Bug #19395 mysqld does not always detect cluster shutdown

sql/ha_ndbcluster.cc

@@ -182,6 +182,8 @@ static const char * ndb_connected_host= 0;
 static long ndb_connected_port= 0;
 static long ndb_number_of_replicas= 0;
 long ndb_number_of_storage_nodes= 0;
+long ndb_number_of_ready_storage_nodes= 0;
+long ndb_connect_count= 0;
 
 static int update_status_variables(Ndb_cluster_connection *c)
 {
@@ -190,6 +192,8 @@ static int update_status_variables(Ndb_cluster_connection *c)
   ndb_connected_host=          c->get_connected_host();
   ndb_number_of_replicas=      0;
   ndb_number_of_storage_nodes= c->no_db_nodes();
+  ndb_number_of_ready_storage_nodes= c->get_no_ready();
+  ndb_connect_count= c->get_connect_count();
   return 0;
 }
 
@@ -9361,11 +9365,15 @@ ndbcluster_show_status(THD* thd, stat_print_fn *stat_print,
                 "cluster_node_id=%u, "
                 "connected_host=%s, "
                 "connected_port=%u, "
-                "number_of_storage_nodes=%u",
+                "number_of_storage_nodes=%u, "
+                "number_of_ready_storage_nodes=%u, "
+                "connect_count=%u",
                 ndb_cluster_node_id,
                 ndb_connected_host,
                 ndb_connected_port,
-                ndb_number_of_storage_nodes);
+                ndb_number_of_storage_nodes,
+                ndb_number_of_ready_storage_nodes,
+                ndb_connect_count);
   if (stat_print(thd, ndbcluster_hton.name, strlen(ndbcluster_hton.name),
                  "connection", strlen("connection"),
                  buf, buflen))

sql/ha_ndbcluster_binlog.cc

@@ -1776,7 +1776,8 @@ ndb_binlog_thread_handle_schema_event(THD *thd, Ndb *ndb,
       break;
     case NDBEVENT::TE_CLUSTER_FAILURE:
       if (ndb_extra_logging)
-        sql_print_information("NDB Binlog: cluster failure for %s.", schema_share->key);
+        sql_print_information("NDB Binlog: cluster failure for %s at epoch %u.",
+                              schema_share->key, (unsigned) pOp->getGCI());
       // fall through
     case NDBEVENT::TE_DROP:
       if (ndb_extra_logging &&
@@ -1785,7 +1786,6 @@ ndb_binlog_thread_handle_schema_event(THD *thd, Ndb *ndb,
                               "read only on reconnect.");
       free_share(&schema_share);
       schema_share= 0;
-      ndb_binlog_tables_inited= FALSE;
       close_cached_tables((THD*) 0, 0, (TABLE_LIST*) 0, FALSE);
       // fall through
     case NDBEVENT::TE_ALTER:
@@ -2829,7 +2829,8 @@ ndb_binlog_thread_handle_non_data_event(THD *thd, Ndb *ndb,
   {
   case NDBEVENT::TE_CLUSTER_FAILURE:
     if (ndb_extra_logging)
-      sql_print_information("NDB Binlog: cluster failure for %s.", share->key);
+      sql_print_information("NDB Binlog: cluster failure for %s at epoch %u.",
+                            share->key, (unsigned) pOp->getGCI());
     if (apply_status_share == share)
     {
       if (ndb_extra_logging &&
@@ -2838,7 +2839,6 @@ ndb_binlog_thread_handle_non_data_event(THD *thd, Ndb *ndb,
                               "read only on reconnect.");
       free_share(&apply_status_share);
       apply_status_share= 0;
-      ndb_binlog_tables_inited= FALSE;
     }
     DBUG_PRINT("info", ("CLUSTER FAILURE EVENT: "
                         "%s  received share: 0x%lx  op: %lx  share op: %lx  "
@@ -2854,7 +2854,6 @@ ndb_binlog_thread_handle_non_data_event(THD *thd, Ndb *ndb,
                               "read only on reconnect.");
       free_share(&apply_status_share);
       apply_status_share= 0;
-      ndb_binlog_tables_inited= FALSE;
     }
     /* ToDo: remove printout */
     if (ndb_extra_logging)
@@ -3267,24 +3266,6 @@ pthread_handler_t ndb_binlog_thread_func(void *arg)
   pthread_mutex_unlock(&injector_mutex);
   pthread_cond_signal(&injector_cond);
 
-  thd->proc_info= "Waiting for ndbcluster to start";
-
-  pthread_mutex_lock(&injector_mutex);
-  while (!schema_share ||
-         (ndb_binlog_running && !apply_status_share))
-  {
-    /* ndb not connected yet */
-    struct timespec abstime;
-    set_timespec(abstime, 1);
-    pthread_cond_timedwait(&injector_cond, &injector_mutex, &abstime);
-    if (abort_loop)
-    {
-      pthread_mutex_unlock(&injector_mutex);
-      goto err;
-    }
-  }
-  pthread_mutex_unlock(&injector_mutex);
-
   /*
     Main NDB Injector loop
   */
@@ -3298,15 +3279,28 @@ pthread_handler_t ndb_binlog_thread_func(void *arg)
   set_thd_ndb(thd, thd_ndb);
   thd_ndb->options|= TNO_NO_LOG_SCHEMA_OP;
   thd->query_id= 0; // to keep valgrind quiet
+
+restart:
   {
-    static char db[]= "";
-    thd->db= db;
-    if (ndb_binlog_running)
-      open_binlog_index(thd, &binlog_tables, &binlog_index);
-    thd->db= db;
+    thd->proc_info= "Waiting for ndbcluster to start";
+
+    pthread_mutex_lock(&injector_mutex);
+    while (!schema_share ||
+           (ndb_binlog_running && !apply_status_share))
+    {
+      /* ndb not connected yet */
+      struct timespec abstime;
+      set_timespec(abstime, 1);
+      pthread_cond_timedwait(&injector_cond, &injector_mutex, &abstime);
+      if (abort_loop)
+      {
+        pthread_mutex_unlock(&injector_mutex);
+        goto err;
+      }
+    }
+    pthread_mutex_unlock(&injector_mutex);
   }
 
-restart:
   {
     // wait for the first event
     thd->proc_info= "Waiting for first event from ndbcluster";
@@ -3337,7 +3331,13 @@ pthread_handler_t ndb_binlog_thread_func(void *arg)
       }
     }
   }
-
+  {
+    static char db[]= "";
+    thd->db= db;
+    if (ndb_binlog_running)
+      open_binlog_index(thd, &binlog_tables, &binlog_index);
+    thd->db= db;
+  }
   do_ndbcluster_binlog_close_connection= BCCC_running;
   for ( ; !((abort_loop || do_ndbcluster_binlog_close_connection) &&
             ndb_latest_handled_binlog_epoch >= g_latest_trans_gci) &&
@@ -3686,7 +3686,12 @@ pthread_handler_t ndb_binlog_thread_func(void *arg)
     ndb_latest_handled_binlog_epoch= ndb_latest_received_binlog_epoch;
   }
   if (do_ndbcluster_binlog_close_connection == BCCC_restart)
+  {
+    ndb_binlog_tables_inited= FALSE;
+    close_thread_tables(thd);
+    binlog_index= 0;
     goto restart;
+  }
 err:
   DBUG_PRINT("info",("Shutting down cluster binlog thread"));
   thd->proc_info= "Shutting down";

storage/ndb/src/ndbapi/NdbEventOperationImpl.cpp

@@ -1107,9 +1107,10 @@ NdbEventBuffer::flushIncompleteEvents(Uint64 gci)
   for(i = 0; i < sz; i++)
   {
     Gci_container* tmp = array + i;
-    if (tmp->m_gci < gci)
+    if (tmp->m_gci && tmp->m_gci < gci)
     {
       // we have found an old not-completed gci, remove it
+      ndbout_c("ndb: flushing incomplete epoch %lld (<%lld)", tmp->m_gci, gci);
       if(!tmp->m_data.is_empty())
       {
         free_list(tmp->m_data);
@@ -1257,7 +1258,6 @@ NdbEventBuffer::deleteUsedEventOperations()
         op->m_prev->m_next = op->m_next;
       else
         m_dropped_ev_op = op->m_next;
-      ndbout_c("deleting NdbEventOperation %p", op->m_facade);
       delete op->m_facade;
     }
   }
@@ -2506,7 +2506,6 @@ NdbEventBuffer::dropEventOperation(NdbEventOperation* tOp)
   {
     DBUG_PRINT("info", ("deleting op: %p", op));
     DBUG_ASSERT(op->m_node_bit_mask.isclear());
-    ndbout_c("deleting NdbEventOperation %p", op->m_facade);
     delete op->m_facade;
   }
   else

storage/ndb/src/ndbapi/NdbEventOperationImpl.hpp

@@ -25,7 +25,7 @@
 #include <UtilBuffer.hpp>
 
 #define NDB_EVENT_OP_MAGIC_NUMBER 0xA9F301B4
-#define EVENT_DEBUG
+//#define EVENT_DEBUG
 #ifdef EVENT_DEBUG
 #define DBUG_ENTER_EVENT(A) DBUG_ENTER(A)
 #define DBUG_RETURN_EVENT(A) DBUG_RETURN(A)