JdbcLockRegistry should retry on DeadlockLoserDataAccessException

[brempusz]

I am using LockRegistryLeaderInitiator with JdbcLockRegistry. It worked fine on 5.6.36 but after the migration to 5.7.19 I noticed flood of "leader granted/revoked" events when more than on instance of my service is running. After some debugging I noticed it is caused by DeadlockLoserDataAccessException not handled as "retry" in JdbcLockRegistry.

MySql 5.7.15 introduced setting innodb_deadlock_detect (enabled by default, see https://dev.mysql.com/doc/refman/5.7/en/innodb-parameters.html#sysvar_innodb_deadlock_detect).

When this setting is enabled exception will be thrown from MySql JDBC driver when deadlock is detected while accessing table row (vendorCode=1213, SQLState=40001, message: Deadlock found when trying to get lock; try restarting transaction). It is then wrapped to spring exception DeadlockLoserDataAccessException.

JdbcLockRegistry does not handle DeadlockLoserDataAccessException with "try again" action (as it does for CannotSerializeTransactionException, TransactionTimedOutException, QueryTimeoutException).

see

spring-integration/spring-integration-jdbc/src/main/java/org/springframework/integration/jdbc/lock/JdbcLockRegistry.java

Lines 218 to 230 in a2f3dbe

	catch (CannotSerializeTransactionException e) {
	// try again
	}
	catch (TransactionTimedOutException e) {
	// try again
	}
	catch (QueryTimeoutException e) {
	// try again
	}
	catch (Exception e) {
	this.delegate.unlock();
	rethrowAsLockException(e);
	}

Spring Boot 1.5.13.RELEASE
Spring Integration JDBC 4.3.16.RELEASE
MySQL 5.6.36/5.7.19

[artembilan]

No objections to have such a fix.
Any chances that you can Contribute?

Thanks

[brempusz]

Sure. Some questions first:

1. Do I have to create issue in JIRA (I don't have an account yet) or just continue with this GH issue?
2. Do I have to create PR for master and 4.3.x branch or just for one branch?

[garyrussell]

Thanks; you don't need a JIRA; generally, just a PR against master is all that is needed and we will cherry-pick. If the cherry-pick is not trivial, we may ask for a separate PR at that time, but that is rare.

[brempusz]

This can't be cherry-picked directly: There is only one catch(Ex1 | Ex2 | Ex3 e) in master but 4.3.x source is 1.6 with multiple catch() clauses. Just let me know and I will create PR to 4.3.x.