Uploaded image for project: 'ZABBIX BUGS AND ISSUES'
  1. ZABBIX BUGS AND ISSUES
  2. ZBX-16517

Incorrect item check scheduler

XMLWordPrintable

    • Icon: Problem report Problem report
    • Resolution: Won't fix
    • Icon: Major Major
    • None
    • 4.0.11
    • Server (S)
    • None

      Steps to reproduce:

      1. Install a clean Debian 9 OS. Mine will be 9.8.0 amd64.
      2. Follow the installation steps. By the time of writing this report, zabbix-release_4.0-3+stretch_all.deb repository package was already released, whereas documentation lists instructions for version 4.0-2. I have made the upgrade during installation, in case it matters.
      3. Once Zabbix Server, Zabbix Agent and Zabbix Web Interface components have been installed, navigate to web interface and create:
        1 host group for your tests. This can be skipped; is done only for filtering convenience.
        1 template containing a single item check with update interval 1 hour. This item has icmpping[google.com] key.
        5 hosts as members of the test host group and with the test template attached to them.
        You can find template and hosts export in attachments for testing convenience.
      4. Wait a couple hours.

      Result:
      Here are results I have had produced:

        HOST 1 HOST 2 HOST 3 HOST 4 HOST 5
      Time of check 1 2019-08-14 08:00:02 2019-08-14 08:00:03 2019-08-14 08:00:04 2019-08-14 08:00:05 2019-08-14 08:00:06
      Time of check 2 2019-08-14 09:00:02 2019-08-14 09:00:03 2019-08-14 09:00:04 2019-08-14 09:00:05 2019-08-14 09:00:06
      Time of check 3 2019-08-14 10:00:02 2019-08-14 10:00:03 2019-08-14 10:00:04 2019-08-14 10:00:05 2019-08-14 10:00:06

      You can also find poller load graph for this period in attachments.

      Expected:
      I remember during Zabbix trainings for version 3.0 we were told Zabbix Server automatically analyzes the amount of hosts, items and their update intervals to automatically allocate checks in time (this would only work in case custom intervals are not used, of course). This ensured Zabbix Server does not perform checks in batches, providing best performance. I have not read all the upgrade notes of all versions between 3.0 and 4.0, but I do believe this is an essential Zabbix feature not worth removing.
      With this being said, I expected the checks to happen evenly thoughout the time: approximately one check per 12 minutes. Not 5 checks per 5 seconds in 1 hour interval as it happens now - there is no scheduling in that.

      I have observed this for quite some time on our pre-production environment, where checks are performed once per 12 hours. You can find a screenshot of poller load in attachments. It shows poller load before we detected and issue and after we disabled the most resource-consuming one. That environment has exactly same symptoms of check intervals being completely off.

      Let me know if you would require additional information.

        1. 16517.ods
          142 kB
        2. pre-prod_pollers.png
          pre-prod_pollers.png
          67 kB
        3. test_pollers.png
          test_pollers.png
          61 kB
        4. zbx_export_hosts.xml
          7 kB
        5. zbx_export_templates.xml
          3 kB

            Unassigned Unassigned
            olegs_vasiljevs Olegs Vasiljevs
            Votes:
            0 Vote for this issue
            Watchers:
            5 Start watching this issue

              Created:
              Updated:
              Resolved: