Uploaded image for project: 'Compass '
  1. Compass
  2. COMPASS-8271

Investigate changes in SERVER-93576: Remove vector embeddings from $vectorSearch explain

    • Type: Icon: Investigation Investigation
    • Resolution: Done
    • Priority: Icon: Major - P3 Major - P3
    • No version
    • Affects Version/s: None
    • Component/s: None
    • None
    • Developer Tools
    • Not Needed

      Original Downstream Change Summary

      This change redacts the queryVector's embedding fields in the explain output (if the field opts.verbosity is present).

      Description of Linked Ticket

      Currently, when running `explain` on a vector search, all the vector embeddings are included. 

      These can be and often are extremely lengthy. E.g., the vector embeddings for mflix data look like this, where there are over 1000 elements. This makes the explain output extremely cumbersome to read.

      When running explain on a basic $unionWith $vectorSearch (as part of SERVER-88046, so using the real mongot rather than a mock):

      On a sharded cluster:

      • The explain output on a is over 9000 lines long.
      • The actual vector we're searching for has its embeddings included 6 separate times in the output (which is questionable in and of itself), and so the length of explain reduces to ~420 lines when you remove the embeddings.

      On a single node:

      • Explain output is ~4800 lines long.
      • The vector embedding is included 3 times, length reduces to ~200 lines without the embeddings.

            Assignee:
            Unassigned Unassigned
            Reporter:
            backlog-server-pm Backlog - Core Eng Program Management Team
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

              Created:
              Updated:
              Resolved: