Opened 9 years ago

Last modified 9 years ago

#3243 assigned defect

Apache hangs due to no way to acquire mutexes

Reported by: alsroot Owned by: bernie
Priority: Unspecified by Maintainer Milestone: Unspecified
Component: infrastructure Version: Unspecified
Severity: Unspecified Keywords:
Cc: Distribution/OS: Unspecified
Bug Status: Unconfirmed

Description

At some point, apache on sunjammer stops reflecting on requests and even SIGTERM can't stop processes, only SIGKILL.

The logs say:

[Fri Nov 25 07:29:17 2011] [emerg] (22)Invalid argument: couldn't grab the accept mutex

Change History (5)

comment:1 Changed 9 years ago by alsroot

  • Owner changed from sascha_silbe to bernie
  • Status changed from new to assigned

comment:2 follow-up: Changed 9 years ago by alsroot

I've increased these values (by guessing) in /etc/sysctl.d/99-apache.conf:

  • semmsl from 250 to 500
  • semmni from 128 to 256

comment:3 in reply to: ↑ 2 Changed 9 years ago by alsroot

Replying to alsroot:

I've increased these values (by guessing) in /etc/sysctl.d/99-apache.conf:

  • semmsl from 250 to 500
  • semmni from 128 to 256

s/semmsl/semmns/

comment:4 Changed 9 years ago by bernie

We can check if sys5 semaphores are really running out with "ipcs -s".

I also have a trick to cleanup the semaphores globally:

for i in ipcs -s | grep www-data | awk '{print $2}'; do ipcrm -s $i; done

comment:5 Changed 9 years ago by bernie

Oops, the wiki syntax mutilated my poor one-liner.

for i in `ipcs -s | grep www-data | awk '{print $2}'`; do ipcrm -s $i; done
Note: See TracTickets for help on using tickets.