galera_toi_drop_database test case fails occasionally

General

Escalation

General

Escalation

Description

There is a timing problem with the test. It starts a long insert operation (or tries to), and then runs a drop database on another node to abort the insert (actually two inserts).

The test will fail occasionally because the insert may actually succeed before the drop database (this can happen when the nodes are loaded so the second node takes a while to actually run).

The solution is to move the test over to using debug_sync, where we can block the insert.

Environment

None

Smart Checklist

Activity

Show:

KennT April 4, 2018 at 1:40 PM

commit 2104f8aa3ef28028ad7d946b5eb6846d3cd00f1a
Author: Kenn Takara <kenn.takara@percona.com>
Date: Tue Apr 3 22:05:30 2018 -0700

: Crash during rollback of a stored procedure called in an event
: BF aborted event does not update wsrep_local_bf_aborts variable

Issue:
PXC was not setting up the proper thread state for the event worker thread.
Thus, various WSREP-related actions were not being performed correctly
(such as rollback and the bf abort counter).

Solution:
Setup the event worker thread state to function more like a normal
local thread (not a replication thread).
Also, call the proper rollback thread if the thread is aborted.