper);
+    return dndOpenNode(pWrapper);
   }
 }
 
@@ -77,6 +77,7 @@ int32_t bmProcessDropReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     dError("failed to drop bnode since %s", terrstr());
     return -1;
   } else {
+    // dndCloseNode(pWrapper);
     return bmDrop(pWrapper);
   }
 }

source/dnode/mgmt/inc/dndInt.h

@@ -126,6 +126,7 @@ typedef struct SDnode {
   int32_t      numOfDisks;
   uint16_t     serverPort;
   bool         dropped;
+  EProcType    procType;
   EDndType     ntype;
   EDndStatus   status;
   EDndEvent    event;

source/dnode/mgmt/main/dndExec.c

@@ -27,7 +27,81 @@ static bool dndRequireNode(SMgmtWrapper *pWrapper) {
   return required;
 }
 
-int32_t dndOpenNode(SMgmtWrapper *pWrapper) {
+static int32_t dndInitNodeProc(SMgmtWrapper *pWrapper) {
+  int32_t shmsize = tsMnodeShmSize;
+  if (pWrapper->ntype == VNODES) {
+    shmsize = tsVnodeShmSize;
+  } else if (pWrapper->ntype == QNODE) {
+    shmsize = tsQnodeShmSize;
+  } else if (pWrapper->ntype == SNODE) {
+    shmsize = tsSnodeShmSize;
+  } else if (pWrapper->ntype == MNODE) {
+    shmsize = tsMnodeShmSize;
+  } else if (pWrapper->ntype == BNODE) {
+    shmsize = tsBnodeShmSize;
+  } else {
+    return -1;
+  }
+
+  if (taosCreateShm(&pWrapper->shm, pWrapper->ntype, shmsize) != 0) {
+    terrno = TAOS_SYSTEM_ERROR(terrno);
+    dError("node:%s, failed to create shm size:%d since %s", pWrapper->name, shmsize, terrstr());
+    return -1;
+  }
+  dInfo("node:%s, shm:%d is created, size:%d", pWrapper->name, pWrapper->shm.id, shmsize);
+
+  SProcCfg cfg = dndGenProcCfg(pWrapper);
+  cfg.isChild = false;
+  pWrapper->procType = PROC_PARENT;
+  pWrapper->pProc = taosProcInit(&cfg);
+  if (pWrapper->pProc == NULL) {
+    dError("node:%s, failed to create proc since %s", pWrapper->name, terrstr());
+    return -1;
+  }
+
+  return 0;
+}
+
+static int32_t dndNewNodeProc(SMgmtWrapper *pWrapper, EDndType n) {
+  char  tstr[8] = {0};
+  char *args[6] = {0};
+  snprintf(tstr, sizeof(tstr), "%d", n);
+  args[1] = "-c";
+  args[2] = configDir;
+  args[3] = "-n";
+  args[4] = tstr;
+  args[5] = NULL;
+
+  int32_t pid = taosNewProc(args);
+  if (pid <= 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    dError("node:%s, failed to exec in new process since %s", pWrapper->name, terrstr());
+    return -1;
+  }
+
+  pWrapper->procId = pid;
+  dInfo("node:%s, continue running in new process:%d", pWrapper->name, pid);
+  return 0;
+}
+
+static int32_t dndRunNodeProc(SMgmtWrapper *pWrapper) {
+  if (pWrapper->pDnode->ntype == NODE_MAX) {
+    dInfo("node:%s, should be started manually", pWrapper->name);
+  } else {
+    if (dndNewNodeProc(pWrapper, pWrapper->ntype) != 0) {
+      return -1;
+    }
+  }
+
+  if (taosProcRun(pWrapper->pProc) != 0) {
+    dError("node:%s, failed to run proc since %s", pWrapper->name, terrstr());
+    return -1;
+  }
+
+  return 0;
+}
+
+static int32_t dndOpenNodeImp(SMgmtWrapper *pWrapper) {
   if (taosMkDir(pWrapper->path) != 0) {
     terrno = TAOS_SYSTEM_ERROR(errno);
     dError("node:%s, failed to create dir:%s since %s", pWrapper->name, pWrapper->path, terrstr());
@@ -44,7 +118,19 @@ int32_t dndOpenNode(SMgmtWrapper *pWrapper) {
   return 0;
 }
 
-void dndCloseNode(SMgmtWrapper *pWrapper) {
+int32_t dndOpenNode(SMgmtWrapper *pWrapper) {
+  SDnode *pDnode = pWrapper->pDnode;
+  if (pDnode->procType == PROC_SINGLE) {
+    return dndOpenNodeImp(pWrapper);
+  } else if (pDnode->procType == PROC_PARENT) {
+    if (dndInitNodeProc(pWrapper) != 0) return -1;
+    if (dndWriteShmFile(pDnode) != 0) return -1;
+    if (dndRunNodeProc(pWrapper) != 0) return -1;
+  }
+  return 0;
+}
+
+static void dndCloseNodeImp(SMgmtWrapper *pWrapper) {
   dDebug("node:%s, mgmt start to close", pWrapper->name);
   pWrapper->required = false;
   taosWLockLatch(&pWrapper->latch);
@@ -65,27 +151,17 @@ void dndCloseNode(SMgmtWrapper *pWrapper) {
   dDebug("node:%s, mgmt has been closed", pWrapper->name);
 }
 
-
-static int32_t dndNewProc(SMgmtWrapper *pWrapper, EDndType n) {
-  char  tstr[8] = {0};
-  char *args[6] = {0};
-  snprintf(tstr, sizeof(tstr), "%d", n);
-  args[1] = "-c";
-  args[2] = configDir;
-  args[3] = "-n";
-  args[4] = tstr;
-  args[5] = NULL;
-
-  int32_t pid = taosNewProc(args);
-  if (pid <= 0) {
-    terrno = TAOS_SYSTEM_ERROR(errno);
-    dError("node:%s, failed to exec in new process since %s", pWrapper->name, terrstr());
-    return -1;
+void dndCloseNode(SMgmtWrapper *pWrapper) {
+  if (pWrapper->pDnode->procType == PROC_PARENT) {
+    if (pWrapper->procId > 0 && taosProcExist(pWrapper->procId)) {
+      dInfo("node:%s, send kill signal to the child process:%d", pWrapper->name, pWrapper->procId);
+      taosKillProc(pWrapper->procId);
+      dInfo("node:%s, wait for child process:%d to stop", pWrapper->name, pWrapper->procId);
+      taosWaitProc(pWrapper->procId);
+      dInfo("node:%s, child process:%d is stopped", pWrapper->name, pWrapper->procId);
     }
-
-  pWrapper->procId = pid;
-  dInfo("node:%s, continue running in new process:%d", pWrapper->name, pid);
-  return 0;
+  }
+  dndCloseNodeImp(pWrapper);
 }
 
 static void dndProcessProcHandle(void *handle) {
@@ -96,13 +172,14 @@ static void dndProcessProcHandle(void *handle) {
 
 static int32_t dndRunInSingleProcess(SDnode *pDnode) {
   dInfo("dnode run in single process");
+  pDnode->procType = PROC_SINGLE;
 
   for (EDndType n = DNODE; n < NODE_MAX; ++n) {
     SMgmtWrapper *pWrapper = &pDnode->wrappers[n];
     pWrapper->required = dndRequireNode(pWrapper);
     if (!pWrapper->required) continue;
 
-    if (dndOpenNode(pWrapper) != 0) {
+    if (dndOpenNodeImp(pWrapper) != 0) {
       dError("node:%s, failed to start since %s", pWrapper->name, terrstr());
       return -1;
     }
@@ -136,8 +213,10 @@ static int32_t dndRunInSingleProcess(SDnode *pDnode) {
 
 static int32_t dndRunInParentProcess(SDnode *pDnode) {
   dInfo("dnode run in parent process");
+  pDnode->procType = PROC_PARENT;
+
   SMgmtWrapper *pDWrapper = &pDnode->wrappers[DNODE];
-  if (dndOpenNode(pDWrapper) != 0) {
+  if (dndOpenNodeImp(pDWrapper) != 0) {
     dError("node:%s, failed to start since %s", pDWrapper->name, terrstr());
     return -1;
   }
@@ -146,36 +225,7 @@ static int32_t dndRunInParentProcess(SDnode *pDnode) {
     SMgmtWrapper *pWrapper = &pDnode->wrappers[n];
     pWrapper->required = dndRequireNode(pWrapper);
     if (!pWrapper->required) continue;
-
-    int32_t shmsize = tsMnodeShmSize;
-    if (n == VNODES) {
-      shmsize = tsVnodeShmSize;
-    } else if (n == QNODE) {
-      shmsize = tsQnodeShmSize;
-    } else if (n == SNODE) {
-      shmsize = tsSnodeShmSize;
-    } else if (n == MNODE) {
-      shmsize = tsMnodeShmSize;
-    } else if (n == BNODE) {
-      shmsize = tsBnodeShmSize;
-    } else {
-    }
-
-    if (taosCreateShm(&pWrapper->shm, n, shmsize) != 0) {
-      terrno = TAOS_SYSTEM_ERROR(terrno);
-      dError("node:%s, failed to create shm size:%d since %s", pWrapper->name, shmsize, terrstr());
-      return -1;
-    }
-    dInfo("node:%s, shm:%d is created, size:%d", pWrapper->name, pWrapper->shm.id, shmsize);
-
-    SProcCfg cfg = dndGenProcCfg(pWrapper);
-    cfg.isChild = false;
-    pWrapper->procType = PROC_PARENT;
-    pWrapper->pProc = taosProcInit(&cfg);
-    if (pWrapper->pProc == NULL) {
-      dError("node:%s, failed to create proc since %s", pWrapper->name, terrstr());
-      return -1;
-    }
+    if (dndInitNodeProc(pWrapper) != 0) return -1;
   }
 
   if (dndWriteShmFile(pDnode) != 0) {
@@ -186,19 +236,7 @@ static int32_t dndRunInParentProcess(SDnode *pDnode) {
   for (EDndType n = DNODE + 1; n < NODE_MAX; ++n) {
     SMgmtWrapper *pWrapper = &pDnode->wrappers[n];
     if (!pWrapper->required) continue;
-
-    if (pDnode->ntype == NODE_MAX) {
-      dInfo("node:%s, should be started manually", pWrapper->name);
-    } else {
-      if (dndNewProc(pWrapper, n) != 0) {
-        return -1;
-      }
-    }
-
-    if (taosProcRun(pWrapper->pProc) != 0) {
-      dError("node:%s, failed to run proc since %s", pWrapper->name, terrstr());
-      return -1;
-    }
+    if (dndRunNodeProc(pWrapper) != 0) return -1;
   }
 
   dndSetStatus(pDnode, DND_STAT_RUNNING);
@@ -239,7 +277,7 @@ static int32_t dndRunInParentProcess(SDnode *pDnode) {
         if (pWrapper->procId <= 0 || !taosProcExist(pWrapper->procId)) {
           dWarn("node:%s, process:%d is killed and needs to be restarted", pWrapper->name, pWrapper->procId);
           taosProcCloseHandles(pWrapper->pProc, dndProcessProcHandle);
-          dndNewProc(pWrapper, n);
+          dndNewNodeProc(pWrapper, n);
         }
       }
     }
@@ -253,6 +291,7 @@ static int32_t dndRunInParentProcess(SDnode *pDnode) {
 static int32_t dndRunInChildProcess(SDnode *pDnode) {
   SMgmtWrapper *pWrapper = &pDnode->wrappers[pDnode->ntype];
   dInfo("%s run in child process", pWrapper->name);
+  pDnode->procType = PROC_CHILD;
 
   pWrapper->required = dndRequireNode(pWrapper);
   if (!pWrapper->required) {
@@ -264,7 +303,7 @@ static int32_t dndRunInChildProcess(SDnode *pDnode) {
   tmsgSetDefaultMsgCb(&msgCb);
   pWrapper->procType = PROC_CHILD;
 
-  if (dndOpenNode(pWrapper) != 0) {
+  if (dndOpenNodeImp(pWrapper) != 0) {
     dError("node:%s, failed to start since %s", pWrapper->name, terrstr());
     return -1;
   }

source/dnode/mgmt/mm/mmHandle.c

@@ -80,6 +80,7 @@ int32_t mmProcessDropReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     dError("failed to drop mnode since %s", terrstr());
     return -1;
   } else {
+    // dndCloseNode(pWrapper);
     return mmDrop(pWrapper);
   }
 }

source/dnode/mgmt/qm/qmHandle.c

@@ -58,7 +58,7 @@ int32_t qmProcessCreateReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     dError("failed to create qnode since %s", terrstr());
     return -1;
   } else {
-    return qmOpen(pWrapper);
+    return dndOpenNode(pWrapper);
   }
 }
 
@@ -77,6 +77,7 @@ int32_t qmProcessDropReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     dError("failed to drop qnode since %s", terrstr());
     return -1;
   } else {
+    // dndCloseNode(pWrapper);
     return qmDrop(pWrapper);
   }
 }

source/dnode/mgmt/sm/smHandle.c

@@ -58,7 +58,7 @@ int32_t smProcessCreateReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     dError("failed to create snode since %s", terrstr());
     return -1;
   } else {
-    return smOpen(pWrapper);
+    return dndOpenNode(pWrapper);
   }
 }
 
@@ -78,6 +78,7 @@ int32_t smProcessDropReq(SMgmtWrapper *pWrapper, SNodeMsg *pMsg) {
     return -1;
   } else {
     return smDrop(pWrapper);
+    // return dndCloseNode(pWrapper);
   }
 }