Uploaded image for project: 'Observium'
  1. Observium
  2. OBS-5300

Poller and discovery times increased ~20x after upgrade from r15142 to r15277

    XMLWordPrintable

Details

    • Bug
    • Resolution: Unresolved
    • Major
    • None
    • None
    • Default

    Description

      ENVIRONMENT

      • Observium Professional, rolling release (SVN trunk)

      VERSIONS

      • Working version before upgrade: 26.9.15142 (3 September 2026)
      • Upgraded to: r15277 (rolling, 5 October 2026)
      • Currently running after rollback: 26.9.15142

      TIMELINE (CEST)

      • Mon 5 Oct 2026, ~12:30: upgraded r15142 -> r15277 (svn up + discovery.php -u)
      • Mon 5 Oct, ~13:00 onwards: poller times rise sharply, see below
      • Tue 6 Oct, ~14:00: rolled back to r15142
      • Tue 6 Oct, ~14:10 onwards: poller times back to normal

      PROBLEM
      Immediately after the upgrade, poller run time increased by more than an order of magnitude. Discovery time followed, and the platform became unresponsive and unusable. The configuration was not changed during the upgrade.

      1) Global "Poller Total time" graph (screenshot)

      • Normal baseline for the whole previous week: approx. 200-250 s (min 183.8 s)
      • After upgrade: approx. 5,000-6,500 s (max 6,493 s)
      • The graph shows gaps after the upgrade, i.e. poll cycles were missed
      • After rollback: 226 s

      2) Per-device poller times, example device (screenshot)

      • On r15277: 20m 56s to 25m 3s per poll, with missing 5-minute cycles
      • After rollback to r15142: 1m 13s, 1m 17s, 1m 45s

      3) Per-device discovery times, same device (screenshot 1)

      • Before upgrade: 29 s to 1m 7s
      • First discovery after upgrade: 1m 33s
      • Subsequent discoveries on r15277: 19m 58s, 21m 41s, 20m 29s, 21m 56s
      • After rollback: back to normal

      ROLLBACK PROCEDURE USED
        cd /opt/observium
        svn up -r 15142
       ./discovery.php u      > "Database is up to date"

      The database schema was not rolled back; r15142 code is running against the schema installed by r15277. Performance returned to normal immediately after the code revert, so the regression appears to be in the code, not in the schema or the data.

       

      QUESTIONS
      1. Which change between r15142 and r15277 could cause this increase in poller and discovery time?
      2. Is there a configuration setting to restore the previous behaviour, so that we can upgrade again?
      3. What debug output would you like from us? We can run poller.php -d against a single device on request.

      ATTACHMENTS
      Discovery Times
      Poller Times
      Poller Total time graph (one week)

      Attachments

        1. screenshot-1.png
          screenshot-1.png
          54 kB
        2. screenshot-2.png
          screenshot-2.png
          30 kB
        3. screenshot-3.png
          screenshot-3.png
          50 kB

        Activity

          People

            landy Mike Stupalov
            quatebriga Marko Zver
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

            Dates

              Created:
              Updated: