BUG#13985 Cluster: ndb_mgm "status" command can return incorrect data node status

partial fix for this bug. more info on what the other half of the fix involves
is in the bug report.


ndb/src/mgmclient/CommandInterpreter.cpp:
  partial fix for bug13985
  
  hold a mutex around printing out events
  hold the mutex also around printing out put of 'status' commands.
  
  this means we don't get
  1 started
  2 started
  2 starting
  
  output.
  
  we'll instead get the event before/after the entire status output.
  Due to the nature of the event arriving before status being updated,
  we'll pretty much always see started AFTER starting.

1 files changed, 24 insertions, 4 deletions

diff --git a/ndb/src/mgmclient/CommandInterpreter.cpp b/ndb/src/mgmclient/CommandInterpreter.cpp
index 14cd3fba42b..cb85cd8f3e6 100644
--- a/ndb/src/mgmclient/CommandInterpreter.cpp
+++ b/ndb/src/mgmclient/CommandInterpreter.cpp
@@ -172,8 +172,15 @@ private:
   bool rep_connected;
 #endif
   struct NdbThread* m_event_thread;
+  NdbMutex *m_print_mutex;
 };
 
+struct event_thread_param {
+  NdbMgmHandle *m;
+  NdbMutex **p;
+};
+
+NdbMutex* print_mutex;
 
 /*
  * Facade object for CommandInterpreter
@@ -409,6 +416,7 @@ CommandInterpreter::CommandInterpreter(const char *_host,int verbose)
   m_connected= false;
   m_event_thread= 0;
   try_reconnect = 0;
+  m_print_mutex= NdbMutex_Create();
 #ifdef HAVE_GLOBAL_REPLICATION
   rep_host = NULL;
   m_repserver = NULL;
@@ -422,6 +430,7 @@ CommandInterpreter::CommandInterpreter(const char *_host,int verbose)
 CommandInterpreter::~CommandInterpreter() 
 {
   disconnect();
+  NdbMutex_Destroy(m_print_mutex);
   ndb_mgm_destroy_handle(&m_mgmsrv);
   ndb_mgm_destroy_handle(&m_mgmsrv2);
 }
@@ -461,11 +470,13 @@ CommandInterpreter::printError()
 
 static int do_event_thread;
 static void*
-event_thread_run(void* m)
+event_thread_run(void* p)
 {
   DBUG_ENTER("event_thread_run");
 
-  NdbMgmHandle handle= *(NdbMgmHandle*)m;
+  struct event_thread_param param= *(struct event_thread_param*)p;
+  NdbMgmHandle handle= *(param.m);
+  NdbMutex* printmutex= *(param.p);
 
   int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_BACKUP,
 		   1, NDB_MGM_EVENT_CATEGORY_STARTUP,
@@ -483,7 +494,11 @@ event_thread_run(void* m)
       {
 	const char ping_token[]= "<PING>";
 	if (memcmp(ping_token,tmp,sizeof(ping_token)-1))
-	  ndbout << tmp;
+	  if(tmp && strlen(tmp))
+          {
+            Guard g(printmutex);
+            ndbout << tmp;
+          }
       }
     } while(do_event_thread);
     NDB_CLOSE_SOCKET(fd);
@@ -516,8 +531,11 @@ CommandInterpreter::connect()
 	assert(m_event_thread == 0);
 	assert(do_event_thread == 0);
 	do_event_thread= 0;
+        struct event_thread_param p;
+        p.m= &m_mgmsrv2;
+        p.p= &m_print_mutex;
 	m_event_thread = NdbThread_Create(event_thread_run,
-					  (void**)&m_mgmsrv2,
+					  (void**)&p,
 					  32768,
 					  "CommandInterpreted_event_thread",
 					  NDB_THREAD_PRIO_LOW);
@@ -607,6 +625,7 @@ CommandInterpreter::execute(const char *_line, int _try_reconnect,
   int result= execute_impl(_line);
   if (error)
     *error= m_error;
+
   return result;
 }
 
@@ -920,6 +939,7 @@ CommandInterpreter::executeForAll(const char * cmd, ExecuteFunction fun,
     ndbout_c("Trying to start all nodes of system.");
     ndbout_c("Use ALL STATUS to see the system start-up phases.");
   } else {
+    Guard g(m_print_mutex);
     struct ndb_mgm_cluster_state *cl= ndb_mgm_get_status(m_mgmsrv);
     if(cl == 0){
       ndbout_c("Unable get status from management server");