MDEV-33211 : Galera SST on maria-backup causes donor node to be unres…

…ponsive

If mariabackup with backup locks is used on SST we do not
pause and desync galera provider at all. If WSREP_MODE_BF_MARIABACKUP
case provider is paused and desync at BLOCK_COMMIT phase. In
other cases provider is paused and desync at BLOCK_DDL phase.

extra/mariabackup/backup_mysql.cc

@@ -1417,12 +1417,103 @@ write_slave_info(ds_ctxt *datasink, MYSQL *connection)
 
 
 /*********************************************************************//**
-Old function, not needed anymore with BACKUP LOCKS
+Retrieves MySQL Galera and saves it in a file. It also prints it to stdout.
+
+We should create xtrabackup_galelera_info file even when backup locks
+are used because donor's wsrep_gtid_domain_id is needed later in joiner.
+Note that at this stage wsrep_local_state_uuid and wsrep_last_committed
+are inconsistent but they are not used in joiner. Joiner will rewrite this file
+at mariabackup --prepare phase and thus there is extra file donor_galera_info.
+Information is needed to maitain wsrep_gtid_domain_id and gtid_binlog_pos
+same across the cluster. If joiner node have different wsrep_gtid_domain_id
+we should still receive effective domain id from the donor node,
+and use it.
 */
 bool
 write_galera_info(ds_ctxt *datasink, MYSQL *connection)
 {
-  return true;                                  // Success
+  char *state_uuid = NULL, *state_uuid55 = NULL;
+  char *last_committed = NULL, *last_committed55 = NULL;
+  char *domain_id = NULL, *domain_id55 = NULL;
+  bool result=true;
+  uint n_values=0;
+  char *wsrep_on = NULL, *wsrep_on55 = NULL;
+
+  mysql_variable vars[] = {
+    {"Wsrep_on", &wsrep_on},
+    {"wsrep_on", &wsrep_on55},
+    {NULL, NULL}
+  };
+
+  mysql_variable status[] = {
+    {"Wsrep_local_state_uuid", &state_uuid},
+    {"wsrep_local_state_uuid", &state_uuid55},
+    {"Wsrep_last_committed", &last_committed},
+    {"wsrep_last_committed", &last_committed55},
+    {NULL, NULL}
+  };
+
+  mysql_variable value[] = {
+    {"Wsrep_gtid_domain_id", &domain_id},
+    {"wsrep_gtid_domain_id", &domain_id55},
+    {NULL, NULL}
+  };
+
+  n_values= read_mysql_variables(connection, "SHOW VARIABLES", vars, true);
+
+  if (n_values == 0 || (wsrep_on == NULL && wsrep_on55 == NULL))
+  {
+    msg("Server is not Galera node thus --galera-info does not "
+	"have any effect.");
+    result = true;
+    goto cleanup;
+  }
+
+  read_mysql_variables(connection, "SHOW STATUS", status, true);
+
+  if ((state_uuid == NULL && state_uuid55 == NULL)
+      || (last_committed == NULL && last_committed55 == NULL))
+  {
+    msg("Warning: failed to get master wsrep state from SHOW STATUS.");
+    result = true;
+    goto cleanup;
+  }
+
+  n_values= read_mysql_variables(connection, "SHOW VARIABLES LIKE 'wsrep%'", value, true);
+
+  if (n_values == 0 || (domain_id == NULL && domain_id55 == NULL))
+  {
+    msg("Warning: failed to get master wsrep state from SHOW VARIABLES.");
+    result = true;
+    goto cleanup;
+  }
+
+  result= datasink->backup_file_printf(XTRABACKUP_GALERA_INFO,
+    "%s:%s %s\n", state_uuid ? state_uuid : state_uuid55,
+    last_committed ? last_committed : last_committed55,
+    domain_id ? domain_id : domain_id55);
+
+  if (result)
+  {
+    result= datasink->backup_file_printf(XTRABACKUP_DONOR_GALERA_INFO,
+      "%s:%s %s\n", state_uuid ? state_uuid : state_uuid55,
+      last_committed ? last_committed : last_committed55,
+      domain_id ? domain_id : domain_id55);
+  }
+
+  if (result)
+    write_current_binlog_file(datasink, connection);
+
+  if (result)
+    msg("Writing Galera info succeeded with %s:%s %s",
+      state_uuid ? state_uuid : state_uuid55,
+      last_committed ? last_committed : last_committed55,
+      domain_id ? domain_id : domain_id55);
+
+cleanup:
+  free_mysql_variables(status);
+
+  return(result);
 }
 
 

mysql-test/include/wait_until_connected_again.inc

@@ -11,7 +11,7 @@ let $counter= 5000;
 let $mysql_errno= 9999;
 while ($mysql_errno)
 {
-  --error 0,ER_ACCESS_DENIED_ERROR,ER_SERVER_SHUTDOWN,ER_CONNECTION_KILLED,ER_LOCK_WAIT_TIMEOUT,2002,2006,2013,HA_ERR_NO_ENCRYPTION
+  --error 0,ER_ACCESS_DENIED_ERROR,ER_SERVER_SHUTDOWN,ER_CONNECTION_KILLED,ER_LOCK_WAIT_TIMEOUT,2002,2006,2013,HA_ERR_NO_ENCRYPTION,2026
   select 1;
 
   dec $counter;

mysql-test/suite/galera/r/galera_bf_abort_mariabackup.result

@@ -12,9 +12,9 @@ connection node_1;
 connection node_2;
 Starting server ...
 connection node_1;
-# Both should return FOUND 2 as we have bootstrap and SST
-FOUND 2 /Desyncing and pausing the provider/ in mysqld.1.err
-FOUND 2 /Resuming and resyncing the provider/ in mysqld.1.err
+# Both should return NOT FOUND as we have mariabackup with backup locks
+NOT FOUND /Desyncing and pausing the provider/ in mysqld.1.err
+NOT FOUND /Resuming and resyncing the provider/ in mysqld.1.err
 connection node_1;
 SET GLOBAL wsrep_mode = "BF_ABORT_MARIABACKUP";
 # Restart node_2, force SST.
@@ -25,9 +25,9 @@ connection node_2;
 Starting server ...
 connection node_2;
 connection node_1;
-# Both should return FOUND 3 as we have 1 new SST
-FOUND 3 /Desyncing and pausing the provider/ in mysqld.1.err
-FOUND 3 /Resuming and resyncing the provider/ in mysqld.1.err
+# Both should return NOT FOUND as we have mariabackup with backup locks
+NOT FOUND /Desyncing and pausing the provider/ in mysqld.1.err
+NOT FOUND /Resuming and resyncing the provider/ in mysqld.1.err
 SET GLOBAL wsrep_mode = "";
 DROP TABLE t;
 # Case 2: MariaBackup backup from node_2
@@ -46,11 +46,13 @@ SET GLOBAL wsrep_mode = "BF_ABORT_MARIABACKUP";
 SELECT @@wsrep_mode;
 @@wsrep_mode
 BF_ABORT_MARIABACKUP
-# Both should return FOUND 1 as node should not desync
-FOUND 1 /Desyncing and pausing the provider/ in mysqld.2.err
-FOUND 1 /Resuming and resyncing the provider/ in mysqld.2.err
-# Should return FOUND 1 because only last backup does not desync
-FOUND 1 /Server not desynched from group because WSREP_MODE_BF_MARIABACKUP used./ in mysqld.2.err
+# Both should return FOUND 2 because both backups do desync but on different points
+FOUND 2 /Desyncing and pausing the provider/ in mysqld.2.err
+FOUND 2 /Resuming and resyncing the provider/ in mysqld.2.err
+# Should return FOUND 1 as server did not desync at BLOCK_DDL
+FOUND 1 /Server not desynched from group at BLOCK_DDL because WSREP_MODE_BF_MARIABACKUP is used./ in mysqld.2.err
+# Should return FOUND 1 as server did desync and pause at BLOCK_COMMIT
+FOUND 1 /Server desynched from group during BACKUP STAGE BLOCK_COMMIT./ in mysqld.2.err
 SET GLOBAL wsrep_mode = "";
 connection node_1;
 DROP TABLE t;

mysql-test/suite/galera/t/galera_bf_abort_mariabackup.test

@@ -13,7 +13,7 @@
 CREATE TABLE t(i INT NOT NULL PRIMARY KEY) ENGINE INNODB;
 INSERT INTO t VALUES(1);
 #
-# In default settings donor should desync
+# In default settings donor should not desync
 #
 --echo # Restart node_2, force SST.
 --connection node_2
@@ -37,7 +37,7 @@ let $restart_noprint=2;
 
 --connection node_1
 let SEARCH_FILE = $MYSQLTEST_VARDIR/log/mysqld.1.err;
---echo # Both should return FOUND 2 as we have bootstrap and SST
+--echo # Both should return NOT FOUND as we have mariabackup with backup locks
 let SEARCH_PATTERN = Desyncing and pausing the provider;
 --source include/search_pattern_in_file.inc
 let SEARCH_PATTERN = Resuming and resyncing the provider;
@@ -76,7 +76,7 @@ let $restart_noprint=2;
 
 --connection node_1
 let SEARCH_FILE = $MYSQLTEST_VARDIR/log/mysqld.1.err;
---echo # Both should return FOUND 3 as we have 1 new SST
+--echo # Both should return NOT FOUND as we have mariabackup with backup locks
 let SEARCH_PATTERN = Desyncing and pausing the provider;
 --source include/search_pattern_in_file.inc
 let SEARCH_PATTERN = Resuming and resyncing the provider;
@@ -117,13 +117,16 @@ let $targetdir=$MYSQLTEST_VARDIR/tmp/backup2;
 --enable_result_log
 
 let SEARCH_FILE = $MYSQLTEST_VARDIR/log/mysqld.2.err;
---echo # Both should return FOUND 1 as node should not desync
+--echo # Both should return FOUND 2 because both backups do desync but on different points
 let SEARCH_PATTERN = Desyncing and pausing the provider;
 --source include/search_pattern_in_file.inc
 let SEARCH_PATTERN = Resuming and resyncing the provider;
 --source include/search_pattern_in_file.inc
---echo # Should return FOUND 1 because only last backup does not desync
-let SEARCH_PATTERN = Server not desynched from group because WSREP_MODE_BF_MARIABACKUP used.;
+--echo # Should return FOUND 1 as server did not desync at BLOCK_DDL
+let SEARCH_PATTERN = Server not desynched from group at BLOCK_DDL because WSREP_MODE_BF_MARIABACKUP is used.;
+--source include/search_pattern_in_file.inc
+--echo # Should return FOUND 1 as server did desync and pause at BLOCK_COMMIT
+let SEARCH_PATTERN = Server desynched from group during BACKUP STAGE BLOCK_COMMIT.;
 --source include/search_pattern_in_file.inc
 
 SET GLOBAL wsrep_mode = "";

sql/backup.cc

@@ -39,6 +39,7 @@
 #ifdef WITH_WSREP
 #include "wsrep_server_state.h"
 #include "wsrep_mysqld.h"
+#include "wsrep_sst.h"
 #endif /* WITH_WSREP */
 
 static const char *stage_names[]=
@@ -293,29 +294,40 @@ static bool backup_block_ddl(THD *thd)
 
 #ifdef WITH_WSREP
   DBUG_ASSERT(thd->wsrep_desynced_backup_stage == false);
-  /*
-    if user is specifically choosing to allow BF aborting for BACKUP STAGE BLOCK_DDL lock
-    holder, then do not desync and pause the node from cluster replication.
-    e.g. mariabackup uses BACKUP STATE BLOCK_DDL; and will be abortable by this.
-    But, If node is processing as SST donor or WSREP_MODE_BF_MARIABACKUP mode is not set,
-    we desync the node for BACKUP STAGE because applier threads
-    bypass backup MDL locks (see MDL_lock::can_grant_lock)
-  */
   if (WSREP_NNULL(thd))
   {
     Wsrep_server_state &server_state= Wsrep_server_state::instance();
 
-    if (!wsrep_check_mode(WSREP_MODE_BF_MARIABACKUP) ||
-        server_state.state() == Wsrep_server_state::s_donor)
+    /*
+      If user is specifically choosing to allow BF aborting for
+      BACKUP STAGE BLOCK_DDL lock holder, then do not desync and
+      pause the node from cluster replication. e.g. mariabackup
+      uses BACKUP STATE BLOCK_DDL; and will be abortable by this.
+    */
+    bool mariabackup= (server_state.state() == Wsrep_server_state::s_donor
+                       && !strcmp(wsrep_sst_method, "mariabackup"));
+    bool allow_bf= wsrep_check_mode(WSREP_MODE_BF_MARIABACKUP);
+    bool pause_and_desync= true;
+
+    if ((allow_bf) || (mariabackup))
+    {
+      pause_and_desync= false;
+    }
+
+    if (pause_and_desync)
     {
-      if (server_state.desync_and_pause().is_undefined()) {
+      if (server_state.desync_and_pause().is_undefined())
         DBUG_RETURN(1);
-      }
+
+      WSREP_INFO("Server desynched from group during BACKUP STAGE BLOCK_DDL.");
       DEBUG_SYNC(thd, "wsrep_backup_stage_after_desync_and_pause");
       thd->wsrep_desynced_backup_stage= true;
     }
     else
-      WSREP_INFO("Server not desynched from group because WSREP_MODE_BF_MARIABACKUP used.");
+    {
+      WSREP_INFO("Server not desynched from group at BLOCK_DDL because %s is used.",
+                 allow_bf ? "WSREP_MODE_BF_MARIABACKUP" : wsrep_sst_method);
+    }
   }
 #endif /* WITH_WSREP */
 
@@ -399,6 +411,28 @@ static bool backup_block_commit(THD *thd)
   }
   thd->clear_error();
 
+#ifdef WITH_WSREP
+  if (WSREP_NNULL(thd) && !thd->wsrep_desynced_backup_stage)
+  {
+    Wsrep_server_state &server_state= Wsrep_server_state::instance();
+    bool mariabackup= (server_state.state() == Wsrep_server_state::s_donor
+                       && !strcmp(wsrep_sst_method, "mariabackup"));
+
+    /* If this node is donor and mariabackup is not used
+       we desync and pause provider here if it is not yet done.
+    */
+    if (!mariabackup)
+    {
+      if (server_state.desync_and_pause().is_undefined())
+        DBUG_RETURN(1);
+
+      WSREP_INFO("Server desynched from group during BACKUP STAGE BLOCK_COMMIT.");
+      thd->wsrep_desynced_backup_stage= true;
+      DEBUG_SYNC(thd, "wsrep_backup_stage_commit_after_desync_and_pause");
+    }
+  }
+#endif /* WITH_WSREP */
+
   DBUG_RETURN(0);
 }