MySQL Bugs: #9712: Difference default / PS protocol for lost connection on IRIX + Mac + hpita2

Bug #9712	Difference default / PS protocol for lost connection on IRIX + Mac + hpita2
Submitted:	7 Apr 2005 10:21	Modified:	9 Jan 2006 9:07
Reporter:	Joerg Bruehe	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server	Severity:	S3 (Non-critical)
Version:	5.0.4 and up	OS:	Other (IRIX (32+64) + Mac OS X + HP-UX 11.23)
Assigned to:	Magnus Blåudd	CPU Architecture:	Any

Description:
First release (test) build for 5.0.4, based on ChangeSet
  1.1882.1.3 05/04/06 23:12:10 serg@serg.mylan +5 -0
  after merge fix

Basically this is still the issue of bug#8866, but due to the several fixes and long history there I rather open a new entry.

On "octane2" (IRIX, bith 32 and 63 bit build) and on both "powermac" (Mac OS X) machines, a "default" build fails in test "kill" when run in default mode, '--ps-protocol' passes.
In all 4 cases, the difference is:
-------------------------------------------------------
*** r/kill.result       Thu Apr  7 02:25:36 2005
--- r/kill.reject       Thu Apr  7 04:46:58 2005
***************
*** 6,12 ****
  0
  kill @id;
  select 1;
- Got one of the listed errors
  select ((@id := kill_id) - kill_id) from t1;
  ((@id := kill_id) - kill_id)
  0
--- 6,11 ----
-------------------------------------------------------

How to repeat:
Build + test on either of these.

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/internals/24397

Patch doesn't work. It passes sometimes, but fails more often than not.

Build of 5.0.6, based on ChangeSet
  1.1901 05/05/26 15:21:01 joerg@mysql.com +1 -0
  Added a missing "make clean" to the RPM "spec" file which had caused the original 4.1.12 RPMs to be broken.
  (bug#10674, bug#10681)

This bug now also occurs on hpita2-64bit (in a "max" build, while "debug" passed!).

this is the infamous kill test failure.

it has something to do with when thd->killed gets noticed by the thread being killed, and whether it interrupts a read/write that may be in process.

on our powermacg5, the test passes about 10% of the time, and fails with either one of two other messages. (no error message, or an error on the wrong statement.)

See also Bug #11549.

This seems a different symptom of the same underlying problem:

kill                           [ fail ]

Errors are (from PATH/mysqltest-time) :
At line 31: query 'select 1' succeeded - should have failed with errno 2006...
At line 34: query 'select ((@id := kill_id) - kill_id) from t1' failed: 2013: Lost connection to MySQL server during query
(the last lines may be the most important ones)

Occurred in 5.0.11 on
powermacg4-5.0-max.log   normal
xserve-5.0-pro.log   normal
xserve-64bit-5.0-debug.log   normal
(also in previous versions, but not listed)

Problem does not exist anymore. The server will now say "goodbye" before closing the connection to the server. Even if it's been compiled with -DSIGNAL_WITH_VIO_CLOSE as was the case on the affected platforms in this bug report.

MySQL-5.0.33
RedHat Enterprise Linux 4 (i386)

Test Fails as described above:

kill                           [ fail ]  timeout

Aborting: kill failed in default mode. To continue, re-run with '--force'.
Stopping All Servers
Shutting-down Instance Manager
make: *** [test] Error 1