ExportXMLWordPrintableJSON

    • Type: Improvement
    • Resolution: Unresolved
    • Priority: Major - P3
    • None
    • Affects Version/s: None
    • Component/s: None
    • None
    • Storage Engines - Server Integration
    • SESINEXT Hi-Jack-ed 2026-12-15
    • None
    • None
    • None
    • None
    • None
    • None
    • None

      This was surfaced from HELP-100704 where SLS resolved page server addresses that didn't resolve ("Domain name not found"). This was an SLS bug but while it was broken, mongod burned 100% CPU on every read

      The sendPageRequestSync loop is while(true) bounded only by the read's totalDeadline and has no sleep in order to keep page reads fast. The exponential backoff that exists is per page server in the writer thread but it fails fast. While a server is in backoff requests are rejected immediately RetryBackoff instead of waiting so the caller loops straight back and retries the next candidate, essentially hot looping when the scenario occurs that all servers are down. queue -> instantly retryBackoff -> 10985387 warn -> retry and deadline expires and we fatal.

      During connect failures the backoff never really accumulates. The backoff never really accumulates. We should add a small backoff/sleep in the page read retry path for connect time failures. Consider eg after 3x RetryBackoff per server sleep ~50ms

            Assignee:
            Unassigned
            Reporter:
            Clarisse Cheah
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

              Created:
              Updated: