Michael Widenius wrote:
>
> Hi!
>
> >>>>> "Tomas" == Tomas Styblo <admin@stripped> writes:
>
> Tomas> I have following problem.
> Tomas> I use MySQL latest stable version from Linux x86 RPM, on RedHat 6.1.
> Tomas> Sometimes (approx. once per every 30 000 generated pages on my web) the
> Tomas> count of mysqld proceses starts slowly grow, to infinite. Six processes
> Tomas> of connected clients are then frozen in read() /found using strace/ call
> Tomas> of their connection to the mysql. They connect to the DB thru unix file
> Tomas> sockets. This has to be an bug in mysqld, because the number of clients
> Tomas> IS limited to six, and there are no more of them, when mysqld starts to
> Tomas> behave this way (spawn more and more processes). The mysqld then can't
> Tomas> be connected to, mysqladmin shutdown frozens when it tries to connect to
> Tomas> the server, the mysqld doesn't respond to -TERM, so I have to kill it
> Tomas> thru killall -KILL. The tables data files are never inconsistent after
> Tomas> that. No long time queries or big tables (20000+ rows) are used. But it
> Tomas> happens only when the system is under heavy load. One of my collegues
> Tomas> has experienced exact the same problem on his box in the past. Everytime
> Tomas> it happens, I can find this in the syslog :
>
> Tomas> "May 9 15:45:16 matrix kernel: SIG: sigpending lied".
>
> Tomas> I am not a kernel guru, but in kernel src I've found that this message
> Tomas> occurs in a part of code which is somehow related to threads handling. I
> Tomas> tried to make mysqld to dump core thru -SEGV, but I can't find it
> Tomas> anywhere. Although the binary in RPM is linked statically, I think some
> Tomas> incompatibility between system libraries (threads/glibc) can be the
> Tomas> problem - my system uses kernel 2.2.15, glibc 2.1.3.
>
> Tomas> Thanks for any help or suggestions.
> Tomas> Tomas Styblo
>
> This really sounds like a kernel problem.
>
> If you can somehow make a client (perl?) program that can get mysqld
> into this behaveour we would of course be very happy to look into
> this. You can probably use the fork_test.pl or mysql-super-smack
> programs to do a test case..
>
> Sasha here at TCX has done a lot of testing of MySQL under high load
> and the only problems he has found is that when you have MANY threads
> running, thread creation may take a long time, but this is not the
> problem you have :(
>
> Regards,
> Monty
Try building from source, and see if the problem persists.
--
Sasha Pachev
+------------------------------------------------------------------+
| /*/\*\/\*\ /*/ \*\ /*/ \*\ |*| Sasha Pachev |
| /*/ /*/ /*/ \*\_ |*| |*||*| mailto:sasha@stripped |
| /*/ /*/ /*/\*\/*/ \*\|*| |*||*| Provo, Utah, USA |
| /*/ /*/ /*/\*\_/*/ \*\_/*/ |*|____ |
| ^^^^^^^^^^^^/*/^^^^^^^^^^^\*\^^^^^^^^^^^ |
| /*/ \*\ Developers Team |
+------------------------------------------------------------------+
