Ad Widget

Collapse

High DB load with 1.4

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • eran
    Member
    • Apr 2007
    • 36

    #1

    High DB load with 1.4

    I've been using 1.3.8 since it was released and last week upgraded to 1.4, I noticed that the queries per second on the DB and the CPU utilization on the DB servers have jumped considerably when switching, and it gets worse with time. With 1.3.8 the server DB server was running at about 300 queries per second, which by itself is way too much for a configuration with 15 monitored devices, and after switching to 1.4 it jumped to over 1000, and kept climbing with time.
    I'm using a dedicated DB server with quad core Xeon CPU and RAIN 10 hard drive and it got to about 90% CPU utilization after a while.

    So I switched back to 1.3.8, it took a while for the queries per second to go back to the normal level but once it did it stayed there. Today I switched back to 1.4, just to make sure it was related. You can see the results of this experiment in the attached graph. Needless to say that the CPU on the DB server jumped as well and it's now on 55%.

    I'm going back to 1.3.8

    Eran
    Attached Files
  • Alexei
    Founder, CEO
    Zabbix Certified Trainer
    Zabbix Certified SpecialistZabbix Certified Professional
    • Sep 2004
    • 5654

    #2
    Something is definitely wrong here. It could be a known issue of 1.4. Do you have any records in table 'nodes'? If so, delete all of them.

    If not, please could you send me last 100-200 lines from zabbix_server.log with Debug set to 4. Thanks.
    Alexei Vladishev
    Creator of Zabbix, Product manager
    New York | Tokyo | Riga
    My Twitter

    Comment

    • eran
      Member
      • Apr 2007
      • 36

      #3
      The nodes table was empty.

      The log files are attached. I switched to 1.4, let it run for a few seconds so I could collect the logs and switched back to 1.3.8, so the log only shows the first few seconds after starting the service. All the entries in both files are from version 1.4.

      One interesting thing to note is that even when I switched back to 1.3.8 the queries per second count didn't drop immediately, it took another 4 hours of slowly going down towards 750 queries per second, and then after exactly 4 hours it dropped to about 400 qps.
      Attached Files

      Comment

      • eran
        Member
        • Apr 2007
        • 36

        #4
        another observation. I left 1.4 to run after all, and i see a very clear pattern on DB usage. Every day at exactly midnight the number of queries per second jump to above 1000, then it slowly goes down during the day until it just again on the next midnight. See attached graph.
        Attached Files

        Comment

        • Alexei
          Founder, CEO
          Zabbix Certified Trainer
          Zabbix Certified SpecialistZabbix Certified Professional
          • Sep 2004
          • 5654

          #5
          How frequently runs ZABBIX housekeeper (HousekeeperFrequency)? Is the QPS an average since MySQL restart or up-to-date value?
          Alexei Vladishev
          Creator of Zabbix, Product manager
          New York | Tokyo | Riga
          My Twitter

          Comment

          • eran
            Member
            • Apr 2007
            • 36

            #6
            In the zabbix_server.conf file I see this parameter, I assume that's what you mean: HousekeepingFrequency=1

            The qps is the avarage queries per second in the previous 60 seconds from when the sample was taken. It reads the queries counter of MySql and uses zabbix' "Delta (speed per second)" configuration.

            Comment

            Working...