Re: [squid-users] FATAL bus error but no core - suggestions?

On Tue, 13 May 2003, Adam wrote:

> Hello again,
>
> Last week squid crashed around 8am (usage is moderate, not high at that
> hour) on my PROD box with the error " FATAL: xmalloc: Unable to allocate
> 24576 bytes!" Per the FAQ it can either be a lack of swap or "maximum data
> segment size." Since ulimit reports unlimited for that, I am assuming that
> swap becomes in short supply. There is only 384MB of on disk swap though
> with 1GB of RAM, there is usually some left over for swap. cache_mem is
> 374MB and cache_dir is "aufs 7072 16 256." I do plan on increasing the
> size of the ondisk swap filesystem during our next maintenance window (Wed
> night next week), so that may also fix this new problem. Is my sizing off.
> It seems odd this error has only happened once so the situations where it
> completely runs out of swap is rare. Will making ondisk swap 1GB help?
>
> What happened today was squid crashed with the error "FATAL: Received Bus
> Error...dying." which, according to the three threads I found on the mailing
> list, would indicate a bug (FAQ shows how to analyze the core).
> Unfortunately no core was generated. I've set my coredump_dir to /logs
> which is owned by squid hence read/writable by squid. Fortunately in both
> instances squid was able to restart itself but I am wondering if there is
> something else I need to do to enable core creation for squid. I doubt it
> but am posting, just in case. I haven't disabled core creation (i.e. using
> coreadm or anything). Server is an Ultra 60, 1GB of RAM, Solaris 8, Squid
> 2.5STABLE2.

I don't have a solution for you but I too get both these errors followed
by a restart of Squid on a fairly regular basis (once a week or so). I run
Squid 2.4S7 on Solaris 8. I did open a bug report but, so far, I have been
unable to convince Squid (or Solaris) to dump Core during a crash. I don't
think my problem is SWAP as I have lots of memory and lot's of Swap space
free. I'm just testing Squid 2.5S2 and was hoping this would resolve my
crashes but based on your report I'm not so sure.

Henrik, are you saying that if I shutdown Squid and restart it under the
Squid userid that it would dump core during a segmentation/bus error?

-Mike

>
> thanks,
>
> Adam
>
> today's cache.log extract:
> 2003/05/13 10:29:11| httpAccept: FD 12: accept failure: (130) Software
> caused connection abort
> FATAL: Received Bus Error...dying.
> 2003/05/13 10:29:25| storeDirWriteCleanLogs: Starting...
> 2003/05/13 10:29:25| WARNING: Closing open FD 12
> 2003/05/13 10:29:25| 65536 entries written so far.
> 2003/05/13 10:29:25| 131072 entries written so far.
>
> cache_mem 374 MB
> cache_dir aufs /cache 7256 16 256
>
> "top" output a few hours after the crash:
> Memory: 1024M real, 461M free, 337M swap in use, 703M swap free
>
> PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
> 25317 root 20 22 0 315M 313M sleep 30:38 13.48% squid
>
>

        ** *** ****** Mike Diggins
       **** **** ******** Analyst, Networks
      ** ******* ** ** Computing and Information Services
     ** ** ** ** ** McMaster University, Hamilton, Ontario
    ** ** ** ** (905) 525-9140 Ext. 27471 ABB131A
   ** ** ********* Fax: (905) 528-3773
  *** *** ******** Email: diggins@mcmaster.ca
Received on Wed May 14 2003 - 11:05:51 MDT