[imp] process / memory problems in new release (?)

[Matthew Dunham]

Michael M Slusarz wrote:
> Quoting Matthew Dunham <matthew.dunham at ic.ucsb.edu>:
> 
>>>>>> I've confirmed that these problems don't exist in Horde 3.2.1/IMP 
>>>>>> 4.2,
>>>>>> so the trigger is clearly something that's been introduced in the 
>>>>>> past
>>>>>> months' development efforts. Just for fun I disabled caching, 
>>>>>> thinking this new feature might be the culprit -- but no dice.
>>>>>
>>>>> Which caching feature did you disable? There are a number of 
>>>>> different ones; several are called out in the accompanying comments 
>>>>> as ones that increase session size, which might lead to memory 
>>>>> growth in httpd processes.
>>>>
>>>> ah, the new horde object caching system (if that's what you call it).
>>>>
>>>> $conf['cache']['driver'] = 'file';
>>>>
>>>> i'm using this setting in production, but see the same behavior with:
>>>>
>>>> $conf['cache']['driver'] = 'none';
>>>
>>> That will definitely not hurt you and may hurt. You should look 
>>> through the IMP settings for things that might increase session size 
>>> - I don't have them in front of me right now, sorry.
>>
>> So I'm not sure what this first sentence means.
>>
>> My tests seem to indicate that setting this cache driver to 'none' 
>> doesn't change the symptom -- process size increases with either of 
>> these settings.
>>
>> I would assume that file-based cache is storing objects in the 
>> filesystem rather than memory. Is this not correct?
>>
>> Also, when you suggest IMP settings may increase session size, do you 
>> suggest the session object is being cached in memory regardless of the 
>> cache driver setting?
> 
> What this means is that caching is not your problem.  And session size 
> shouldn't be either - there was definitely nothing added to increase 
> session size (at least in IMP) between 4.2 and 4.3 and if anything, 
> session size would have been reduced.  Your problem lies elsewhere.  
> FWIW, in the past I know there have been problems with Solaris/IMP with 
> - at a minimum - database related stuff and the spellchecker.
> 
> Another thing you might want to look at is your PHP optimizer/cacher.  
> Seeing a process' resident memory jump 17MB on the first access sounds 
> suspiciously like that PHP/Apache combo is caching the opcodes for that 
> process only and is not using shared memory for this cache.  if every 
> process is caching its own opcodes - that is very no bueno.

Was running APC, but of course disabled that straightaway thinking it 
was a culprit. It's not.

Further, I can fire up the server and hit non-Horde php pages a-ok. It's 
only when I log into Horde 3.3 that I see the process size grow as 
described previously.

Like I said, the change occured between Horde 3.2.1 -> 3.3 / IMP 4.2 -> 
4.3. I've been fairly rigorous in my testing to narrow this down.

-m