github.com/pingcap/tiflow@v0.0.0-20240520035814-5bf52d54e205/dm/tests/duplicate_event/run.sh (about)

     1  #!/bin/bash
     2  
     3  set -eu
     4  
     5  cur=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
     6  source $cur/../_utils/test_prepare
     7  WORK_DIR=$TEST_DIR/$TEST_NAME
     8  
     9  function run_with_prepared_source_config() {
    10  	# 1. test sync fetch binlog met error and reset binlog streamer with remote binlog
    11  
    12  	# with a 5 rows insert txn: 1 * FormatDesc + 1 * PreviousGTID + 1 * GTID + 1 * BEGIN + 5 * (Table_map + Write_rows) + 1 * XID
    13  	# here we fail at the third write rows event, sync should retry and auto recover without any duplicate event
    14  	export GO_FAILPOINTS="github.com/pingcap/tiflow/dm/syncer/GetEventErrorInTxn=return(3)"
    15  
    16  	run_sql_file $cur/data/db1.prepare.sql $MYSQL_HOST1 $MYSQL_PORT1 $MYSQL_PASSWORD1
    17  	check_contains 'Query OK, 2 rows affected'
    18  
    19  	run_dm_master $WORK_DIR/master $MASTER_PORT $cur/conf/dm-master.toml
    20  	check_rpc_alive $cur/../bin/check_master_online 127.0.0.1:$MASTER_PORT
    21  	check_metric $MASTER_PORT 'start_leader_counter' 3 0 2
    22  	run_dm_worker $WORK_DIR/worker1 $WORKER1_PORT $cur/conf/dm-worker1.toml
    23  	check_rpc_alive $cur/../bin/check_worker_online 127.0.0.1:$WORKER1_PORT
    24  
    25  	dmctl_operate_source create $WORK_DIR/source1.yaml $SOURCE_ID1
    26  
    27  	run_dm_ctl $WORK_DIR "127.0.0.1:$MASTER_PORT" \
    28  		"start-task $cur/conf/dm-task.yaml --remove-meta"
    29  	check_metric $WORKER1_PORT "dm_worker_task_state{source_id=\"mysql-replica-01\",task=\"test\",worker=\"worker1\"}" 10 1 3
    30  
    31  	# wait safe-mode pass
    32  	check_log_contain_with_retry "disable safe-mode after task initialization finished" $WORK_DIR/worker1/log/dm-worker.log
    33  
    34  	run_sql_file $cur/data/db1.increment.sql $MYSQL_HOST1 $MYSQL_PORT1 $MYSQL_PASSWORD1
    35  
    36  	check_log_contain_with_retry "reset replication binlog puller" $WORK_DIR/worker1/log/dm-worker.log
    37  	check_log_contain_with_retry "discard event already consumed" $WORK_DIR/worker1/log/dm-worker.log
    38  	check_sync_diff $WORK_DIR $cur/conf/diff_config.toml
    39  
    40  	# 2. test relay log retry relay with GTID
    41  
    42  	# with a 5 rows insert txn: 1 * FormatDesc + 1 * PreviousGTID + 1 * GTID + 1 * BEGIN + 5 * (Table_map + Write_rows) + 1 * XID
    43  	# here we fail at the third write rows event, sync should retry and auto recover without any duplicate event
    44  	export GO_FAILPOINTS="github.com/pingcap/tiflow/dm/relay/RelayGetEventFailedAt=return(3);github.com/pingcap/tiflow/dm/relay/RelayAllowRetry=return"
    45  
    46  	run_dm_worker $WORK_DIR/worker2 $WORKER2_PORT $cur/conf/dm-worker2.toml
    47  	check_rpc_alive $cur/../bin/check_worker_online 127.0.0.1:$WORKER2_PORT
    48  
    49  	dmctl_operate_source create $WORK_DIR/source2.yaml $SOURCE_ID2
    50  
    51  	run_sql_file $cur/data/db1.prepare.sql $MYSQL_HOST2 $MYSQL_PORT2 $MYSQL_PASSWORD2
    52  	check_contains 'Query OK, 2 rows affected'
    53  
    54  	run_dm_ctl $WORK_DIR "127.0.0.1:$MASTER_PORT" \
    55  		"start-task $cur/conf/dm-task-relay.yaml --remove-meta"
    56  	check_metric $WORKER2_PORT "dm_worker_task_state{source_id=\"mysql-replica-02\",task=\"test_relay\",worker=\"worker2\"}" 10 1 3
    57  
    58  	check_sync_diff $WORK_DIR $cur/conf/diff_relay_config.toml
    59  
    60  	run_sql_source2 "flush logs;"
    61  	run_dm_ctl_with_retry $WORK_DIR "127.0.0.1:$MASTER_PORT" \
    62  		"start-relay -s $SOURCE_ID2 worker2" \
    63  		"\"result\": true" 2
    64  	run_dm_ctl_with_retry $WORK_DIR "127.0.0.1:$MASTER_PORT" \
    65  		"query-status -s $SOURCE_ID2" \
    66  		"\"relayCatchUpMaster\": true" 1
    67  
    68  	run_sql_file $cur/data/db1.increment.sql $MYSQL_HOST2 $MYSQL_PORT2 $MYSQL_PASSWORD2
    69  
    70  	check_sync_diff $WORK_DIR $cur/conf/diff_relay_config.toml
    71  
    72  	# check relay log binlog file size is the same as master size
    73  	run_sql_source2 "show master status;"
    74  	binlog_file=$(grep "File" $TEST_DIR/sql_res.$TEST_NAME.txt | awk -F: '{print $2}' | xargs)
    75  	binlog_pos=$(grep "Position" $TEST_DIR/sql_res.$TEST_NAME.txt | awk -F: '{print $2}' | xargs)
    76  	server_uuid=$(tail -n 1 $WORK_DIR/worker2/relay-dir/server-uuid.index)
    77  
    78  	succ=0
    79  	for ((k = 1; k < 6; k++)); do
    80  		relay_log_size=$(ls -al $WORK_DIR/worker2/relay-dir/$server_uuid/$binlog_file | awk '{print $5}')
    81  		echo "binlog_pos: $binlog_pos relay_log_size: $relay_log_size"
    82  		if [[ "$binlog_pos" -eq "$relay_log_size" ]]; then
    83  			succ=1
    84  			break
    85  		fi
    86  		sleep 1
    87  	done
    88  	if [[ $succ -eq 0 ]]; then
    89  		echo "binlog_pos is not equal to relay_log_size"
    90  		exit 1
    91  	fi
    92  
    93  	echo "============== run_with_prepared_source_config success ==================="
    94  }
    95  
    96  function run() {
    97  	# test source enable gtid
    98  	cp $cur/conf/source1.yaml $WORK_DIR/source1.yaml
    99  	cp $cur/conf/source2.yaml $WORK_DIR/source2.yaml
   100  
   101  	run_with_prepared_source_config
   102  
   103  	cleanup_process $*
   104  	cleanup_data dup_event1 dup_event_relay
   105  
   106  	# test source disable gtid
   107  	cp $cur/conf/source1.yaml $WORK_DIR/source1.yaml
   108  	cp $cur/conf/source2.yaml $WORK_DIR/source2.yaml
   109  	sed -i "s/enable-gtid: true/enable-gtid: false/g" $WORK_DIR/source1.yaml
   110  	sed -i "s/enable-gtid: true/enable-gtid: false/g" $WORK_DIR/source2.yaml
   111  
   112  	run_with_prepared_source_config
   113  }
   114  
   115  # also cleanup dm processes in case of last run failed
   116  cleanup_process $*
   117  cleanup_data dup_event1 dup_event_relay
   118  run
   119  cleanup_process $*
   120  
   121  echo "[$(date)] <<<<<< test case $TEST_NAME success! >>>>>>"