Uploaded image for project: 'WiredTiger'
  1. WiredTiger
  2. WT-13185

Improve random cursors for deletion heavy workloads

    • Type: Icon: Improvement Improvement
    • Resolution: Unresolved
    • Priority: Icon: Major - P3 Major - P3
    • None
    • Affects Version/s: None
    • Component/s: APIs
    • Storage Engines
    • StorEng - Defined Pipeline

      Random cursors behave poorly in workloads that have many deletions.  This ticket points out some of the anomalies and some potential fixes and/or improvements.  Part of of random iteration code includes code that selects a random leaf page.  As eviction uses this same random leaf selection, some improvements here may help with eviction for these workloads.

      These workloads may not be uncommon at all.  Consider a collection that contains orders being processed.  An order comes in, a new order is created.  When the order is fulfilled, it is removed from this collection, maybe moved to a different collection of past orders.  If the key is always advancing, then over time get a tree with lots of pages that have no entries, or few entries.

      In the comments will be a program that does a simple simulation of this situation and the surprising results of testing random cursors on this.

      It's possible that multiple smaller tickets will spin off of this investigation.

       

            Assignee:
            backlog-server-storage-engines [DO NOT USE] Backlog - Storage Engines Team
            Reporter:
            donald.anderson@mongodb.com Donald Anderson
            Votes:
            0 Vote for this issue
            Watchers:
            5 Start watching this issue

              Created:
              Updated: