Bug #20294 Instance manager tests fail randomly
Submitted: 6 Jun 2006 14:24 Modified: 5 Oct 2007 15:16
Reporter: Michael Widenius Email Updates:
Status: Unsupported Impact on me:
None 
Category:Instance Manager Severity:S1 (Critical)
Version:5.0,5.1.12 OS:Any
Assigned to: CPU Architecture:Any
Tags: rt_q1_2007

[6 Jun 2006 14:24] Michael Widenius
Description:
From time to time, im_life_cycle fails and when it fails, a lot of other test also starts failing

(In my case index_merge index_merge_bdb index_merge_innodb index_merge_innodb2 index_merge_ror index_merge_ror_cpk information_schema information_schema_chmod information_schema_db information_schema_inno information_schema_par)

How to repeat:
make test

Suggested fix:
[6 Jun 2006 14:30] Michael Widenius
Sorry, meant im_instance_conf, not im_life_cycle
[14 Jun 2006 18:40] Alexander Nozdrin
The problem here is not only with im_instance_conf.imtest itself.
There are race conditions in IM tests and all tests suffer from them.
This seems to be the same bug as BUG#18023 and BUG#19391.

The other tests fail because the race condition leads to having
mysqld-instance, which is "lost" for test driver (mysql-test-run.pl).
The driver tries to start another mysqld, but failes because TCP-port
is bound.

I've put it in progress, since there are patches for 18023 and 19391.
After they will be pushed, I will request re-verification of this bug.
[20 Jun 2006 22:13] Alexander Nozdrin
Patches for BUG#18023 and BUG#19391 have been pushed.
This bug should be reverified.
[21 Jun 2006 10:29] Alexander Nozdrin
The problem now is the following: im_instance_conf.imtest
expects that no other test suites are running at the same
time.

The problem is that a new mysqld-instance with default
parameters is created in im_instance_conf.imtest.
As parameters are by defaults, TCP port and socket-file-name
are not specified, so TCP port is 3306 (which is used by other
running test-suites).

The test should be updated.
[21 Jun 2006 12:02] Bugs System
A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/8015
[22 Jun 2006 14:56] Bugs System
A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/8083
[27 Jun 2006 8:57] Alexander Nozdrin
Merged into 5.1 tree, currently tagged 5.1.12-beta.
[6 Jul 2006 22:41] Paul DuBois
Changes are to test case only.
No changelog entry needed.
[23 Jul 2006 23:33] Stewart Smith
Still occurs.

Description:
When running im_options, it will sometimes fail with:

-------------------------------------------------------
*** r/im_options.result 2006-06-28 11:13:44.000000000 +0300
--- r/im_options.reject 2006-07-24 02:22:03.316783750 +0300
***************
*** 7,13 ****
  server_id     1
  SHOW INSTANCES;
  instance_name state
! mysqld1       starting
  mysqld2       offline
  UNSET mysqld1.server_id;
  ERROR HY000: The instance is active. Stop the instance first
--- 7,13 ----
  server_id     1
  SHOW INSTANCES;
  instance_name state
! mysqld1       online
  mysqld2       offline
  UNSET mysqld1.server_id;
  ERROR HY000: The instance is active. Stop the instance first
-------------------------------------------------------

If the machine is more loaded, the test will pass. I assume this is because it
takes longer for mysqld1 to start

How to repeat:
run im_options test when machine is idle.
run im_options test when machine is more loaded

Suggested fix:
allow starting or online in results, or wait for started
[23 Jul 2006 23:38] Bugs System
A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/9465
[24 Jul 2006 6:11] Stewart Smith
my patch disables the test, doesn't fix anything.
[20 Mar 2007 10:59] Jani Tolonen
Comment from Alexander:

The previous comment was wrong, this is a bug and will be fixed later.

Estimate Q2/2007.
[5 Oct 2007 15:16] Konstantin Osipov
We do not plan to fix this.