Busy loop with memcached 1.2.2

[Kari Lavikka]

On Tue, 21 Aug 2007, Paul Lindner wrote:

> Okay.  1.2.1 -> 1.2.2 had quite a few changes. So it will be hard to
> pin it on any specific thing.  Did you enable threads on your build?

Ah, forgot to mention that. Threads were not enabled.

>> We have debian linux on 64bit dual opterons with 8GB of memory. Each
>> server runs two instances. Libevent is: libevent-1.1a.so.1.0.2
>
> That's a pretty old libevent, 1.1b has some bugfixes

Yes, seems to be. Have to upgrade...

>
>> Command line arguments were:
>> memcached -m 3500 -c 8192 -d -P /var/run/memcached2.pid -p 11212 -u
>> memcached
>
> You might try removing the -d flag if you plan to use the -P flag at
> the same time.  There was a bug that surfaced with that in 1.2 on
> x86_64 due to some odd interaction between libevent and open file
> descriptors.

Sound quite weird. I don't actually need pid file for anything so I'll try 
without that.. and update only one instance this time.

One small details was missing. I used -m 3400 instead of -m 3500 with 
1.2.1.

I still have to emphasize the fact that according to "top", resident size 
of each memcached instance was 1.4GB on the event of crash. Maybe there 
was something strange going on when expanding the hashtable and a loop 
appeared to the linked list of hash bucket... or some other corruption..?

Thanks for these tips anyway :)

>
> Thanks
> Paul

     |\__/|
     ( oo )    Kari Lavikka - tuner at bdb.fi - (050) 380 3808
__ooO(  )Ooo_______ _____ ___ _ _  _   _    _      _                  _
       ""