[Labs-l] Memcached issues on piramido

[Andrew Bogott]

On 1/30/13 3:37 PM, Matthew Flaschen wrote:
> On 01/30/2013 05:11 PM, Ryan Lane wrote:
>
>> Looking on the instance, in dmesg I see segfaults:
>>
>> [3014246.521105] memcached[13216]: segfault at 0 ip 000000000040d525 sp
>> 00007ffbc8336a90 error 4 in memcached[400000+18000]
>>
>> That said, memcache is configured on that instance to use 2GB of memory:
>>
>> /usr/bin/memcached -m 2000 -p 11000 -u nobody -l 127.0.0.1 -c 25000 -n 5
>> -f 1.05
>>
>> The system only has 1GB total.
> It says  2048 Mb  at
> https://labsconsole.wikimedia.org/wiki/Nova_Resource:I-000004bd .
>
> Granted, it might not be a good idea to allocate it all to memcached
> (not sure how smart it is about not filling the RAM).
>
> Do you recommend we decrease the amount memcached can use?  If so, where
> is the setting?
It looks like in this case memcached was installed via the 
mediawiki::singlenode class.  Since that class is only used in labs, 
I've changed it to use 128M of cache.

https://gerrit.wikimedia.org/r/#/c/47680/

If this turns out not to be the cause or if you need more flexibility in 
this, let me know!

-Andrew