Apache + Fcgi , slow site loading

Discussion in 'Server Operation' started by invader7, Nov 28, 2011.

  1. invader7

    invader7 New Member

    Hello, i'm facing a problem with my site , it has a lot of traffic and it loads very slow , ~30-40sec.

    First i found the problem when i got some 500 internal errors , i found that IPCCommTimeout 31 , was to low and i changed it to 60 , now i don't get 500 errors but the site loads really slow , ~40 sec ...

    I think fcgi is under heavy presure and i doesnt work as expected...

    can you provide me some information in order to make things better ?
  2. till

    till Super Moderator Staff Member ISPConfig Developer

    Have you installed a php binary cache like xcache or eaccelerator?
  3. invader7

    invader7 New Member

    no i didnt , is that so necessary ?? its a joomla site
  4. till

    till Super Moderator Staff Member ISPConfig Developer

    It will speed up php processing a lot. Which Linux Distribution do you use?
  5. invader7

    invader7 New Member

    i use debian , the site is updating minute per minute , if i enable cache it wont take longer for the user to "see" the new content ?
  6. falko

    falko Super Moderator ISPConfig Developer

    No, updates are still shown to users immediately.

    php-apc is another PHP opcode cacher, it is available as a package for Debian Squeeze.
  7. invader7

    invader7 New Member

    i tried with another joomla site , different machine and the results are big difference , ~2 sec site loading....

    Now with the problem of the thread , i think its code problem because i moved the site at a machine with xeon 8 core processor with 4 GB ram , and i got very little changes..
  8. ircf

    ircf Member

    Same problem, how to configure fcgid ?


    I have the same problem : running Debian Squeeze + ISP Config 3.0.3 + Apache 2 + PHP + fcgid + suexec and approx 200 websites hosted on the same machine with only 4GB RAM. Here is my fcgi config (from a website vhost config) :
          IdleTimeout 3600
          ProcessLifeTime 7200
          # MaxProcessCount 1000
          DefaultMinClassProcessCount 3
          DefaultMaxClassProcessCount 100
          IPCConnectTimeout 8
          IPCCommTimeout 360
          BusyTimeout 300
    When apache starts all goes fine, but after few hours (2-3 hours) a huge number of process runs and all websites answer very slow (about 30-60s before display a page) :

    [email protected]:~# pstree
         │         ├─apache2───302*[php-cgi]
    My question are :
    - Which fcgi parameters should I modify to fix this and how ? I'd couldn't find a tutorial to tweak my fcgi config depending on my ressources (websites/RAM)
    - Where should I modify it ? Maybe /usr/local/ispconfig/server/conf(-custom)/vhost.conf.master ?
    - Once modified how I can re-generate each vhost config file ? (remember I have 200 vhosts)
    - Should I install php-apc too, or will this add significant overhead ?

    Notes :
    - My boss is negociating with our ISP to increase RAM, we'll try to add as much as possible...
    - This problem happens since yesterday when we imported about 70 new websites from an old server.

    Thank you for your support.

    Edit :
    - I've found the official apache fcgi documentation , it helped me a to understand fcgi config
    - I've found a (temporary ?) solution :
    1- I have commented fcgi config (see lines above) from /usr/local/ispconfig/server/conf/vhost.conf.master
    2- I modified each website config so the new vhost file can be generated for each website
    3- I have copied fcgi config from last ispconfig version into /etc/apache2/mods-enabled/fcgi.conf so I can globally tweak fcgi without having to modify each website :
            FcgidIdleTimeout 30 # Default is 300, process weren't recycled enough, I will set it higher when I have more RAM
            FcgidIdleScanInterval 30 # Default is 120, even if I set FcgidIdleTimeout < 120 process were removed only every 120 seconds
            FcgidProcessLifeTime 3600
            FcgidMaxProcesses 100  # Default is 1000, there were too many processes for my server to handle, I will set it higher when I have more RAM
            FcgidMinProcessesPerClass 0
            FcgidMaxProcessesPerClass 10 # Default is 100, I had to limit each site so that every site could have a chance
            FcgidConnectTimeout 3
            FcgidIOTimeout 360
            FcgidBusyTimeout 300
            FcgidMaxRequestLen 1073741824 # Avoid some HTTP 500 errors for upload programs
    Any other idea is welcomed. Thank you again.
    Last edited: Jan 18, 2012
  9. ircf

    ircf Member

    Help needed...

    Unfortunately, after trying different settings (see last settings in post above) the best I could do is avoiding the whole server to crash by setting a "sure" max process limit (100). The problem is all 200 sites can't run simultaneously. Besides some sites may take lots of process while others won't have one.

    I also installed APC (apt-get install php-apc) that really improved site performance, assuming you have a process for this site, so that didn't solve my problem.

    Tomorrow I'll try to increase a little max process and to set max process per site, and upgrade RAM asap. If anyone had a clue to that would save my life ! ty :)

    PS : Previous server ran the 200 websites with mod_php and only 2GB with no problem. Indeed security and quota handling was problematic...

    Edit :
    - This morning I set FcgidIdleScanInterval and FcgidMaxProcessesPerClass (see modified config above), wait and see...
    - I noticed some (most ?) process are not killed after IdleTimeout, especially when ps count reaches maxProcess, I seems not to decrease and stick to maxProcess (some process are still recycled anyway). I couldn't find a forum thread about this (mod_fcgid bug ?). To "fix" this I added "0 * * * * /etc/init.d/apache reload" to my crontab, that will kill processes each hours, I couldn't find a better way until now. Since I added this, processes are always under maxProcess (around 50-80 processes) and all sites can work smoothly, I don't think this is a good solution anyway, besides it may disconnect people each hour
    - Maybe the "process not killed" bug is related to the Debian mod_fcgid version (2.3.6), the mod_fcgid changelog shows a similar bug solved in next version (2.3.7)
    - Finally found a fix for the "process not killed" bug : Some websites vhosts weren't updated and still had the old fcgi config. After having updated all website, processes are killed perfectly, then I removed my crontab entry. I hope this bug won't get back if I set specific fcgi config on some website (according to mod_fcgid changelog)
    Last edited: Jan 19, 2012

Share This Page