Bug #48236 InnoDB fails in "os_file_readdir_next_file()" on AIX and i5os
Submitted: 22 Oct 2009 15:22 Modified: 2 Feb 2010 17:03
Reporter: Joerg Bruehe Email Updates:
Status: Duplicate Impact on me:
Category:MySQL Server: InnoDB storage engine Severity:S3 (Non-critical)
Version:5.1.40 + 5.5.0 OS:Other (AIX + i5os)
Assigned to: Assigned Account CPU Architecture:Any

[22 Oct 2009 15:22] Joerg Bruehe
All runs of test "crash_commit_before" fail on AIX and IBM i5os (all configurations) for both the built-in and the plugin version of InnoDB.

The scheme is:
     1  main.crash_commit_before                 [ retry-fail ]  Found warnings/errors in server log file!
     2          Test ended at YYYY-MM-DD HH:MM:SS
     3  mysqltest: Logging to ''.
     4  mysqltest: Results saved in ''.
     5  mysqltest: Connecting to server localhost:13000 (socket /PATH/mysql-test/var/tmp/mysqld.1.sock) as 'root', connection 'default', attempt 0 ...
     6  mysqltest: ... Connected.
     7  mysqltest: Start processing test commands from './include/check-warnings.test' ...
     8  file_name       line
     9  /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysqld.1.err      InnoDB: Error: os_file_readdir_next_file() returned -1 in
    10  /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysqld.1.err      InnoDB: Error: os_file_readdir_next_file() returned -1 in
    11  /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysqld.1.err      InnoDB: Error: os_file_readdir_next_file() returned -1 in
    ... ...
   406  /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysqld.1.err      InnoDB: Error: os_file_readdir_next_file() returned -1 in
   407  /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysqld.1.err      InnoDB: Error: os_file_readdir_next_file() returned -1 in
   408  /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysqld.1.err      InnoDB: Error: os_file_readdir_next_file() returned -1 in
   409  ^ Found warnings!!
   410  mysqltest: ... Done processing test commands.
   411  mysqltest: Test has succeeded!
   412  ok

   413   - saving '/PATH/mysql-test/var/log/main.crash_commit_before/' to '/PATH/mysql-test/var/log/main.crash

   414  Test has failed 2 times, no more retries!
   415  ===---===---===---===---===---===--- BUGDB:
   416  ===---===---===---===---===---===--- BUILD-ID: 2009100621142320647
   417  ===---===---===---===---===---===--- OCCURRED:
   418  i5os-02.mysql.com,i5os-power-32bit,5.1-community.log   debug

Similar for other configurations, whether on AIX or i5os, 32 or 64 bit, built-in or plugin.

How to repeat:
Run the test suite on a debug server.
[23 Oct 2009 11:14] Mikhail Izioumtchenko
just to make sure:
the problem is that on specified platforms there are error messages in
mysqld log:

InnoDB: Error: os_file_readdir_next_file() returned -1

but the execution continues, no crash or assert?
[23 Oct 2009 17:06] Joerg Bruehe
I have quoted everything that got logged for this test,
just eliminated the lines 12 through 405 which are identical to 9-11 and 406-408.

So yes, there was no crash or assert logged, but MTR considered these lines to indicate enough of a problem to mark that as a test failure.

Maybe these messages are to be expected on these platforms, and the code handles all this just fine (you sure can check that), then we need some way to tell this to MTR.
[23 Oct 2009 17:38] Mikhail Izioumtchenko
Joerg, as I don't have this kind of hardware readily available could
you provide the entire mysqld log, in your example this would be /home/mysqldev/tmp-200910061726-5.1.40-20647/i5os-power-32bit/test/mysql-test/var/log/mysq
ld.1.err? In fact I'm mostly interested in a few lines
before and after the "InnoDB: Error: os_file_readdir_next_file() returned -1 in
" message which hopefully will provide some additional information.
[23 Oct 2009 19:48] Hery Ramilison
mysqld.1.err error file

Attachment: mysqld.1.err (application/octet-stream, text), 3.65 KiB.

[23 Oct 2009 19:52] Hery Ramilison
We just ran on aix entreprise in debug mode the crash_commit_before. Error only occurs when running in debug mode. Complete log in mysqld.1.err file.
[26 Oct 2009 12:54] Mikhail Izioumtchenko
Vasil, could you try to figure this out? IBM likes to do things a bit differently, maybe we are missing an error code here.
[19 Nov 2009 16:26] Joerg Bruehe
Same problem in the build of 5.5.0-beta,
tests "crash_commit_before" and "myisam_crash_before_flush_keys".
[28 Jan 2010 14:48] Andrew Hutchings
Possibly due to bug#50691
[28 Jan 2010 15:11] Mikhail Izioumtchenko
likely a duplicate of that bug, reassigning to Jimmy as well
[29 Jan 2010 3:37] Jimmy Yang
Certainly looks like a dup of bug #50691, whose fix is being reviewed.
[2 Feb 2010 17:03] Andrew Hutchings
Tested with debug build of fix for bug #50691.  This test now passes.  Closing as duplicate