MDEV-38907 Optimistic Relay Log Crash Recovery

mysql-test/suite/rpl/r/auto_relay_log_recovery.result

@@ -0,0 +1,69 @@
+include/master-slave.inc
+[connection master]
+# Case 0: No event groups and no trailing incomplete event
+connection slave;
+# Case 1: No event groups but trailing incomplete event
+include/stop_slave_sql.inc
+SET @@GLOBAL.debug_dbug= '+d,fail_rotate_mid_write';
+connection master;
+FLUSH BINARY LOGS;
+connection slave;
+include/wait_for_slave_io_error.inc [errno=1595]
+include/rpl_restart_server.inc [server_number=2]
+include/show_events.inc
+Log_name	Pos	Event_type	Server_id	End_log_pos	Info
+slave-relay-bin.000002	#	Rotate	#	#	master-bin.000001;pos=POS
+slave-relay-bin.000002	#	Format_desc	#	#	SERVER_VERSION, BINLOG_VERSION
+slave-relay-bin.000002	#	Gtid_list	#	#	[]
+slave-relay-bin.000002	#	Binlog_checkpoint	#	#	master-bin.000001
+# Case 2: Complete event group but trailing incomplete event
+include/start_slave.inc
+include/stop_slave_sql.inc
+SET @@GLOBAL.debug_dbug= '+d,fail_rotate_mid_write';
+connection master;
+CREATE TABLE t (a TEXT) ENGINE=innodb;
+FLUSH BINARY LOGS;
+connection slave;
+include/wait_for_slave_io_error.inc [errno=1595]
+include/rpl_restart_server.inc [server_number=2]
+include/show_events.inc
+Log_name	Pos	Event_type	Server_id	End_log_pos	Info
+slave-relay-bin.000002	#	Gtid	#	#	GTID #-#-#
+slave-relay-bin.000002	#	Query	#	#	use `test`; CREATE TABLE t (a TEXT) ENGINE=innodb
+# Case 3: Trailing incomplete event group AND trailing incomplete event
+SET @@max_relay_log_size=4096;
+include/start_slave.inc
+include/stop_slave_sql.inc
+SET @@GLOBAL.debug_dbug= '+d,fail_rotate_mid_write';
+connection master;
+BEGIN;
+INSERT INTO t VALUES (REPEAT('3', 4096));
+INSERT INTO t VALUES ('3');
+COMMIT;
+connection slave;
+include/wait_for_slave_io_error.inc [errno=1595]
+include/rpl_restart_server.inc [server_number=2]
+include/show_events.inc
+# Case 4: Trailing incomplete event group but complete event
+include/start_slave.inc
+include/stop_slave_sql.inc
+SET @@GLOBAL.debug_dbug= '+d,kill_slave_io_before_commit';
+connection master;
+INSERT INTO t VALUES ('4');
+connection slave;
+include/wait_for_slave_io_error.inc [errno=1595]
+include/rpl_restart_server.inc [server_number=2]
+include/show_events.inc
+# Double-check that all INSERTs have applied
+include/start_slave.inc
+SELECT LEFT(a, 2) FROM t;
+LEFT(a, 2)
+33
+3
+4
+# Case NaN: Cleanup
+connection master;
+DROP TABLE t;
+CALL mtr.add_suppression('Could not use ');
+CALL mtr.add_suppression('Slave I/O: Relay log write failure: ');
+include/rpl_end.inc

mysql-test/suite/rpl/t/auto_relay_log_recovery.test

@@ -0,0 +1,109 @@
+# MDEV-38907: Relay Logs can now automatically remove
+# an incomplete event or event group left by a crash.
+#
+# Each case must first sync the slave up
+# so it does not inject failures at the previous case's unsynchronized section.
+# Due to START SLAVE restarting relay logs in GTID mode (MDEV-4698),
+# it must also show truncation results directly via
+# SHOW RELAYLOG EVENTS before the next case's START SLAVE and sync.
+# (Due to specifying `$binlog_start`, the listing may be empty;
+#  this is expected when truncating a trailing incomplete event group.)
+
+--source include/have_debug.inc
+--source include/have_innodb.inc
+# format-agnostic, but easier to test rotate crashes with the row format
+--source include/have_binlog_format_row.inc
+--source include/master-slave.inc
+
+--echo # Case 0: No event groups and no trailing incomplete event
+--sync_slave_with_master
+
+
+--echo # Case 1: No event groups but trailing incomplete event
+--source include/stop_slave_sql.inc
+--let $binlog_file= `SELECT Relay_Log_File FROM information_schema.slave_status`
+
+SET @@GLOBAL.debug_dbug= '+d,fail_rotate_mid_write';
+--connection master
+  FLUSH BINARY LOGS;
+  --save_master_pos
+--connection slave
+--let $slave_io_errno= 1595
+--source include/wait_for_slave_io_error.inc
+
+--let $rpl_server_number= 2
+--source include/rpl_restart_server.inc
+--source include/show_relaylog_events.inc
+
+
+--echo # Case 2: Complete event group but trailing incomplete event
+--source include/start_slave.inc
+--sync_with_master
+--source include/stop_slave_sql.inc
+--let $binlog_file= `SELECT Relay_Log_File FROM information_schema.slave_status`
+--let $binlog_start= `SELECT Relay_Log_Pos FROM information_schema.slave_status`
+
+SET @@GLOBAL.debug_dbug= '+d,fail_rotate_mid_write';
+--connection master
+  CREATE TABLE t (a TEXT) ENGINE=innodb;
+  FLUSH BINARY LOGS;
+  --save_master_pos
+--connection slave
+--source include/wait_for_slave_io_error.inc
+
+--source include/rpl_restart_server.inc
+--source include/show_relaylog_events.inc
+
+
+--echo # Case 3: Trailing incomplete event group AND trailing incomplete event
+SET @@max_relay_log_size=4096;
+--source include/start_slave.inc
+--sync_with_master
+--source include/stop_slave_sql.inc
+--let $binlog_file= `SELECT Relay_Log_File FROM information_schema.slave_status`
+--let $binlog_start= `SELECT Relay_Log_Pos FROM information_schema.slave_status`
+
+SET @@GLOBAL.debug_dbug= '+d,fail_rotate_mid_write';
+--connection master
+  BEGIN;
+    INSERT INTO t VALUES (REPEAT('3', 4096)); # slave's `@@max_relay_log_size`
+    INSERT INTO t VALUES ('3');
+  COMMIT;
+  --save_master_pos
+--connection slave
+--source include/wait_for_slave_io_error.inc
+
+--source include/rpl_restart_server.inc
+--source include/show_relaylog_events.inc
+
+
+--echo # Case 4: Trailing incomplete event group but complete event
+--source include/start_slave.inc
+--sync_with_master
+--source include/stop_slave_sql.inc
+--let $binlog_file= `SELECT Relay_Log_File FROM information_schema.slave_status`
+--let $binlog_start= `SELECT Relay_Log_Pos FROM information_schema.slave_status`
+
+SET @@GLOBAL.debug_dbug= '+d,kill_slave_io_before_commit';
+--connection master
+  INSERT INTO t VALUES ('4');
+  --save_master_pos
+--connection slave
+--source include/wait_for_slave_io_error.inc
+
+--source include/rpl_restart_server.inc
+--source include/show_relaylog_events.inc
+
+
+--echo # Double-check that all INSERTs have applied
+--source include/start_slave.inc
+--sync_with_master
+SELECT LEFT(a, 2) FROM t;
+
+--echo # Case NaN: Cleanup
+--connection master
+  DROP TABLE t;
+  CALL mtr.add_suppression('Could not use ');
+  CALL mtr.add_suppression('Slave I/O: Relay log write failure: ');
+
+--source include/rpl_end.inc