Performance drop after migrating from 2.4 to 2.6

XMLWordPrintableJSON

    • Type: Bug
    • Resolution: Cannot Reproduce
    • Priority: Major - P3
    • None
    • Affects Version/s: 2.6.8, 2.6.9
    • Component/s: Replication
    • None
    • ALL
    • None
    • 3
    • None
    • None
    • None
    • None
    • None
    • None

      Immediately after migrating a mongodb replica set from 2.4.5 to 2.6.9 we observed an increased lock % on all members, even on secondaries that were only synchronizing without serving queries.
      Our replica set is composed by 4 nodes -> 1 primary, 2 slaves and an hidden one. We do not use sharding. We rolled back all nodes except a slave in order to troubleshoot and diagnose the issue.
      We attached iostats output and mongo mms (please consider interval 09-11 AM) graphs collected at the same time on the two slave nodes with different mongo versions.
      As you can see the node with mongo 2.6 shows increased lock % and background flush avg. Also the amount of data written to the recovery log and database datafiles per second are much more bigger on mongo 2.6.

        1. 500k_populator.js
          0.2 kB
        2. 500k_updates_benchRun.js
          0.4 kB
        3. iostats.xlsx
          18 kB
        4. mongo_replicaset_test_results.xlsx
          9 kB
        5. mongo-2.4.5-secondary.tar.gz
          1 kB
        6. mongo-2.6.9-primary.tar.gz
          5 kB
        7. mongo-2.6.9-secondary.tar.gz
          1 kB
        8. mongo24.png
          mongo24.png
          297 kB
        9. mongo26.png
          mongo26.png
          271 kB
        10. mongod.conf
          0.2 kB
        11. mongostat.png
          mongostat.png
          307 kB
        12. replicaset-configure.js
          0.4 kB

            Assignee:
            Sam Kleinman (Inactive)
            Reporter:
            Marzio Rizzo
            Votes:
            0 Vote for this issue
            Watchers:
            5 Start watching this issue

              Created:
              Updated:
              Resolved: