How to handling workload conflicts on Percona XtraDB Cluster

有天在看 Percona XtraDB Cluster (PXC) 101 簡報的時候
發現 PXC 如何處理 Update race condition 的疑問
索性就來做個實驗，也把看到的 Log 記錄下來分享
條件如下，
1. Generate 1 million rows test data.

CREATE TABLE t1 (
    colume_1 INT NOT NULL AUTO_INCREMENT PRIMARY KEY
)

2. At the same time delete the data and then node1 and node2.

DELETE FROM t1 WHERE 1 LIMIT 500000;

3. node[1,2] mysqld.log will show the procedure.

$ cat mysqld_conflicts.log
*** Priority TRANSACTION:
TRANSACTION 5514, ACTIVE 0 sec starting index read
mysql tables in use 1, locked 1
MySQL thread id 5, OS thread handle 139807762601728, query id 1472 System lock
*** Victim TRANSACTION:
TRANSACTION 5509, ACTIVE 7 sec updating or deleting
mysql tables in use 1, locked 1
, undo log entries 452668
MySQL thread id 65, OS thread handle 139807629027072, query id 1461 node1.localhost 172.19.7.215 username updating
delete from t1 where 1 limit 500000
*** WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 31 page no 760 n bits 792 index PRIMARY of table `test`.`t1` trx id 5509 lock_mode X
2018-01-24T02:52:58.451550Z 5 [Note] WSREP: --------- CONFLICT DETECTED --------
2018-01-24T02:52:58.451605Z 5 [Note] WSREP: cluster conflict due to high priority abort for threads:
2018-01-24T02:52:58.451835Z 5 [Note] WSREP: Winning thread:
   THD: 5, mode: applier, state: executing, conflict: no conflict, seqno: 58
   SQL: (null)
2018-01-24T02:52:58.451929Z 5 [Note] WSREP: Victim thread:
   THD: 65, mode: local, state: executing, conflict: no conflict, seqno: -1
   SQL: delete from t1 where 1 limit 500000

4. Just need to wait & no intervention.
5. Execute complete.
以前可能需要人為介入找到產生 Lock 的Queries，處理 Lock 的 Query ID，等等作業
而實作經驗上，我們也不樂見存取衝突產生的 Lock
所以 PXC 已經節省掉很多事情了
—-
結論怎麼弱成這樣?