# ==== Purpose ====
# Prove successful slave service recovery after a simulated failure
# at handling recoverable DDL queries.
# The recoverable DDL:s are those that are binary-logged with xid info
# with enables the server to decide on their commit or rollback at server
# recovery.
#
# ==== Implementation ====
# In the test the slave applier processes a sequence of recoverable DDL:s
# whose applying  is interruped by a few simulated failures.
# When a failure is an ordinary error slave stops, the failed DDL
# group is retried to success, provided the failure condition was
# withdrawn.
# The crash error is sorted out depending on its timing.
# When it happens at time before the slave applier has logged the DDL
# group into its binary log, first the server recovery rolls back and then
# the slave service restarts at the beginning of the crashed DDL's group.
# When the applier has logged and yet not committed the DDL statement,
# the slave server recovery completes with commit and the slave service
# resumes at the next group of events following the crashed DDL's group.
#
# The test requires the TABLE slave info repository type through
# the opt file and --log-bin through a have macro
# because of crash-restart actions on the slave side.
#
# ==== References ====
# WL#9175  Correct recovery of DDL statements/transactions by binary log
#
# Params:
# --let $rpl_atomic_ddl_print_verbose # get out more info
#

--let $rpl_atomic_ddl_print_verbose=1

--source include/big_test.inc
--source include/not_crashrep.inc
--source include/not_valgrind.inc
--source include/have_log_bin.inc
--source include/have_debug.inc
--source include/have_binlog_format_mixed.inc
--source include/have_udf.inc
--let $rpl_gtid_utils= 1
--source include/master-slave.inc

--source include/rpl_connection_slave.inc

# Simulated data inconsistency on slave causes few errors
call mtr.add_suppression("Operation CREATE USER failed");
call mtr.add_suppression("Operation RENAME USER failed");
call mtr.add_suppression("The slave coordinator and worker threads are stopped");
call mtr.add_suppression("Slave worker thread has failed to apply an event");
call mtr.add_suppression("Error writing relay log configuration");
call mtr.add_suppression("Error 'Unknown error' on query");
--disable_query_log
call mtr.add_suppression("You need to use --log-bin to make.*work");
--enable_query_log

--source include/rpl_connection_master.inc

# instructing gtid_step_assert that is called by a sourced file
# to satisfy to gtid-mode OFF as well.
--let $gtid_mode_on= `SELECT @@GLOBAL.GTID_MODE = 'ON'`
--let $gtid_mode= `SELECT @@GLOBAL.GTID_MODE`
--let $gtid_step_gtid_mode_agnostic=`SELECT '$gtid_mode' != 'ON'`

if (!$crash_commit)
{
  --let $commit_message=Crash right after the query has been binary-logged before committed in the engine
  --let $crash_commit="+d,crash_commit_after_log"
}
if (!$crash_rollback)
{
  --let $rollback_message=Crash right after the query has been prepared in the engine before being logged
  --let $crash_rollback="+d,crash_commit_before_log"
}

#
# The overall logics of the following part of the tests boils down to
# this pseudo-code block:
#
# Run load generator on master to prepare binlog events;
# For a number of simulated of failures
#    turn ON simulator,
#    start the slave,
#    wait for an expected slave error,
#    turn OFF the current simulator
#  end for.
#
# Errors due to Interruptive actions described below are meaningful in
# the context of the sourced load generator.
# In  each loop the server crashes with roll-back and roll-forward following recovery
# which is asserted.

# Master load
#
# Start off with the master side logger table (see the master opt
# file) that record DDL queries that will be subject to crash
# simulation on the slave side.  The table will hold prescribed
# post-recovery checks to be performed on the slave side.
#
--let $master_log_table=t_checks
--let $master_log_db=master_db
--eval CREATE DATABASE $master_log_db

--let $save_curr_db=`SELECT database()`
--eval USE $master_log_db
--eval CREATE TABLE IF NOT EXISTS $master_log_table (id INT AUTO_INCREMENT NOT NULL PRIMARY KEY,  ddl_query TEXT,  pre_binlog_check TEXT, post_binlog_check TEXT);

--eval USE $save_curr_db

# required for the extra section of the being included binlog_ddl.inc
--let $table=t_1
--eval CREATE TABLE $table (a int)
--source include/sync_slave_sql_with_master.inc

--source include/stop_slave_sql.inc

--source include/rpl_connection_master.inc

--let $do_post_prepare=0
--let $do_post_binlog=0
--let $do_only_regular_logging=1
--let $do_show_binlog_events= 0
--let $do_count_queries=1
--let $count_ddl_queries=0
--source extra/binlog_tests/binlog_ddl.inc

--source include/rpl_connection_slave.inc
--source include/start_slave_io.inc

--source include/rpl_connection_master.inc
--source include/sync_slave_io_with_master.inc

--source include/rpl_connection_slave.inc
# First, a regular error
CREATE USER user1;
START SLAVE SQL_THREAD;
--let $slave_sql_errno=convert_error(ER_CANNOT_USER)
--source include/wait_for_slave_sql_error.inc
DROP USER user1;

#
# Simulated error in pre_commit hook stops the slave applier.
#
SET @@GLOBAL.debug="+d,rli_pre_commit_error";
START SLAVE SQL_THREAD;
--let $slave_sql_errno=convert_error(ER_UNKNOWN_ERROR)
--source include/wait_for_slave_sql_error.inc
#
# Like in above, the fact that there has been no advance of the last
# executed coordinates is proved later by successful resuming/recovery
# from the error.
# The error simulation is lifted:
SET @@GLOBAL.debug="-d,rli_pre_commit_error";

# The rest is crash-restart in loop of number of DDL queries
# in the master load.
--let $loops=$count_ddl_queries
--let $id=0
while ($loops)
{
  --inc $id

  # Find the loop's query and it post-recovery checks upon the first recovery
  --source include/rpl_connection_master.inc

  --let $current_query=`SELECT ddl_query FROM $master_log_db.$master_log_table WHERE id = $id `
  --let $pre_binlog_check=`SELECT pre_binlog_check FROM $master_log_db.$master_log_table WHERE id = $id`
  --let $post_binlog_check=`SELECT post_binlog_check FROM $master_log_db.$master_log_table WHERE id = $id`
  if ($rpl_atomic_ddl_print_verbose)
  {
  # 
  --let gen_current_query = `SELECT REPLACE("$current_query",'.so','_LIB')`
  --let gen_current_query = `SELECT REPLACE("$gen_current_query",'.dll','_LIB')`
  --echo Loop: $id out of $count_ddl_queries; current query: '$gen_current_query'; pre-binlog check: '$pre_binlog_check'; pre-commit check: '$post_binlog_check';
  }
  --source include/rpl_connection_slave.inc

  #
  # The rollback recovery upon the pre-binlog crash
  #
  --source include/expect_crash.inc
  --echo *** Query: '$gen_current_query'. $rollback_message. ***
  --eval SET @@GLOBAL.debug=$crash_rollback

  # Memorizing execution status right before "scheduled" failure
  --source include/gtid_step_reset.inc
  --error 0,2013
  START SLAVE SQL_THREAD;

  # Server wait for stop and restart
  --enable_reconnect
  --source include/wait_until_disconnected.inc
  --let $rpl_server_number= 2
  --source include/rpl_start_server.inc
  --disable_reconnect

  # Proof of the rollback recovery is no exec progress
  --let $gtid_step_count=0
  --source include/gtid_step_assert.inc

  # Proof by the master side post-recovery checks
  if ($pre_binlog_check)
  {
    if (!`$pre_binlog_check`)
    {
        --echo *** State check upon recovery after the pre-binlog crash fails ***
        --die
    }
  }

  #
  # Commit recovery upon the pre-commit crash.
  #
  --source include/expect_crash.inc
  --echo *** Query: '$gen_current_query'. $commit_message. ***
  --eval SET @@GLOBAL.debug=$crash_commit
  --source include/gtid_step_reset.inc
  --error 0,2013
  START SLAVE SQL_THREAD;

  # Server wait for stop and restart
  --enable_reconnect
  --source include/wait_until_disconnected.inc
  --let $rpl_server_number= 2
  --source include/rpl_start_server.inc
  --disable_reconnect

  --source include/rpl_connection_slave.inc
  # Proof of the commit recovery is one committed gtid
  --let $gtid_step_only_count=1
  --let $gtid_step_count=1
  --source include/gtid_step_assert.inc

  # Proof by the master side post-recovery checks
  if ($post_binlog_check)
  {
    if (!`$post_binlog_check`)
    {
       --echo *** State check upon recovery after the pre-commit crash fails ***
        --die
    }
  }

  --dec $loops
} # end of while

# Eventually:

--source include/rpl_connection_slave.inc
--source include/start_slave.inc

--source include/rpl_connection_master.inc
--eval DROP table $table
--eval DROP DATABASE $master_log_db

--source include/sync_slave_sql_with_master.inc


--source include/rpl_end.inc