Merged with lp:~codership/codership-mysql/5.5-23, up to revision 3903

Docs/README-wsrep

@@ -377,11 +377,12 @@ the other cluster members by transferring state snapshot from one of them.
 The options below govern how this happens and should be set up before attempting
 to join or start a cluster.
 
-wsrep_sst_method=mysqldump
+wsrep_sst_method=rsync
    What method to use to copy database state to a newly joined node. Supported
    methods:
-   - mysqldump:  slow (except for small datasets) but most tested.
-   - rsync:      much faster on large datasets.
+   - mysqldump:  slow (except for small datasets) but allows for upgrade
+                 between major MySQL versions or InnoDB features.
+   - rsync:      much faster on large datasets (default).
    - rsync_wan:  same as rsync but with deltaxfer to minimize network traffic.
    - xtrabackup: very fast and practically non-blocking SST method based on
                  Percona's xtrabackup tool.

scripts/wsrep_sst_common.sh

@@ -1,4 +1,4 @@
-# Copyright (C) 2010 Codership Oy
+# Copyright (C) 2012 Codership Oy
 #
 # This program is free software; you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -19,6 +19,7 @@
 set -u
 
 WSREP_SST_OPT_BYPASS=0
+WSREP_SST_OPT_DATA=""
 
 while [ $# -gt 0 ]; do
 case "$1" in
@@ -86,7 +87,12 @@ shift
 done
 readonly WSREP_SST_OPT_BYPASS
 
+if [ -n "$WSREP_SST_OPT_DATA" ]
+then
     SST_PROGRESS_FILE="$WSREP_SST_OPT_DATA/sst_in_progress"
+else
+    SST_PROGRESS_FILE=""
+fi
 
 wsrep_log()
 {
@@ -108,6 +114,6 @@ wsrep_log_info()
 
 wsrep_cleanup_progress_file()
 {
-    rm -f $SST_PROGRESS_FILE 2>/dev/null
+    [ -n "$SST_PROGRESS_FILE" ] && rm -f "$SST_PROGRESS_FILE" 2>/dev/null
 }
 

sql/sql_class.cc

@@ -1116,6 +1116,8 @@ THD::THD()
   wsrep_status_vars       = 0;
   wsrep_mysql_replicated  = 0;
   wsrep_bf_thd            = NULL;
+  wsrep_TOI_pre_query     = NULL;
+  wsrep_TOI_pre_query_len = 0;
 #endif
   /* Call to init() below requires fully initialized Open_tables_state. */
   reset_open_tables_state(this);
@@ -1477,6 +1479,8 @@ void THD::init(void)
   wsrep_consistency_check = NO_CONSISTENCY_CHECK;
   wsrep_mysql_replicated  = 0;
   wsrep_bf_thd = NULL;
+  wsrep_TOI_pre_query     = NULL;
+  wsrep_TOI_pre_query_len = 0;
 #endif
   if (variables.sql_log_bin)
     variables.option_bits|= OPTION_BIN_LOG;

sql/sql_class.h

@@ -2381,6 +2381,9 @@ public:
   wsrep_stats_var*          wsrep_status_vars;
   int                       wsrep_mysql_replicated;
   THD*                      wsrep_bf_thd;
+  const char*               wsrep_TOI_pre_query; /* a query to apply before 
+						    the actual TOI query */
+  size_t                    wsrep_TOI_pre_query_len;
 #endif /* WITH_WSREP */
   /**
     Internal parser state.

sql/sql_parse.cc

@@ -2893,12 +2893,6 @@ case SQLCOM_PREPARE:
       if (create_info.options & HA_LEX_CREATE_TMP_TABLE)
         thd->variables.option_bits|= OPTION_KEEP_LOG;
       /* regular create */
-#ifdef WITH_WSREP
-      if (!thd->is_current_stmt_binlog_format_row() ||
-	  !(create_info.options & HA_LEX_CREATE_TMP_TABLE))
-       WSREP_TO_ISOLATION_BEGIN(create_table->db, create_table->table_name,
-                                 NULL)
-#endif /* WITH_WSREP */
       if (create_info.options & HA_LEX_CREATE_TABLE_LIKE)
       {
         /* CREATE TABLE ... LIKE ... */
@@ -2907,6 +2901,12 @@ case SQLCOM_PREPARE:
       }
       else
       {
+#ifdef WITH_WSREP
+        if (!thd->is_current_stmt_binlog_format_row() ||
+            !(create_info.options & HA_LEX_CREATE_TMP_TABLE))
+          WSREP_TO_ISOLATION_BEGIN(create_table->db, create_table->table_name,
+                                   NULL)
+#endif /* WITH_WSREP */
         /* Regular CREATE TABLE */
         res= mysql_create_table(thd, create_table,
                                 &create_info, &alter_info);
@@ -8260,6 +8260,15 @@ wsrep_status_t wsrep_apply_cb(void* const ctx,
 #endif /* WSREP_PROC_INFO */
 
   if (WSREP_OK != rcode) wsrep_write_rbr_buf(thd, buf, buf_len);
+  TABLE *tmp;
+  while ((tmp = thd->temporary_tables))
+  {
+    WSREP_DEBUG("Applier %lu, has temporary tables: %s.%s",
+      thd->thread_id, 
+      (tmp->s) ? tmp->s->db.str : "void",
+      (tmp->s) ? tmp->s->table_name.str : "void");
+    close_temporary_table(thd, tmp, 1, 1);    
+  }
 
   return rcode;
 }
@@ -8447,6 +8456,10 @@ void wsrep_replication_process(THD *thd)
   mysql_cond_broadcast(&COND_thread_count);
   mysql_mutex_unlock(&LOCK_thread_count);
 
+  if (thd->temporary_tables)
+  {
+    WSREP_DEBUG("Applier %lu, has temporary tables at exit", thd->thread_id);
+  }
   wsrep_return_from_bf_mode(thd, &shadow);
   DBUG_VOID_RETURN;
 }

sql/sql_table.cc

@@ -4752,6 +4752,49 @@ bool mysql_create_like_table(THD* thd, TABLE_LIST* table, TABLE_LIST* src_table,
   uint not_used;
   DBUG_ENTER("mysql_create_like_table");
 
+#ifdef WITH_WSREP
+  if (!thd->wsrep_applier)
+  {
+    TABLE *tmp_table;
+    bool is_tmp_table= FALSE;
+
+    for (tmp_table= thd->temporary_tables; tmp_table; tmp_table=tmp_table->next)
+    {
+      if (!strcmp(src_table->db, tmp_table->s->db.str)     &&
+          !strcmp(src_table->table_name, tmp_table->s->table_name.str))
+      {
+        is_tmp_table= TRUE;
+        break;
+      }
+    }
+    if (!is_tmp_table)
+    {
+      WSREP_TO_ISOLATION_BEGIN(table->db, table->table_name, NULL);
+    }
+    else
+    {
+      TABLE_LIST tbl;
+      bzero((void*) &tbl, sizeof(tbl));
+      tbl.db= src_table->db;
+      tbl.table_name= tbl.alias= src_table->table_name;
+      tbl.table= tmp_table;
+      char buf[2048];
+      String query(buf, sizeof(buf), system_charset_info);
+      query.length(0);  // Have to zero it since constructor doesn't
+
+      (void)  store_create_info(thd, &tbl, &query, NULL, TRUE);
+      WSREP_DEBUG("TMP TABLE: %s", query.ptr());
+
+      thd->wsrep_TOI_pre_query=     query.ptr();
+      thd->wsrep_TOI_pre_query_len= query.length();
+      
+      WSREP_TO_ISOLATION_BEGIN(table->db, table->table_name, NULL);
+
+      thd->wsrep_TOI_pre_query=      NULL;
+      thd->wsrep_TOI_pre_query_len= 0;
+    }
+  }
+#endif
 
   /*
     We the open source table to get its description in HA_CREATE_INFO
@@ -4899,6 +4942,12 @@ bool mysql_create_like_table(THD* thd, TABLE_LIST* table, TABLE_LIST* src_table,
 
 err:
   DBUG_RETURN(res);
+#ifdef WITH_WSREP
+ error:
+  thd->wsrep_TOI_pre_query= NULL;
+  DBUG_RETURN(TRUE);
+#endif /* WITH_WSREP */
+
 }
 
 

sql/wsrep_check_opts.cc

@@ -40,7 +40,7 @@ static struct opt opts[] =
 {
     { "wsrep_slave_threads",     "1" }, // mysqld.cc
     { "bind_address",      "0.0.0.0" }, // mysqld.cc
-    { "wsrep_sst_method","mysqldump" }, // mysqld.cc
+    { "wsrep_sst_method",    "rsync" }, // mysqld.cc
     { "wsrep_sst_receive_address","AUTO"}, // mysqld.cc
     { "binlog_format",         "ROW" }, // mysqld.cc
     { "wsrep_provider",       "none" }, // mysqld.cc

sql/wsrep_hton.cc

@@ -75,7 +75,12 @@ void wsrep_register_hton(THD* thd, bool all)
     if (i->ht()->db_type == DB_TYPE_INNODB)
     {
       trans_register_ha(thd, all, wsrep_hton);
+
+      /* follow innodb read/write settting */
+      if (i->is_trx_read_write())
+      {
 	thd->ha_data[wsrep_hton->slot].ha_info[all].set_trx_read_write();
+      }
       break;
     }
   }

sql/wsrep_mysqld.cc

@@ -983,16 +983,23 @@ int wsrep_to_buf_helper(
   if (open_cached_file(&tmp_io_cache, mysql_tmpdir, TEMP_PREFIX,
                        65536, MYF(MY_WME)))
     return 1;
-  Query_log_event ev(thd, query, query_len, FALSE, FALSE, FALSE, 0);
+
   int ret(0);
-
-  Format_description_log_event *tmp_fd = new Format_description_log_event(4);
-  tmp_fd->checksum_alg = binlog_checksum_options;
-  tmp_fd->write(&tmp_io_cache);
-  delete tmp_fd;
-
+  /* if there is prepare query, add event for it */
+  if (thd->wsrep_TOI_pre_query)
+  {
+    Query_log_event ev(thd, thd->wsrep_TOI_pre_query, 
+		       thd->wsrep_TOI_pre_query_len, 
+		       FALSE, FALSE, FALSE, 0);
     if (ev.write(&tmp_io_cache)) ret= 1;
+  }
+
+  /* append the actual query */
+  Query_log_event ev(thd, query, query_len, FALSE, FALSE, FALSE, 0);
+  if (ev.write(&tmp_io_cache)) ret= 1;
+
   if (!ret && wsrep_write_cache(&tmp_io_cache, buf, buf_len)) ret= 1;
+
   close_cached_file(&tmp_io_cache);
   return ret;
 }

sql/wsrep_sst.cc

@@ -45,8 +45,9 @@ extern const char wsrep_defaults_file[];
 #define WSREP_SST_OPT_BYPASS   "--bypass"
 
 #define WSREP_SST_MYSQLDUMP    "mysqldump"
+#define WSREP_SST_RSYNC        "rsync"
 #define WSREP_SST_SKIP         "skip"
-#define WSREP_SST_DEFAULT      WSREP_SST_MYSQLDUMP
+#define WSREP_SST_DEFAULT      WSREP_SST_RSYNC
 #define WSREP_SST_ADDRESS_AUTO "AUTO"
 #define WSREP_SST_AUTH_MASK    "********"
 
@@ -691,9 +692,11 @@ static int sst_donate_mysqldump (const char*         addr,
               WSREP_SST_OPT_PORT" '%s' "
               WSREP_SST_OPT_LPORT" '%u' "
               WSREP_SST_OPT_SOCKET" '%s' "
+              WSREP_SST_OPT_DATA" '%s' "
               WSREP_SST_OPT_GTID" '%s:%lld'"
               "%s",
-              user, pswd, host, port, mysqld_port, mysqld_unix_port, uuid_str,
+              user, pswd, host, port, mysqld_port, mysqld_unix_port,
+              mysql_real_data_home, uuid_str,
               (long long)seqno, bypass ? " "WSREP_SST_OPT_BYPASS : "");
 
     WSREP_DEBUG("Running: '%s'", cmd_str);

sql/wsrep_var.cc

@@ -236,7 +236,17 @@ bool wsrep_provider_update (sys_var *self, THD* thd, enum_var_type type)
 
   WSREP_DEBUG("wsrep_provider_update: %s", wsrep_provider);
 
+  /* stop replication is heavy operation, and includes closing all client 
+     connections. Closing clients may need to get LOCK_global_system_variables
+     at least in MariaDB.
+
+     Note: releasing LOCK_global_system_variables may cause race condition, if 
+     there can be several concurrent clients changing wsrep_provider
+  */
+  mysql_mutex_unlock(&LOCK_global_system_variables);
   wsrep_stop_replication(thd);
+  mysql_mutex_lock(&LOCK_global_system_variables);
+
   wsrep_deinit();
 
   char* tmp= strdup(wsrep_provider); // wsrep_init() rewrites provider 

storage/innobase/row/row0upd.c

@@ -52,6 +52,7 @@ Created 12/27/1996 Heikki Tuuri
 #include "eval0eval.h"
 #include "buf0lru.h"
 #ifdef WITH_WSREP
+#include "ha_prototypes.h"
 extern my_bool wsrep_debug;
 #endif
 

storage/xtradb/row/row0upd.c

@@ -52,6 +52,7 @@ Created 12/27/1996 Heikki Tuuri
 #include "eval0eval.h"
 #include "buf0lru.h"
 #ifdef WITH_WSREP
+#include "ha_prototypes.h"
 extern my_bool wsrep_debug;
 #endif
 

support-files/mysql.spec.sh

@@ -300,11 +300,14 @@ documentation and the manual for more information.
 %package -n MySQL-server%{product_suffix}
 %if %{defined with_wsrep}
 Version:        %{mysql_version}
-#Release:        %{wsrep_version}.%{release}
 %endif
 Summary:        MySQL: a very fast and reliable SQL database server
 Group:          Applications/Databases
+%if %{defined with_wsrep}
+Requires:       %{distro_requires} rsync
+%else
 Requires:       %{distro_requires}
+%endif
 %if %{defined susever}
 Provides:       msqlormysql MySQL MySQL-server
 Conflicts:      mysql mysql-server mysql-advanced mysql-server-advanced

support-files/wsrep.cnf.sh

@@ -108,7 +108,7 @@ wsrep_notify_cmd=
 ##
 
 # State Snapshot Transfer method
-wsrep_sst_method=mysqldump
+wsrep_sst_method=rsync
 
 # Address which donor should send State Snapshot to.
 # Should be the address of THIS node. DON'T SET IT TO DONOR ADDRESS!!!