numfile issue

[Drazen Korda]

I have a new VPS server handling mail for about 30 users and a few websites, nothing fancy, not even a database. I am getting a hi number of QoS Alerts pointing to numfile value. The Yellow zone alerts are very numerous, about 10 per hour during the prime-time but the drop and even disappear overnight. But it gets into black zone about 3-5 times a day. Normally, it clears itself within minutes.

Looking around I got an impression that the numfile count should be set higher but my provider, 1and1 is not willing to do it, well they want me to upgrade.

Normally the numfile count is about 2-3k. I am not experienced enough to figure out what causes it to sump to 4k. At those times system becomes unresponsive anyway.

Looking al lsof output there is a large number of files opened by the httpd and I am trying to figure out how to reduce that. Can anyone suggest where else I should look?

This is the output of beancounters.

# cat /proc/user_beancounters
Version: 2.5                                                                                                                     
       uid  resource                     held              maxheld              barrier                limit              failcnt
  52728218: kmemsize                 12336739             12357932             20971520             23068672                    4
            lockedpages                     0                    0                  256                  256                    0
            privvmpages                 61170                61213               262144               288358                    0
            shmpages                     2834                 2834                20480                20480                  220
            dummy                           0                    0                    0                    0                    0
            numproc                        54                   54                   96                   96                    0
            physpages                   28770                28773           2147483647           2147483647                    0
            vmguarpages                     0                    0                65536           2147483647                    0
            oomguarpages                28779                28782                65536           2147483647                    0
            numtcpsock                     28                   28                  360                  360                    0
            numflock                       11                   11                  188                  206                    0
            numpty                          4                    4                   16                   16                    0
            numsiginfo                      0                    0                  256                  256                    0
            tcpsndbuf                  218560               218560              1720320              2703360                    0
            tcprcvbuf                  214176               218568              1720320              2703360                  651
            othersockbuf               203120               203120              1126080              2097152                    0
            dgramrcvbuf                     0                    0               262144               288358                  190
            numothersock                  124                  124                  360                  360                    0
            dcachesize                1070195              1073192              3145728              3460300                    0
            numfile                      2828                 2837                 4096                 4096                89024
            dummy                           0                    0                    0                    0                    0
            dummy                           0                    0                    0                    0                    0
            dummy                           0                    0                    0                    0                    0
            numiptent                      58                   58                  100                  105                    0

 

[Denis Sirochenko]

Hello,

Check your kernel version. CU-2.6.18-028stab060.8 Parallels Virtuozzo Containers 4.0 kernel issued on 2009-02-23
provides a number of important stability fixes, including the fix for the following issue:

The "numfile" and "kmemsize" usage could grow without limit due to a bug
in the file precharging mechanism. In particular, the "named" process could
trigger this growth under certain circumstances.

For more details, look at the Knowledge Base article describing the kernel:

http://kb.odin.com/en/6066

P.S: Isn't it a Virtuozzo, not Plesk related issue?

Regards,
Denis

 

[IrwinS]

resolved?

Wondering if you ever figured this out? I have a similar problem where my numfile limit climbs very quickly and hits the yellow zone about 6 times an hour during peak usage.

I have been unable to find a resolution yet. One thing I did notice though was that the high volume of files open may be linked to mail accounts. As the admin for the server I have 6 different email accounts. If I turn off my mail client the numfiles count decreases by 25%!

Any thoughts on how to resolve this would be greatly appreciated.