Mysql test suite tuning

[Seppo Jaakola]

wsrep patched mysqld does not pass all mysql test suite tests, when run in source tree with no replication provider assigned.
In principle. when running with no replication provider, mysqld should behave as native mysql server. However, wsrep patch adds a number of wsrep options and status variables and their mere presence seems to make some tests to fail.

The purpose of this bug task is to find all mysql test suite tests which currently fail for a safe wsrep behavior difference and modify expected result accordingly so that the test will pass in the future.

How to reproduce:
in the source tree, mysql-test directory, run:

./mtr

[Vadim Tkachenko (vadim-tk)]

I run ./mysql-test-run.pl --force
and also see a lot of test failures.

One of tests
main.merge crashes mysqld.

[Seppo Jaakola (seppo-jaakola)]

Here is a simple way to reproduce the main.merge test failure:

DROP TABLE IF EXISTS t1, t2, m1, m2;
CREATE TABLE t1 (c1 INT, c2 INT) ENGINE=MyISAM;
CREATE TABLE t2 (c1 INT, c2 INT) ENGINE=MyISAM;
CREATE TEMPORARY TABLE m1 (c1 INT, c2 INT) ENGINE=MRG_MyISAM UNION=(t1,t2)
  INSERT_METHOD=LAST;

CREATE TABLE m2 LIKE m1;

Test succeeds, if m1 is created like this:
CREATE TABLE m1 (c1 INT, c2 INT) ENGINE=MRG_MyISAM UNION=(t1,t2)
  INSERT_METHOD=LAST;

=> temporary tables are not handled correctly

[Seppo Jaakola (seppo-jaakola)]

looks like the main.merge makes a native MySQL bug to surface.
 wsrep patch can be made to work like native MySQl with an easy fix, and the test passes, underlying bug remains though...

[Seppo Jaakola (seppo-jaakola)]

./mtr --force completes now with status:

The servers were restarted 369 times
Spent 7337.783 of 9138 seconds executing testcases

Completed: Failed 4/1662 tests, 99.76% were successful.

Failing test(s): sys_vars.all_vars main.mysqld--help-notwin main.sp-threads perfschema.dml_setup_instruments

This result is from my private branch, push to lp:codership-mysql/5.5 will happen after release 1.1 build.

The failing tests seem to be due to excessive wsrep status and variables and differences in processlist output. For wsrep variable differences, I will modify the test results, as hiding wsrep variables would not be feasible, we need to be able to promote a MySQL server to a wsrep replication node by setting the wsrep_provider.

[Seppo Jaakola (seppo-jaakola)]

FIx for main.merge pushed in: http://bazaar.launchpad.net/~codership/codership-mysql/wsrep-5.5/revision/3655

[Philip Stoev (philip-stoev-f)]

MTR test fixes around wsrep_ variables have been pushed to 5.6 . The tests pass there to the best of my understanding.