[4.0] Performance regression processing large aggregation commands

XMLWordPrintableJSON

    • Type: Bug
    • Resolution: Fixed
    • Priority: Major - P3
    • 4.0.16
    • Affects Version/s: 4.0.15
    • Component/s: None
    • None
    • Fully Compatible
    • ALL
    • Execution Team 2020-02-10
    • 0
    • None
    • 3
    • None
    • None
    • None
    • None
    • None
    • None

      Due to the backport of SERVER-43910, all aggregation commands now call BSONObj::toString() via redact() on execution. For large BSON objects, this is a very expensive operation.

      This performance regression will be most noticeable for large aggregation commands that execute quickly. This is not related to pipeline stages necessarily, but command object size. For example, $in on a large array will observe this regression.

      This is caused by the absence of changes from SERVER-35912 in 4.0, which changed the inheritance of PipelineCommand from BasicCommand to Command. This performance regression does not exist in 4.2 and master because PipelineCommand does not inherit from BasicCommand, which calls into redact() and BSONObj::toString().

            Assignee:
            Benety Goh
            Reporter:
            Louis Williams
            Votes:
            0 Vote for this issue
            Watchers:
            24 Start watching this issue

              Created:
              Updated:
              Resolved: