Merge 10.0-galera into 10.1

14 files changed, 112 insertions, 87 deletions

diff --git a/sql/keycaches.cc b/sql/keycaches.cc
index 9426ccc9e95..78e64ebac72 100644
--- a/sql/keycaches.cc
+++ b/sql/keycaches.cc
@@ -223,6 +223,7 @@ Rpl_filter *get_or_create_rpl_filter(const char *name, uint length)
 void free_rpl_filter(const char *name, Rpl_filter *filter)
 {
   delete filter;
+  filter= 0;
 }
 
 void free_all_rpl_filters()
diff --git a/sql/mysqld.cc b/sql/mysqld.cc
index 423d07b02cd..6d8305dea91 100644
--- a/sql/mysqld.cc
+++ b/sql/mysqld.cc
@@ -4976,7 +4976,7 @@ a file name for --log-bin-index option", opt_binlog_index_name);
       - SST may modify binlog index file, so it must be opened
         after SST has happened
      */
-  if (WSREP_ON && !wsrep_recovery) /* WSREP BEFORE SE */
+  if (WSREP_ON && !wsrep_recovery && !opt_abort) /* WSREP BEFORE SE */
   {
     if (opt_bootstrap) // bootsrap option given - disable wsrep functionality
     {
@@ -9240,6 +9240,16 @@ static int get_options(int *argc_ptr, char ***argv_ptr)
   else
     global_system_variables.option_bits&= ~OPTION_BIG_SELECTS;
 
+  if (!opt_bootstrap && WSREP_PROVIDER_EXISTS &&
+      global_system_variables.binlog_format != BINLOG_FORMAT_ROW)
+  {
+
+    WSREP_ERROR ("Only binlog_format = 'ROW' is currently supported. "
+                 "Configured value: '%s'. Please adjust your configuration.",
+                 binlog_format_names[global_system_variables.binlog_format]);
+    return 1;
+  }
+
   // Synchronize @@global.autocommit on --autocommit
   const ulonglong turn_bit_on= opt_autocommit ?
     OPTION_AUTOCOMMIT : OPTION_NOT_AUTOCOMMIT;
diff --git a/sql/rpl_mi.cc b/sql/rpl_mi.cc
index e97b5ecc8fe..8ce67bbf4b8 100644
--- a/sql/rpl_mi.cc
+++ b/sql/rpl_mi.cc
@@ -92,6 +92,13 @@ Master_info::Master_info(LEX_STRING *connection_name_arg,
 
 Master_info::~Master_info()
 {
+#ifdef WITH_WSREP
+  /*
+    Do not free "wsrep" rpl_filter. It will eventually be freed by
+    free_all_rpl_filters() when server terminates.
+  */
+  if (strncmp(connection_name.str, STRING_WITH_LEN("wsrep")))
+#endif
   rpl_filters.delete_element(connection_name.str, connection_name.length,
                              (void (*)(const char*, uchar*)) free_rpl_filter);
   my_free(connection_name.str);
diff --git a/sql/sql_alter.cc b/sql/sql_alter.cc
index cfe360217c2..e413361191d 100644
--- a/sql/sql_alter.cc
+++ b/sql/sql_alter.cc
@@ -307,16 +307,12 @@ bool Sql_cmd_alter_table::execute(THD *thd)
 #ifdef WITH_WSREP
   TABLE *find_temporary_table(THD *thd, const TABLE_LIST *tl);
 
-  if (WSREP(thd) &&
-      (!thd->is_current_stmt_binlog_format_row() ||
-       !find_temporary_table(thd, first_table))  &&
-      wsrep_to_isolation_begin(thd,
-                               lex->name.str ? select_lex->db : NULL,
-                               lex->name.str ? lex->name.str : NULL,
-                               first_table))
+  if ((!thd->is_current_stmt_binlog_format_row() ||
+       !find_temporary_table(thd, first_table)))
   {
-    WSREP_WARN("ALTER TABLE isolation failure");
-    DBUG_RETURN(TRUE);
+    WSREP_TO_ISOLATION_BEGIN(((lex->name.str) ? select_lex->db : NULL),
+                             ((lex->name.str) ? lex->name.str : NULL),
+                             first_table);
   }
 #endif /* WITH_WSREP */
 
@@ -329,6 +325,12 @@ bool Sql_cmd_alter_table::execute(THD *thd)
                             lex->ignore);
 
   DBUG_RETURN(result);
+
+#ifdef WITH_WSREP
+error:
+  WSREP_WARN("ALTER TABLE isolation failure");
+  DBUG_RETURN(TRUE);
+#endif /* WITH_WSREP */
 }
 
 bool Sql_cmd_discard_import_tablespace::execute(THD *thd)
diff --git a/sql/sql_class.cc b/sql/sql_class.cc
index 4a66542b71f..9c457557bcb 100644
--- a/sql/sql_class.cc
+++ b/sql/sql_class.cc
@@ -6451,6 +6451,14 @@ int THD::binlog_query(THD::enum_binlog_query_type qtype, char const *query_arg,
       The MYSQL_LOG::write() function will set the STMT_END_F flag and
       flush the pending rows event if necessary.
     */
+    /*
+      Even though wsrep only supports ROW binary log format, a user can set
+      binlog format to STATEMENT (wsrep_forced_binlog_format). In which case
+      the control might reach here even when binary logging (--log-bin) is
+      not enabled. This is possible because wsrep patch partially enables
+      binary logging by setting wsrep_emulate_binlog.
+    */
+    if (mysql_bin_log.is_open())
     {
       Query_log_event qinfo(this, query_arg, query_len, is_trans, direct,
                             suppress_use, errcode);
diff --git a/sql/sql_parse.cc b/sql/sql_parse.cc
index 4f88f8d5c96..64e016f8311 100644
--- a/sql/sql_parse.cc
+++ b/sql/sql_parse.cc
@@ -2820,8 +2820,6 @@ mysql_execute_command(THD *thd)
 #endif
   case SQLCOM_SHOW_STATUS:
   {
-    if (WSREP_CLIENT(thd) && wsrep_sync_wait(thd))
-      goto error;
     execute_show_status(thd, all_tables);
     break;
   }
@@ -6965,7 +6963,7 @@ static void wsrep_mysql_parse(THD *thd, char *rawbuf, uint length,
         WSREP_DEBUG("abort in exec query state, avoiding autocommit");
       }
 
-      if (thd->wsrep_conflict_state== MUST_REPLAY) 
+      if (thd->wsrep_conflict_state == MUST_REPLAY)
       {
         wsrep_replay_transaction(thd);
       }
diff --git a/sql/sql_prepare.cc b/sql/sql_prepare.cc
index cf92b016a9f..7b157f98431 100644
--- a/sql/sql_prepare.cc
+++ b/sql/sql_prepare.cc
@@ -3657,7 +3657,7 @@ reexecute:
         break;
 
       case MUST_REPLAY:
-        (void)wsrep_replay_transaction(thd);
+        (void) wsrep_replay_transaction(thd);
         break;
 
       default:
diff --git a/sql/sys_vars.cc b/sql/sys_vars.cc
index cf2eabb5691..315dc01952b 100644
--- a/sql/sys_vars.cc
+++ b/sql/sys_vars.cc
@@ -437,6 +437,24 @@ static bool binlog_format_check(sys_var *self, THD *thd, set_var *var)
   if (check_has_super(self, thd, var))
     return true;
 
+  /*
+    MariaDB Galera does not support STATEMENT or MIXED binlog format currently.
+  */
+  if (WSREP(thd) && var->save_result.ulonglong_value != BINLOG_FORMAT_ROW)
+  {
+    // Push a warning to the error log.
+    push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, ER_UNKNOWN_ERROR,
+                        "MariaDB Galera does not support binlog format: %s",
+                        binlog_format_names[var->save_result.ulonglong_value]);
+
+    if (var->type == OPT_GLOBAL)
+    {
+      WSREP_ERROR("MariaDB Galera does not support binlog format: %s",
+                  binlog_format_names[var->save_result.ulonglong_value]);
+      return true;
+    }
+  }
+
   if (var->type == OPT_GLOBAL)
     return false;
 
@@ -465,26 +483,6 @@ static bool binlog_format_check(sys_var *self, THD *thd, set_var *var)
          ER_INSIDE_TRANSACTION_PREVENTS_SWITCH_BINLOG_FORMAT))
     return true;
 
-#ifdef WITH_WSREP
-  /* MariaDB Galera does not support STATEMENT or MIXED binlog
-  format currently */
-  if (WSREP(thd) &&
-     (var->save_result.ulonglong_value == BINLOG_FORMAT_STMT ||
-      var->save_result.ulonglong_value == BINLOG_FORMAT_MIXED))
-  {
-    WSREP_DEBUG("MariaDB Galera does not support binlog format : %s",
-                var->save_result.ulonglong_value == BINLOG_FORMAT_STMT ?
-                "STATEMENT" : "MIXED");
-    /* Push also warning, because error message is general */
-    push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN,
-                        ER_UNKNOWN_ERROR,
-                        "MariaDB Galera does not support binlog format: %s",
-                        var->save_result.ulonglong_value == BINLOG_FORMAT_STMT ?
-                        "STATEMENT" : "MIXED");
-    return true;
-  }
-#endif
-
   return false;
 }
 
@@ -4587,7 +4585,7 @@ static Sys_var_charptr Sys_wsrep_provider_options(
 static Sys_var_charptr Sys_wsrep_data_home_dir(
        "wsrep_data_home_dir", "home directory for wsrep provider",
        READ_ONLY GLOBAL_VAR(wsrep_data_home_dir), CMD_LINE(REQUIRED_ARG),
-       IN_FS_CHARSET, DEFAULT(""));
+       IN_FS_CHARSET, DEFAULT(mysql_real_data_home));
 
 static Sys_var_charptr Sys_wsrep_cluster_name(
        "wsrep_cluster_name", "Name for the cluster",
diff --git a/sql/wsrep_applier.cc b/sql/wsrep_applier.cc
index 23687e98c32..03d356aa046 100644
--- a/sql/wsrep_applier.cc
+++ b/sql/wsrep_applier.cc
@@ -95,7 +95,8 @@ static wsrep_cb_status_t wsrep_apply_events(THD*        thd,
 
   DBUG_ENTER("wsrep_apply_events");
 
-  if (thd->killed == KILL_CONNECTION)
+  if (thd->killed == KILL_CONNECTION &&
+      thd->wsrep_conflict_state != REPLAYING)
   {
     WSREP_INFO("applier has been aborted, skipping apply_rbr: %lld",
                (long long) wsrep_thd_trx_seqno(thd));
diff --git a/sql/wsrep_mysqld.cc b/sql/wsrep_mysqld.cc
index 599b9be69c4..19736cef7bf 100644
--- a/sql/wsrep_mysqld.cc
+++ b/sql/wsrep_mysqld.cc
@@ -333,7 +333,13 @@ wsrep_view_handler_cb (void*                    app_ctx,
              wsrep_cluster_size, wsrep_local_index, view->proto_ver);
 
   /* Proceed further only if view is PRIMARY */
-  if (WSREP_VIEW_PRIMARY != view->status) {
+  if (WSREP_VIEW_PRIMARY != view->status)
+  {
+#ifdef HAVE_QUERY_CACHE
+    // query cache must be initialised by now
+    query_cache.flush();
+#endif /* HAVE_QUERY_CACHE */
+
     wsrep_ready_set(FALSE);
     memb_status= WSREP_MEMBER_UNDEFINED;
     /* Always record local_uuid and local_seqno in non-prim since this
@@ -380,9 +386,16 @@ wsrep_view_handler_cb (void*                    app_ctx,
     wsrep_ready_set(FALSE);
 
     /* Close client connections to ensure that they don't interfere
-     * with SST */
-    WSREP_DEBUG("[debug]: closing client connections for PRIM");
-    wsrep_close_client_connections(TRUE);
+     * with SST. Necessary only if storage engines are initialized
+     * before SST.
+     * TODO: Just killing all ongoing transactions should be enough
+     * since wsrep_ready is OFF and no new transactions can start.
+     */
+    if (!wsrep_before_SE())
+    {
+        WSREP_DEBUG("[debug]: closing client connections for PRIM");
+        wsrep_close_client_connections(TRUE);
+    }
 
     ssize_t const req_len= wsrep_sst_prepare (sst_req);
 
@@ -662,9 +675,6 @@ int wsrep_init()
             wsrep->provider_vendor,  sizeof(provider_vendor) - 1);
   }
 
-  if (!wsrep_data_home_dir || strlen(wsrep_data_home_dir) == 0)
-    wsrep_data_home_dir = mysql_real_data_home;
-
   char node_addr[512]= { 0, };
   size_t const node_addr_max= sizeof(node_addr) - 1;
   if (!wsrep_node_address || !strcmp(wsrep_node_address, ""))
@@ -1686,7 +1696,6 @@ wsrep_grant_mdl_exception(MDL_context *requestor_ctx,
 pthread_handler_t start_wsrep_THD(void *arg)
 {
   THD *thd;
-  rpl_sql_thread_info sql_info(NULL);
   wsrep_thd_processor_fun processor= (wsrep_thd_processor_fun)arg;
 
   if (my_thread_init())
@@ -1717,7 +1726,6 @@ pthread_handler_t start_wsrep_THD(void *arg)
   thd->bootstrap=1;
   thd->max_client_packet_length= thd->net.max_packet;
   thd->security_ctx->master_access= ~(ulong)0;
-  thd->system_thread_info.rpl_sql_info= &sql_info;
 
   /* from handle_one_connection... */
   pthread_detach_this_thread();
diff --git a/sql/wsrep_mysqld.h b/sql/wsrep_mysqld.h
index a6430560e3e..085dc5de453 100644
--- a/sql/wsrep_mysqld.h
+++ b/sql/wsrep_mysqld.h
@@ -162,7 +162,7 @@ extern wsrep_seqno_t wsrep_locked_seqno;
   (global_system_variables.wsrep_on)
 
 #define WSREP(thd) \
-  (WSREP_ON && (thd && thd->variables.wsrep_on))
+  (WSREP_ON && wsrep && (thd && thd->variables.wsrep_on))
 
 #define WSREP_CLIENT(thd) \
     (WSREP(thd) && thd->wsrep_client_thread)
@@ -315,6 +315,7 @@ int wsrep_create_trigger_query(THD *thd, uchar** buf, size_t* buf_len);
 #define WSREP_EMULATE_BINLOG(thd) (0)
 #define WSREP_CLIENT(thd) (0)
 #define WSREP_FORMAT(my_format) ((ulong)my_format)
+#define WSREP_PROVIDER_EXISTS (0)
 #define wsrep_emulate_bin_log (0)
 #define wsrep_xid_seqno(X) (0)
 #define wsrep_to_isolation (0)
diff --git a/sql/wsrep_thd.cc b/sql/wsrep_thd.cc
index 2b4980ae05f..7ef31171846 100644
--- a/sql/wsrep_thd.cc
+++ b/sql/wsrep_thd.cc
@@ -89,32 +89,6 @@ void wsrep_client_rollback(THD *thd)
 
 #define NUMBER_OF_FIELDS_TO_IDENTIFY_COORDINATOR 1
 #define NUMBER_OF_FIELDS_TO_IDENTIFY_WORKER 2
-//#include "rpl_info_factory.h"
-
-#ifdef NOT_USED
-static Relay_log_info* wsrep_relay_log_init(const char* log_fname)
-{
-
-  /* MySQL 5.6 version has rli factory: */
-#ifdef MYSQL_56
-  uint rli_option = INFO_REPOSITORY_DUMMY;
-  Relay_log_info *rli= NULL;
-  rli = Rpl_info_factory::create_rli(rli_option, false);
-  rli->set_rli_description_event(
-      new Format_description_log_event(BINLOG_VERSION));
-#endif
-  Relay_log_info* rli= new Relay_log_info(false);
-  rli->sql_driver_thd= current_thd;
- 
-  rli->no_storage= true;
-  rli->relay_log.description_event_for_exec=
-    new Format_description_log_event(4);
-
-  return rli;
-}
-#endif
-
-class Master_info;
 
 static rpl_group_info* wsrep_relay_group_init(const char* log_fname)
 {
@@ -126,13 +100,27 @@ static rpl_group_info* wsrep_relay_group_init(const char* log_fname)
     rli->relay_log.description_event_for_exec=
       new Format_description_log_event(4);
   }
-  static LEX_STRING dbname= { C_STRING_WITH_LEN("mysql") };
 
-  rli->mi = new Master_info( &dbname,  false);
-  //rli->mi = new Master_info( &(C_STRING_WITH_LEN("wsrep")),  false);
+  static LEX_STRING connection_name= { C_STRING_WITH_LEN("wsrep") };
+
+  /*
+    Master_info's constructor initializes rpl_filter by either an already
+    constructed Rpl_filter object from global 'rpl_filters' list if the
+    specified connection name is same, or it constructs a new Rpl_filter
+    object and adds it to rpl_filters. This object is later destructed by
+    Mater_info's destructor by looking it up based on connection name in
+    rpl_filters list.
+
+    However, since all Master_info objects created here would share same
+    connection name ("wsrep"), destruction of any of the existing Master_info
+    objects (in wsrep_return_from_bf_mode()) would free rpl_filter referenced
+    by any/all existing Master_info objects.
 
-  rli->mi->rpl_filter = new Rpl_filter;
-  copy_filter_setting(rli->mi->rpl_filter, get_or_create_rpl_filter("", 0));
+    In order to avoid that, we have added a check in Master_info's destructor
+    to not free the "wsrep" rpl_filter. It will eventually be freed by
+    free_all_rpl_filters() when server terminates.
+  */
+  rli->mi = new Master_info(&connection_name, false);
 
   rli->sql_driver_thd= current_thd;
 
@@ -154,9 +142,11 @@ static void wsrep_prepare_bf_thd(THD *thd, struct wsrep_thd_shadow* shadow)
   else
     thd->variables.option_bits&= ~(OPTION_BIN_LOG);
 
-  //if (!thd->wsrep_rli) thd->wsrep_rli= wsrep_relay_log_init("wsrep_relay");
   if (!thd->wsrep_rgi) thd->wsrep_rgi= wsrep_relay_group_init("wsrep_relay");
-  //  thd->wsrep_rli->info_thd = thd;
+
+  /* thd->system_thread_info.rpl_sql_info isn't initialized. */
+  thd->system_thread_info.rpl_sql_info=
+    new rpl_sql_thread_info(thd->wsrep_rgi->rli->mi->rpl_filter);
 
   thd->wsrep_exec_mode= REPL_RECV;
   thd->net.vio= 0;
@@ -180,12 +170,11 @@ static void wsrep_return_from_bf_mode(THD *thd, struct wsrep_thd_shadow* shadow)
   thd->variables.tx_isolation = shadow->tx_isolation;
   thd->reset_db(shadow->db, shadow->db_length);
 
-  delete thd->wsrep_rgi->rli->mi->rpl_filter;
+  delete thd->system_thread_info.rpl_sql_info;
   delete thd->wsrep_rgi->rli->mi;
   delete thd->wsrep_rgi->rli;
   delete thd->wsrep_rgi;
   thd->wsrep_rgi = NULL;
-;
 }
 
 void wsrep_replay_transaction(THD *thd)
@@ -283,6 +272,10 @@ void wsrep_replay_transaction(THD *thd)
         WSREP_ERROR("trx_replay failed for: %d, query: %s",
                     rcode, thd->query() ? thd->query() : "void");
         /* we're now in inconsistent state, must abort */
+
+        /* http://bazaar.launchpad.net/~codership/codership-mysql/5.6/revision/3962#sql/wsrep_thd.cc */
+        mysql_mutex_unlock(&thd->LOCK_wsrep_thd);
+
         unireg_abort(1);
         break;
       }
diff --git a/sql/wsrep_utils.cc b/sql/wsrep_utils.cc
index c4a992c751a..2c3fc0d77fd 100644
--- a/sql/wsrep_utils.cc
+++ b/sql/wsrep_utils.cc
@@ -325,10 +325,6 @@ thd::~thd ()
 /* Returns INADDR_NONE, INADDR_ANY, INADDR_LOOPBACK or something else */
 unsigned int wsrep_check_ip (const char* const addr)
 {
-#if 0
-  if (addr && 0 == strcasecmp(addr, MY_BIND_ALL_ADDRESSES)) return INADDR_ANY;
-#endif
-
   unsigned int ret = INADDR_NONE;
   struct addrinfo *res, hints;
 
@@ -380,11 +376,12 @@ size_t wsrep_guess_ip (char* buf, size_t buf_len)
     unsigned int const ip_type= wsrep_check_ip(my_bind_addr_str);
 
     if (INADDR_NONE == ip_type) {
-      WSREP_ERROR("Networking not configured, cannot receive state transfer.");
+      WSREP_ERROR("Networking not configured, cannot receive state "
+                  "transfer.");
       return 0;
     }
 
-    if (INADDR_ANY != ip_type) {;
+    if (INADDR_ANY != ip_type) {
       strncpy (buf, my_bind_addr_str, buf_len);
       return strlen(buf);
     }
diff --git a/sql/wsrep_var.cc b/sql/wsrep_var.cc
index 1c0e51e50ad..ca921b7ef18 100644
--- a/sql/wsrep_var.cc
+++ b/sql/wsrep_var.cc
@@ -66,6 +66,7 @@ bool wsrep_causal_reads_update (SV *sv)
   } else {
     sv->wsrep_sync_wait &= ~WSREP_SYNC_WAIT_BEFORE_READ;
   }
+
   return false;
 }