dded method to retrieve mysqld, ndbd, and ndb_mgmd aliases for API,MGM and DB

added sanity checks for config file
changed so that ndb_0 is replaced by ndb_pid#### in ndb files
added error_string for to some functions, alloc_node_id(), start()
added better error printout for failed alloc node id
better error printouts for failures when starting ndb_mgmd

ndb/include/mgmapi/mgmapi.h

@@ -394,6 +394,14 @@ extern "C" {
    */
   const char * ndb_mgm_get_node_type_string(enum ndb_mgm_node_type type);
 
+  /**
+   * Convert an ndb_mgm_node_type to a alias string
+   *
+   * @param   type          Node type.
+   * @return                NULL if invalid id.
+   */
+  const char * ndb_mgm_get_node_type_alias_string(enum ndb_mgm_node_type type, const char **str);
+
   /**
    * Convert a string to a ndb_mgm_node_status
    *

ndb/src/common/mgmcommon/ConfigInfo.cpp

@@ -161,6 +161,9 @@ const int ConfigInfo::m_NoOfRules = sizeof(m_SectionRules)/sizeof(SectionRule);
 /****************************************************************************
  * Config Rules declarations
  ****************************************************************************/
+static bool sanity_checks(Vector<ConfigInfo::ConfigRuleSection>&sections, 
+			  struct InitConfigFileParser::Context &ctx, 
+			  const char * rule_data);
 static bool add_node_connections(Vector<ConfigInfo::ConfigRuleSection>&sections, 
 				 struct InitConfigFileParser::Context &ctx, 
 				 const char * rule_data);
@@ -173,6 +176,7 @@ static bool check_node_vs_replicas(Vector<ConfigInfo::ConfigRuleSection>&section
 
 const ConfigInfo::ConfigRule 
 ConfigInfo::m_ConfigRules[] = {
+  { sanity_checks, 0 },
   { add_node_connections, 0 },
   { add_server_ports, 0 },
   { check_node_vs_replicas, 0 },
@@ -2958,6 +2962,29 @@ saveInConfigValues(InitConfigFileParser::Context & ctx, const char * data){
   return true;
 }
 
+static bool
+sanity_checks(Vector<ConfigInfo::ConfigRuleSection>&sections, 
+	      struct InitConfigFileParser::Context &ctx, 
+	      const char * rule_data)
+{
+  Uint32 db_nodes = 0;
+  Uint32 mgm_nodes = 0;
+  Uint32 api_nodes = 0;
+  if (!ctx.m_userProperties.get("DB", &db_nodes)) {
+    ctx.reportError("At least one database node should be defined in config file");
+    return false;
+  }
+  if (!ctx.m_userProperties.get("MGM", &mgm_nodes)) {
+    ctx.reportError("At least one management server node should be defined in config file");
+    return false;
+  }
+  if (!ctx.m_userProperties.get("API", &api_nodes)) {
+    ctx.reportError("At least one application node (for the mysqld) should be defined in config file");
+    return false;
+  }
+  return true;
+}
+
 static bool
 add_node_connections(Vector<ConfigInfo::ConfigRuleSection>&sections, 
 		   struct InitConfigFileParser::Context &ctx, 
@@ -3091,7 +3118,9 @@ check_node_vs_replicas(Vector<ConfigInfo::ConfigRuleSection>&sections,
 {
   Uint32 db_nodes = 0;
   Uint32 replicas = 0;
-  ctx.m_userProperties.get("DB", &db_nodes);
+  if (!ctx.m_userProperties.get("DB", &db_nodes)) {
+    return true;
+  }
   ctx.m_userProperties.get("NoOfReplicas", &replicas);
   if((db_nodes % replicas) != 0){
     ctx.reportError("Invalid no of db nodes wrt no of replicas.\n"

ndb/src/common/mgmcommon/NdbConfig.c

@@ -53,58 +53,73 @@ NdbConfig_NdbCfgName(int with_ndb_home){
   return buf;
 }
 
+static
+char *get_prefix_buf(int len, int node_id)
+{
+  char tmp_buf[sizeof("ndb_pid#########")+1];
+  if (node_id > 0)
+    snprintf(tmp_buf, sizeof(tmp_buf), "ndb_%u", node_id);
+  else
+    snprintf(tmp_buf, sizeof(tmp_buf), "ndb_pid%u", getpid());
+  tmp_buf[sizeof(tmp_buf)-1]= 0;
+
+  char *buf= NdbConfig_AllocHomePath(len+strlen(tmp_buf));
+  strcat(buf, tmp_buf);
+  return buf;
+}
+
 char* 
 NdbConfig_ErrorFileName(int node_id){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u_error.log", node_id);
+  snprintf(buf+len, 128, "_error.log");
   return buf;
 }
 
 char*
 NdbConfig_ClusterLogFileName(int node_id){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u_cluster.log", node_id);
+  snprintf(buf+len, 128, "_cluster.log");
   return buf;
 }
 
 char*
 NdbConfig_SignalLogFileName(int node_id){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u_signal.log", node_id);
+  snprintf(buf+len, 128, "_signal.log");
   return buf;
 }
 
 char*
 NdbConfig_TraceFileName(int node_id, int file_no){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u_trace.log.%u", node_id, file_no);
+  snprintf(buf+len, 128, "_trace.log.%u", file_no);
   return buf;
 }
 
 char*
 NdbConfig_NextTraceFileName(int node_id){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u_trace.log.next", node_id);
+  snprintf(buf+len, 128, "_trace.log.next");
   return buf;
 }
 
 char*
 NdbConfig_PidFileName(int node_id){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u.pid", node_id);
+  snprintf(buf+len, 128, ".pid");
   return buf;
 }
 
 char*
 NdbConfig_StdoutFileName(int node_id){
-  char *buf= NdbConfig_AllocHomePath(128);
+  char *buf= get_prefix_buf(128, node_id);
   int len= strlen(buf);
-  snprintf(buf+len, 128, "ndb_%u_out.log", node_id);
+  snprintf(buf+len, 128, "_out.log");
   return buf;
 }

ndb/src/kernel/error/ErrorReporter.cpp

@@ -139,7 +139,7 @@ ErrorReporter::formatMessage(ErrorCategory type,
 	   objRef, 
 	   programName, 
 	   processId, 
-	   theNameOfTheTraceFile);
+	   theNameOfTheTraceFile ? theNameOfTheTraceFile : "<no tracefile>");
 
   // Add trailing blanks to get a fixed lenght of the message
   while (strlen(messptr) <= MESSAGE_LENGTH-3){
@@ -217,8 +217,10 @@ WriteMessage(ErrorCategory thrdType, int thrdMessageID,
   /**
    * Format trace file name
    */
-  int file_no= ErrorReporter::get_trace_no();
-  char *theTraceFileName= NdbConfig_TraceFileName(globalData.ownId, file_no);
+  char *theTraceFileName= 0;
+  if (globalData.ownId > 0)
+    theTraceFileName= NdbConfig_TraceFileName(globalData.ownId,
+					      ErrorReporter::get_trace_no());
   NdbAutoPtr<char> tmp_aptr1(theTraceFileName);
   
   // The first 69 bytes is info about the current offset
@@ -291,26 +293,28 @@ WriteMessage(ErrorCategory thrdType, int thrdMessageID,
   fflush(stream);
   fclose(stream);
   
-  // Open the tracefile...
-  FILE *jamStream = fopen(theTraceFileName, "w");
+  if (theTraceFileName) {
+    // Open the tracefile...
+    FILE *jamStream = fopen(theTraceFileName, "w");
   
-  //  ...and "dump the jam" there.
-  //  ErrorReporter::dumpJam(jamStream);
-  if(thrdTheEmulatedJam != 0){
-    dumpJam(jamStream, thrdTheEmulatedJamIndex, thrdTheEmulatedJam);
-  }
-  
-  /* Dont print the jobBuffers until a way to copy them, 
-     like the other variables,
-     is implemented. Otherwise when NDB keeps running, 
-     with this function running
-     in the background, the jobBuffers will change during runtime. And when
-     they're printed here, they will not be correct anymore.
-  */
-  globalScheduler.dumpSignalMemory(jamStream);
+    //  ...and "dump the jam" there.
+    //  ErrorReporter::dumpJam(jamStream);
+    if(thrdTheEmulatedJam != 0){
+      dumpJam(jamStream, thrdTheEmulatedJamIndex, thrdTheEmulatedJam);
+    }
   
-  fclose(jamStream);
+    /* Dont print the jobBuffers until a way to copy them, 
+       like the other variables,
+       is implemented. Otherwise when NDB keeps running, 
+       with this function running
+       in the background, the jobBuffers will change during runtime. And when
+       they're printed here, they will not be correct anymore.
+    */
+    globalScheduler.dumpSignalMemory(jamStream);
   
+    fclose(jamStream);
+  }
+
   return 0;
 }
 

ndb/src/mgmapi/mgmapi.cpp

@@ -403,14 +403,15 @@ ndb_mgm_disconnect(NdbMgmHandle handle)
 struct ndb_mgm_type_atoi 
 {
   const char * str;
+  const char * alias;
   enum ndb_mgm_node_type value;
 };
 
 static struct ndb_mgm_type_atoi type_values[] = 
 {
-  { "NDB", NDB_MGM_NODE_TYPE_NDB},
-  { "API", NDB_MGM_NODE_TYPE_API },
-  { "MGM", NDB_MGM_NODE_TYPE_MGM }
+  { "NDB", "ndbd", NDB_MGM_NODE_TYPE_NDB},
+  { "API", "mysqld", NDB_MGM_NODE_TYPE_API },
+  { "MGM", "ndb_mgmd", NDB_MGM_NODE_TYPE_MGM }
 };
 
 const int no_of_type_values = (sizeof(type_values) / 
@@ -440,6 +441,20 @@ ndb_mgm_get_node_type_string(enum ndb_mgm_node_type type)
   return 0;
 }
 
+extern "C"
+const char * 
+ndb_mgm_get_node_type_alias_string(enum ndb_mgm_node_type type, const char** str)
+{
+  for(int i = 0; i<no_of_type_values; i++)
+    if(type_values[i].value == type)
+      {
+	if (str)
+	  *str= type_values[i].str;
+	return type_values[i].alias;
+      }
+  return 0;
+}
+
 struct ndb_mgm_status_atoi {
   const char * str;
   enum ndb_mgm_node_status value;

ndb/src/mgmsrv/MgmtSrvr.cpp

@@ -584,8 +584,10 @@ MgmtSrvr::MgmtSrvr(NodeId nodeId,
 
   _ownNodeId= 0;
   NodeId tmp= nodeId;
-  if (!alloc_node_id(&tmp, NDB_MGM_NODE_TYPE_MGM, 0, 0)){
-    ndbout << "Unable to obtain requested nodeid " << nodeId;
+  BaseString error_string;
+  if (!alloc_node_id(&tmp, NDB_MGM_NODE_TYPE_MGM, 0, 0, error_string)){
+    ndbout << "Unable to obtain requested nodeid: "
+	   << error_string.c_str() << endl;
     exit(-1);
   }
   _ownNodeId = tmp;
@@ -606,17 +608,20 @@ MgmtSrvr::check_start()
 }
 
 bool 
-MgmtSrvr::start()
+MgmtSrvr::start(BaseString &error_string)
 {
   if (_props == NULL) {
-    if (!check_start())
+    if (!check_start()) {
+      error_string.append("MgmtSrvr.cpp: check_start() failed.");
       return false;
+    }
   }
   theFacade = TransporterFacade::start_instance
     (_ownNodeId,(ndb_mgm_configuration*)_config->m_configValues);
   
   if(theFacade == 0) {
     DEBUG("MgmtSrvr.cpp: theFacade is NULL.");
+    error_string.append("MgmtSrvr.cpp: theFacade is NULL.");
     return false;
   }  
 
@@ -630,6 +635,7 @@ MgmtSrvr::start()
   
   if(_blockNumber == -1){
     DEBUG("MgmtSrvr.cpp: _blockNumber is -1.");
+    error_string.append("MgmtSrvr.cpp: _blockNumber is -1.");
     theFacade->stop_instance();
     theFacade = 0;
     return false;
@@ -2304,7 +2310,8 @@ bool
 MgmtSrvr::alloc_node_id(NodeId * nodeId, 
 			enum ndb_mgm_node_type type,
 			struct sockaddr *client_addr, 
-			SOCKET_SIZE_TYPE *client_addr_len)
+			SOCKET_SIZE_TYPE *client_addr_len,
+			BaseString &error_string)
 {
   Guard g(&f_node_id_mutex);
 #if 0
@@ -2322,19 +2329,26 @@ MgmtSrvr::alloc_node_id(NodeId * nodeId,
       }
   }
 
+  bool found_matching_id= false;
+  bool found_matching_type= false;
+  bool found_free_node= false;
+  unsigned type_c= 0;
+
   ndb_mgm_configuration_iterator iter(*(ndb_mgm_configuration *)_config->m_configValues,
 				      CFG_SECTION_NODE);
   for(iter.first(); iter.valid(); iter.next()) {
     unsigned tmp= 0;
     if(iter.get(CFG_NODE_ID, &tmp)) abort();
-    if (connected_nodes.get(tmp))
-      continue;
     if (*nodeId && *nodeId != tmp)
       continue;
-    unsigned type_c;
+    found_matching_id= true;
     if(iter.get(CFG_TYPE_OF_SECTION, &type_c)) abort();
     if(type_c != type)
       continue;
+    found_matching_type= true;
+    if (connected_nodes.get(tmp))
+      continue;
+    found_free_node= true;
     const char *config_hostname = 0;
     if(iter.get(CFG_NODE_HOST, &config_hostname)) abort();
 
@@ -2374,6 +2388,43 @@ MgmtSrvr::alloc_node_id(NodeId * nodeId,
 #endif
     return true;
   }
+
+  BaseString type_string, type_c_string;
+  {
+    const char *alias, *str;
+    alias= ndb_mgm_get_node_type_alias_string(type, &str);
+    type_string.assfmt("%s(%s)", alias, str);
+    alias= ndb_mgm_get_node_type_alias_string((enum ndb_mgm_node_type)type_c, &str);
+    type_c_string.assfmt("%s(%s)", alias, str);
+  }
+
+  if (*nodeId == 0) {
+    if (found_matching_id)
+      if (found_matching_type)
+	if (found_free_node)
+	  error_string.appfmt("Connection done from wrong host %s.",
+			      inet_ntoa(((struct sockaddr_in *)(client_addr))->sin_addr));
+	else
+	  error_string.appfmt("No free node id found for %s.", type_string.c_str());
+      else
+	error_string.appfmt("No %s node defined in config file.", type_string.c_str());
+    else
+      error_string.append("No nodes defined in config file.");
+  } else {
+    if (found_matching_id)
+      if (found_matching_type)
+	if (found_free_node)
+	  error_string.appfmt("Connection with id %d done from wrong host %s, expected host XX.",
+			      *nodeId, inet_ntoa(((struct sockaddr_in *)(client_addr))->sin_addr));
+	else
+	  error_string.appfmt("Id %d already allocated by another node.", *nodeId);
+      else
+	error_string.appfmt("Id %d configured as %s, connect attempted as %s.",
+			    *nodeId, type_c_string.c_str(), type_string.c_str());
+    else
+      error_string.appfmt("No node defined with id=%d in config file.", *nodeId);
+  }
+
   return false;
 }
 

ndb/src/mgmsrv/MgmtSrvr.hpp

@@ -179,7 +179,7 @@ public:
    *   @return true if succeeded, otherwise false
    */
   bool check_start(); // may be run before start to check that some things are ok
-  bool start();
+  bool start(BaseString &error_string);
 
   ~MgmtSrvr();
 
@@ -467,7 +467,8 @@ public:
    */
   bool getNextNodeId(NodeId * _nodeId, enum ndb_mgm_node_type type) const ;
   bool alloc_node_id(NodeId * _nodeId, enum ndb_mgm_node_type type,
-		     struct sockaddr *client_addr, SOCKET_SIZE_TYPE *client_addr_len);
+		     struct sockaddr *client_addr, SOCKET_SIZE_TYPE *client_addr_len,
+		     BaseString &error_string);
   
   /**
    *

ndb/src/mgmsrv/Services.cpp

@@ -413,11 +413,14 @@ MgmApiSession::get_nodeid(Parser_t::Context &,
 
   NodeId tmp= nodeid;
   if(tmp == 0 || !m_allocated_resources->is_reserved(tmp)){
+    BaseString error_string;
     if (!m_mgmsrv.alloc_node_id(&tmp, (enum ndb_mgm_node_type)nodetype, 
-				&addr, &addrlen)){
+				&addr, &addrlen, error_string)){
+      const char *alias;
+      const char *str;
+      alias= ndb_mgm_get_node_type_alias_string((enum ndb_mgm_node_type)nodetype, &str);
       m_output->println(cmd);
-      m_output->println("result: no free nodeid %d for nodetype %d",
-			nodeid, nodetype);
+      m_output->println("result: %s", error_string.c_str());
       m_output->println("");
       return;
     }

ndb/src/mgmsrv/main.cpp

@@ -183,7 +183,7 @@ NDB_MAIN(mgmsrv){
     if(!glob.socketServer->tryBind(glob.port, glob.interface_name)){
       ndbout_c("Unable to setup port: %s:%d!\n"
 	       "Please check if the port is already used,\n"
-	       "(perhaps a mgmtsrvr is already running),\n"
+	       "(perhaps a ndb_mgmd is already running),\n"
 	       "and if you are executing on the correct computer", 
 	       (glob.interface_name ? glob.interface_name : "*"), glob.port);
       goto error_end;
@@ -195,7 +195,7 @@ NDB_MAIN(mgmsrv){
   if(!glob.socketServer->setup(mapi, glob.port, glob.interface_name)){
     ndbout_c("Unable to setup management port: %d!\n"
 	     "Please check if the port is already used,\n"
-	     "(perhaps a mgmtsrvr is already running),\n"
+	     "(perhaps a ndb_mgmd is already running),\n"
 	     "and if you are executing on the correct computer", 
 	     glob.port);
     delete mapi;
@@ -228,10 +228,14 @@ NDB_MAIN(mgmsrv){
   }
 
   signal(SIGPIPE, SIG_IGN);
-  if(!glob.mgmObject->start()){
-    ndbout_c("Unable to start management server.");
-    ndbout_c("Probably caused by illegal initial configuration file.");
-    goto error_end;
+  {
+    BaseString error_string;
+    if(!glob.mgmObject->start(error_string)){
+      ndbout_c("Unable to start management server.");
+      ndbout_c("Probably caused by illegal initial configuration file.");
+      ndbout_c(error_string.c_str());
+      goto error_end;
+    }
   }
 
   //glob.mgmObject->saveConfig();