Bug #20080 16 mysql-test-run tests hang, resulting in corrupt tables
Submitted: 26 May 2006 1:01 Modified: 26 May 2006 2:50
Reporter: J.L. Hill Email Updates:
Status: Not a Bug Impact on me:
None 
Category:MySQL Server Severity:S1 (Critical)
Version:5.0.21 debug x86_64 binary OS:Linux (Debian Sarge-- Kernel 2.6.16.17)
Assigned to: CPU Architecture:Any

[26 May 2006 1:01] J.L. Hill
Description:
16 mysql-test-run tests hang. These tests  corrupt tables when run singly and when run singly in debug mode. Per MySQL's debug instructions, I have uploaded a trace of one failed test (auto_increment) to ftp.mysql.com/pub/mysql/upload/ (as jhill-master.trace).

One test fails, is rpl_rotate_logs (no table corruption). The tests that hang are: 
          auto_increment              index_merge_ror                 insert_select
          rpl_ignore_grant             trigger                               delete
          information_schema_db   insert                                rpl_ignore_revoke
          grant2                           information_schema_inno     merge.test            
          grant                             information_schema             myisam
          sp

After MySQL shutdown, 

        myisamchk -e -s master-data/test/*.MYI
        myisamchk: MyISAM file master-data/test/t1.MYI
        myisamchk: warning: 1 client is using or hasn't closed the table properly
        MyISAM-table 'master-data/test/t1.MYI' is usable but should be fixed

While this is only a warning, the table cannot be accessed from mysqld. Select, insert and delete commands simply hang.

I have ran filesystem checks and bonnie++ with no errors reported. No errors are reported in system debug, kernel or syslogs.

The system is: Dell PE2850 Xeon, vanilla kernel 2.6.16.17, software raid-1 on ReiserFS3. This problem began a month ago on a system that was stable. I am reporting this as suggested by the mysql-test-run Readme. 

I am not concluding that this is a MySQL, bug but it is a problem as MySQL hangs daily with corrupt tables. I have tried different versions of MySQL and various 2.4 and 2.6 kernels and configurations with the same result.

I have used MySQL since 1999 and this is the first time I've had a critical bug. I believe the only way to resolve this problem is for the assistance of someone who can understand the uploaded master.trace well enough to provide some educated direction to resolve the problem.

This comes at the worst time: I just proposed that the company I'm developing for purchase MySQL Network.

Thank you,

Jeff Hill

How to repeat:
mysql-test-run --local --debug t/auto_increment.test

or any of the tests noted in my description. 

The tests hang indefinitely (I've left them to run for up to five hours). The resulting myisamchk (run with mysql shutdown [confirmed with a ps check]) warning states that the table is usable, which it is not.

The database the table can be used, other tables and databases can be used, but not the corrupted table.

Suggested fix:
None.
[26 May 2006 1:08] Jim Winstead
This is a bug in Debian's glibc.

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=314408
[26 May 2006 2:50] J.L. Hill
Thank you very, very much. I swear I searched the Debian bug reports, obviously incorrectly (too much stress I guess). I've spent days searching. The MySQL Readme's request to submit failed tests was just too much to resist.

I know I'm pushing my luck to ask for more advice, but is the only reasonable solution to upgrade to Debian Etch? I can't see a better way, although I may have to fight other bugs in Etch.

Thank you again. I promise not to bother further.

Jeff