MDEV-31517 Wrong variable name in the configuration leads Galera to think SST/IST failed, at next restart will request a full SST

mysql-test/suite/galera/r/galera_mdev_31517.result

@@ -0,0 +1,24 @@
+connection node_2;
+connection node_1;
+connection node_2;
+CALL mtr.add_suppression("unknown variable 'non_existing_variable=ON'");
+CALL mtr.add_suppression("Aborting");
+CALL mtr.add_suppression("sst_received failed: State wait was interrupted");
+CALL mtr.add_suppression("State transfer interrupted, shutting down gracefully");
+connection node_1;
+CREATE TABLE t(i INT NOT NULL PRIMARY KEY) ENGINE INNODB;
+INSERT INTO t VALUES(1);
+connection node_2;
+connection node_1;
+connection node_2;
+connection node_1;
+connection node_2;
+connection node_1;
+connection node_2;
+Starting server ...
+Starting server ...
+SET GLOBAL wsrep_mode = DEFAULT;
+connection node_1;
+DROP TABLE t;
+disconnect node_2;
+disconnect node_1;

mysql-test/suite/galera/t/galera_mdev_31517.test

@@ -0,0 +1,129 @@
+#
+# Test for MDEV-31517: Wrong variable name in the configuration leads
+# Galera to think SST/IST failed, at next restart will request a full
+# SST.
+#
+# To reproduce:
+#
+#     1. Start Galera cluster
+#     2. Stop a Node
+#     3. Start the node
+#     4. Stop a Node
+#     5. Add non_existing_variable=ON in the config
+#     6. Start the node, this will fail
+#     7  Remove non_existing_variable=ON from the config file
+#     8  Restart the server
+#     9. Observe a full SST happening
+#
+# This test checks that an IST takes place at Step 9 instead of a full SST.
+
+# Step 1: Start Galera cluster
+--source include/galera_cluster.inc
+--source include/have_mariabackup.inc
+
+# Suppress expected errors and warnings:
+--connection node_2
+CALL mtr.add_suppression("unknown variable 'non_existing_variable=ON'");
+CALL mtr.add_suppression("Aborting");
+CALL mtr.add_suppression("sst_received failed: State wait was interrupted");
+CALL mtr.add_suppression("State transfer interrupted, shutting down gracefully");
+
+# Count the number of "SST completed" messages in the log file before
+# and after testing. To do this we need to save original log file
+# before testing:
+#
+--let TEST_LOG=$MYSQLTEST_VARDIR/log/mysqld.2.err
+--perl
+   use strict;
+   my $test_log=$ENV{'TEST_LOG'} or die "TEST_LOG not set";
+   my $test_log_copy=$test_log . '.copy';
+   if (-e $test_log_copy) {
+      unlink $test_log_copy;
+   }
+EOF
+--copy_file $TEST_LOG $TEST_LOG.copy
+
+--connection node_1
+CREATE TABLE t(i INT NOT NULL PRIMARY KEY) ENGINE INNODB;
+INSERT INTO t VALUES(1);
+
+# Step 2: Stop node 2
+--connection node_2
+--source include/shutdown_mysqld.inc
+
+--connection node_1
+--let $wait_condition = SELECT VARIABLE_VALUE = 1 FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+--source include/wait_condition.inc
+
+# Step 3: Start node 2
+--connection node_2
+--source include/start_mysqld.inc
+
+--connection node_1
+--let $wait_condition = SELECT VARIABLE_VALUE = 2 FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+--source include/wait_condition.inc
+
+# Step 4: Stop node 2
+--connection node_2
+let $MYSQLD_DATADIR= `SELECT @@datadir`;
+--source include/shutdown_mysqld.inc
+
+--connection node_1
+--let $wait_condition = SELECT VARIABLE_VALUE = 1 FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+--source include/wait_condition.inc
+
+# Step 5: Add non_existing_variable=ON in the config
+--exec cp  $MYSQLTEST_VARDIR/my.cnf $MYSQLTEST_VARDIR/my.cnf-orig
+--exec echo '[mysqld.2]' >> $MYSQLTEST_VARDIR/my.cnf
+--exec echo 'non_existing_variable=ON' >> $MYSQLTEST_VARDIR/my.cnf
+
+# Step 6: start the stopped node, this will fail
+--connection node_2
+--exec cp -p $MYSQLD_DATADIR/grastate.dat $MYSQLD_DATADIR/grastate.dat_before
+--echo Starting server ...
+--error 1
+--exec $MYSQLD --defaults-group-suffix=.2 --defaults-file=$MYSQLTEST_VARDIR/my.cnf | grep 'non_existing_variable'
+--exec cp -p $MYSQLD_DATADIR/grastate.dat $MYSQLD_DATADIR/grastate.dat_after
+
+# Step 7: remove the wrong variable in the config file
+--exec cp  $MYSQLTEST_VARDIR/my.cnf-orig $MYSQLTEST_VARDIR/my.cnf
+
+# Step 8: Start the node
+--echo Starting server ...
+let $restart_noprint=2;
+--source include/start_mysqld.inc
+
+--let $wait_condition = SELECT VARIABLE_VALUE = 2 FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_cluster_size';
+--source include/wait_condition.inc
+
+--let $wait_condition = SELECT VARIABLE_VALUE = 'ON' FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_ready';
+--source include/wait_condition.inc
+
+# cleanup
+SET GLOBAL wsrep_mode = DEFAULT;
+
+--connection node_1
+DROP TABLE t;
+
+# Count the number of "SST completed" messages in the log file during
+# test phase - to print the error message if the number of such
+# messages in log file increased at the end of the test:
+#
+--perl
+   use strict;
+   my $test_log=$ENV{'TEST_LOG'} or die "TEST_LOG not set";
+   my $test_log_copy=$test_log . '.copy';
+   open(FILE, $test_log_copy) or die("Unable to open $test_log_copy: $!\n");
+   my $initial=grep(/SST completed/gi,<FILE>);
+   close(FILE);
+   open(FILE, $test_log) or die("Unable to open $test_log: $!\n");
+   my $final=grep(/SST completed/gi,<FILE>);
+   close(FILE);
+   if ($final != $initial) {
+      my $diff=$final-$initial;
+      print("Full WSREP SST performed $diff times.\n");
+   }
+EOF
+--remove_file $TEST_LOG.copy
+
+--source include/galera_end.inc

sql/mysqld.cc

@@ -5241,8 +5241,14 @@ static int init_server_components()
 #endif
 
     if ((ho_error= handle_options(&remaining_argc, &remaining_argv, removed_opts,
-                                  mysqld_get_one_option)))
+                                  mysqld_get_one_option))) {
+#ifdef WITH_WSREP
+      Wsrep_server_state::instance().disable_node_reset();
+#endif
+
       unireg_abort(ho_error);
+    }
+
     /* Add back the program name handle_options removes */
     remaining_argc++;
     remaining_argv--;

sql/wsrep_client_service.cc

@@ -392,3 +392,9 @@ int Wsrep_client_service::bf_rollback()
 
   DBUG_RETURN(ret);
 }
+
+void Wsrep_client_service::notify_state_change()
+{
+  DBUG_ENTER("Wsrep_client_service::notify_state_change");
+  DBUG_VOID_RETURN;
+}

sql/wsrep_client_service.h

@@ -68,6 +68,7 @@ class Wsrep_client_service : public wsrep::client_service
   void debug_sync(const char*) override;
   void debug_crash(const char*) override;
   int bf_rollback() override;
+  void notify_state_change() override;
 private:
   friend class Wsrep_server_service;
   THD* m_thd;