ExportXMLWordPrintableJSON

    • Type: Improvement
    • Resolution: Unresolved
    • Priority: Major - P3
    • None
    • Affects Version/s: None
    • Component/s: None
    • None
    • Storage Engines - Server Integration
    • SESIKhuuBeanz 2026-10-06, SESIroflcopter 2026-11-03, SESIWhinery 2026-11-17, SESINickBeckenhome 2026-12-01
    • None
    • None
    • None
    • None
    • None
    • None
    • None

      (Previous this was a ticket in SLS-11996, as this is a PALI request I am create a SERVER ticket here. See some initial discussion as comments on SLS-11996)

      Coming as part of investigation done in https://jira.mongodb.org/browse/SLS-11818 

      Phylog/checkpoint write bursts can reach ~691 MiB/s and coincide with increased peer-message delivery latency and elevated p99 commit latency. Lower bursts around 430–462 MB/s do not show the same impact.

      Investigate changes in PALI/log-service scheduling or QoS to prioritize oplog traffic over phylog writes. PALI should spread, defer, or throttle phylog writes during contention so they do not interfere with oplog commit and materialization.

      Acceptance Criteria

      • Oplog traffic receives higher priority than phylog traffic under contention.
      • Phylog writes cannot starve oplog commit or materialization.
      • Validate using the TSBS workload on M60 compute with burst traffic.
      • Compare p99 commit latency, peer-message delivery latency, queue depth, and phylog catch-up time before and after the change.
      • Add metrics showing phylog throttling, queueing, and the impact on oplog latency.

            Assignee:
            Aaron Balsara
            Reporter:
            Sunil Narasimhamurthy
            Votes:
            0 Vote for this issue
            Watchers:
            6 Start watching this issue

              Created:
              Updated: