[horde] Slow response times

Fri Apr 18 21:21:58 UTC 2014

Sent from Samsung tablet

-------- Original message --------
From: Michael M Slusarz 
Date:19/04/2014 01:54 (GMT+10:00) 
To: horde at lists.horde.org 
Subject: Re: [horde] Slow response times 

Quoting Simon Wilson <simon at simonandkate.net>:

> ----- Message from Simon Wilson <simon at simonandkate.net> ---------
>     Date: Fri, 18 Apr 2014 23:21:48 +1000
>     From: Simon Wilson <simon at simonandkate.net>
> Reply-To: simon at simonandkate.net
>  Subject: [horde] Slow response times
>       To: horde at lists.horde.org
>
>
>> I have been running Horde for quite a few years now, but have never  
>> had this issue before. I would appreciate some help if anyone can...
>>
>> I've moved my Horde web server (that has been running Horde 5 for a  
>> year or more) from being reverse proxied behind another box to now  
>> being 'direct', so I have made a couple of Apache config changes.  
>> That was a week ago, and it has been running fast and fine. Today  
>> Horde has slowed to a crawl, from multiple PCs and browsers.
>>
>> The server is running fully up to date Horde 5.1.6 on a CentOS 6.5  
>> KVM virtual machine with 8GB RAM (the host has 32GB), as it has  
>> been for ages. PHP is 5.3.28. Horde VM average load is under 5%.  
>> The host's average load is under 10%.
>>
>> Horde apps and pages are taking between 15 and 30 seconds to load.  
>> Logout is instantaneous. Test.php is instantaneous. GET login.php  
>> is instantaneous. POST login takes 30 seconds, but loading the CSS  
>> and JS is quick. Basic mode is as slow as dynamic. If I log into  
>> Smart phone mobile mode it is fast (under 3 seconds even for  
>> 'heavy' pages). ActiveSync is still flying, with logged response  
>> times under 0.5 sec. ActiveSync refreshes fast on the phones.  
>> Connectivity within Horde apps remains fast and responsive, so if I  
>> load Imp (which takes 15 seconds) emails appear almost instantly  
>> once Imp loads, and can be read, refreshed, deleted, etc., all as  
>> per normal speed.
>>
>> The Apache server on the VM serves other web sites, and they  
>> respond in under a second, including a Wordpress blog that runs on  
>> PHP and that is plenty quick enough. PHPLdapAdmin runs internally  
>> on the same server, and responds (including with a https login to  
>> the separate LDAP server that Horde auths users to) in under 0.5  
>> seconds).
>>
>> For troubleshooting I am connecting inside the LAN with just http  
>> instead of https and with no port forwarding via my router. I have  
>> disabled / re-enabled caching (currently xcache). I have restarted  
>> apache, checked its config, and rebooted the server. I have trimmed  
>> down php loaded modules. Still, horde load times are between 12  
>> seconds and 30 seconds.
>>
>> Horde logs show nothing abnormal at INFO level, and at DEBUG level  
>> so much is logged it is hard to follow anything for the 30 or 40  
>> seconds I have it at DEBUG.
>>
>> If anyone has any ideas where I can look next please let me know!  
>> :) Log detail below.
>>
>> Thanks
>> Simon
>>
>
> Answered it myself.
>
> Installed xdebug, and drilled down into the resultant file. Found:
>
> Gollem_Application -> authTransparent was taking time.
>
> My Samba server had an issue which I was not aware of, and because  
> Horde couldn't see the three SMB shares, the entire Horde website  
> ground to almost a halt.
>
> It's certainly a drastic reaction to an app in the background not  
> being able to see a share...

Not sure what else we can do.  You have to have some kind of  
reasonable timeout that allows a connection to the Gollem VFS backend  
if it doesn't immediately respond.  I'm guessing that timeout is 5  
seconds which may be too long by default (can't remember if it is  
configurable).  But it can't be 0.1 seconds either.

Even if it was 2 seconds, that would still be 6 seconds of wait time  
on your particular installation.  Which means the remedy is the same:  
either fix the VFS connection or remove it from the backend list.

In the past, I believe we were trying to connect to the gollem VFS  
backends even if they weren't used in a page access, but I believe  
that has been fixed.

michael

___________________________________
Michael Slusarz [slusarz at horde.org]

-- 
Horde mailing list
Frequently Asked Questions: http://horde.org/faq/
To unsubscribe, mail: horde-unsubscribe at lists.horde.org

Thanks Michael. I wasn't going near Gollem, and every page access was doing it, even saving configuration. 

One thing I noticed was what appears to be a lack of logging, even at high log levels. 

Should a Gollem vfs backend fail like this not be notified to admin through log entries? Would have saved a lot of time and troubleshooting. 

Simon 