ExportXMLWordPrintableJSON

    • Type: Technical Debt
    • Resolution: Unresolved
    • Priority: Minor - P4
    • None
    • Affects Version/s: None
    • None
    • Storage Engines - Persistence
    • 706.491
    • StorEng - Defined Pipeline
    • None

      Problem

      On 7.0, checkpoint-cleanup: most recent duration on all eligible files (usecs) (cc_duration) is zeroed at each checkpoint start and accumulates cleanup time for the whole checkpoint interval (inline checkpoint cleanup plus every utility-thread pass). On 8.0.21+, 8.3, 9.0 and develop the same stat is set once per full cleanup pass.

      Evidence: in 7.0.45 FTDC, all 70 resets land on the same sample as a checkpoint-lock acquisition, i.e. the value resets at checkpoint start and grows continuously until the next one.

      Confusion

      The name and description promise the duration of the most recent cleanup pass, and FTDC/Atlas dashboards read it that way ("Cleanup Pass Duration"). On 7.0 it instead reports seconds of accumulated cleanup time, so the metric cannot be used to read per-pass duration and is not comparable with other versions under the same key.

      Definition of done

      • Choose and implement one:
        • Keep the 7.0 per-checkpoint measurement, but give the stat an accurate description and update the consumers; or
        • Align 7.0 with develop: WT_STAT_CONN_SET one full __checkpoint_cleanup_int() pass and remove the per-page increment and per-checkpoint reset.
      • Remove the unused per-data-source copies of cc_duration / cc_handle_processed.
      • Add a test that triggers cleanup repeatedly and locks in the chosen semantics; it must fail on the current 7.0 code.

        1. test_cc07.py
          3 kB
        2. test_cc08.py
          4 kB

            Assignee:
            Etienne Petrel
            Reporter:
            Etienne Petrel
            Votes:
            0 Vote for this issue
            Watchers:
            1 Start watching this issue

              Created:
              Updated: