List:Internals« Previous MessageNext Message »
From:Michael Widenius Date:October 7 2002 4:40pm
Subject:RE: Load problems with 3.23.51 (syscall/lack thereof)
View as plain text  
Hi!

>>>>> "Steven" == Steven Roussey <sroussey@stripped> writes:

>> -----Original Message-----
>> From: Michael Widenius [mailto:monty@stripped]

Steven> This is obviously different. 3.23.47 is in tcp_data_wait.
Steven> 3.23.51 is usually doing nothing (!) or in suspend. Odd.
>> 
>> I agree that this is really strange.
>> 
>> Could you please do a test with MySQL 3.23.52 and see if you get
>> similar results with this. (We have switched to an older glibc library
>> in an effort to get this fixed)

Steven> Just an update. We are back to 3.23.49a, the last version to work
Steven> flawlessly in a heavy load environment for us. The odd TCP state problem
Steven> that kills MySQL under load, can be found in versions 3.23.51, 3.23.52
Steven> (the one you asked about above), 4.0.3, and 4.0.4 (just released).

Thanks for the update.  It seams that something is strange with our
new glibc library and we have two guys trying to simulate this on our
machine.

Unfortunately we don't have a copy of the old working glibc due to a
hardware disk crash, which makes it a bit hard for us to reproduce
the old binaries.  Hopefully we get this solved soon.

We will inform you as soon as we have managed to solve this.  The
current state is that wWe know that one can get a working MySQL server
by compiling MySQL on a different build machine without our optimized
glibc library.  The problem is that the optimized glibc library
(similar one to the ones used in 3.23.49a) is needed for a lot of our
'heavy' users and we really need to get this working.

Steven> Working ones: 3.23.49a (currently using) and 4.0.2. We can't use 4.0.2
Steven> with recent versions of PHP as we get loads of 'out of sync' errors when
Steven> using persistent connections. While persistent connects to MySQL can be
Steven> bad when you have four web servers connecting to the db, it is good
Steven> thing when you have better hardware and can have just one webserver.

Did you get out of sync errors with 4.0.4 too ?
(Just curious)

Regards,
Monty
Thread
RE: Load problems with 3.23.51 (syscall/lack thereof)Steven Roussey3 Oct
  • RE: Load problems with 3.23.51 (syscall/lack thereof)Michael Widenius7 Oct
RE: Load problems with 3.23.51 (syscall/lack thereof)Steven Roussey4 Oct
Re: Load problems with 3.23.51 (syscall/lack thereof)Jocelyn Fournier7 Oct
  • Re: Load problems with 3.23.51 (syscall/lack thereof)Michael Widenius8 Oct
    • RE: Load problems with 3.23.51 (syscall/lack thereof)Steven Roussey9 Oct
    • RE: Load problems with 3.23.51 (syscall/lack thereof)Steven Roussey9 Oct
Re: Load problems with 3.23.51 (syscall/lack thereof)Jocelyn Fournier10 Oct