Uploaded image for project: 'WiredTiger'
  1. WiredTiger
  2. WT-2764

Optimize checkpoints to reduce throughput disruption

    XMLWordPrintableJSON

Details

    • Bug
    • Status: Closed
    • Major - P3
    • Resolution: Fixed
    • None
    • WT2.9.0, 3.2.10, 3.3.11
    • None
    • None

    Description

      There are a large set of changes in this ticket including:

      • Release checkpoint transaction before syncing.
      • Replace write leaves phase of checkpoints with a mode that steps down the dirty content in cache using eviction.
      • Change the default eviction dirty trigger (20) and target (5)

      While measuring the workload in WT-2747, I noticed that transaction IDs stayed pinned while a checkpoint was waiting for updates to sync.

      This dates back to WT-1982 and SERVER-18316: the checkpoint's transaction snapshot is released before the sync, but that no longer clears the global "checkpoint pinned" transaction ID.

      Attachments

        1. 2764.ckpt.tar.gz
          51 kB
          Susan LoVerso

        Issue Links

          Activity

            People

              michael.cahill@mongodb.com Michael Cahill
              michael.cahill@mongodb.com Michael Cahill
              Votes:
              0 Vote for this issue
              Watchers:
              6 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: