No announcement yet.

Busy Polling with PERFdata turned on

  • Filter
  • Time
  • Show
Clear All
new posts

  • Busy Polling with PERFdata turned on

    Hi All;

    We are running a monitoring solution with around 9000 service checks over 500 hosts, system is based a ESXI/Redhat 5.5 host (8G mem) with all the normal bells 'n' whistles.

    We are using a mysql slave config, local RRD files and drbd with another system
    containing Nagios config for a manual failover if required. (poor man's cluster!)

    For the centstatus database, we have this in 'mem' as apposed to mysql db
    to try and improve performance but the only way to reduce the polling delay
    and the check's backlog was to turn perfdata off.

    Anyone give any example installations of this scale in production and
    how we may approach reducing the polling backlog. We have looked
    at the DB and possibly reducing the check data we require.....

    Looking for the 'golden' bottleneck.

    Last edited by gt71027; 24 May 2011, 14:32.

  • #2
    In fact perfdata export in Nagios is done with command that block the checks execution during run (export-serviceperfdata), you can try to launch it with some & so they will not block the check execution, and so will reduce the latency.
    Auteur de Shinken, outil de supervision compatible avec Nagios et orientée supervision distribuée hautement disponible et mulitplateforme.


    • #3

      first of all, migrate your MySQL instance to a dedicated host. It can be on the same ESXi host, but practical experience shows that it should be a seperate virtual machine. Then you should tune your MySQL instance. For the beginning you could use the MySQL tuning primer at for some basic hints. Then see Bug #2393 at

      We're monitoring with such an tweaked environment at a client site approx. 2500 hosts and 13000 services without performance issues.



      • #4
        Hi Frank,

        Thanks for the insight, when you mean migrate mysql instance, you mean move it off the polling box right ?

        For your monitoring, do you have 'PERF' data turned on ? if we turn it off then 'polling' times drop right off.



        • #5
          And which database engine are you using ? assuming it is InnoDB, we tried Heap (mem based) with no improvement.


          • #6
            Sorry for the late answer.

            Yes, i mean move it off the polling box to a dedicated machine.

            We're using MyISAM for the centstorage db.