Skip to content

Benchmark regression: query #487

Closed as not planned
Closed as not planned

Description

@github-actions

Bencher flagged these benchmarks in the nightly run (query) on two runs in a row:

  • query/predicate_pushdown/100k

Run: https://github.com/project-minigraf/minigraf/actions/runs/37749964713
Previous run: https://github.com/project-minigraf/minigraf/actions/runs/37592142842

Please review the Bencher dashboard to confirm the regression.

Activity

  1. adityamukho commented on Oct 8, 2026

    @adityamukho
    Collaborator

    Not a regression. The jump on query/predicate_pushdown/100k (≈195 ms → ≈710 ms, first seen on the 2026-10-06 run) comes from the #445 fix in v2.0.3 (6739a94).

    The 100k fixture writes 200,000 uncheckpointed in-memory facts. Before #445, every pending fact past position 65,535 shared one index slot, and every row this query should return (?v > 90000, facts past position 180,000) was in that range. So the old benchmark timed an empty result:

    time rows
    before #445 fix (6739a94^) 198 ms 0 (wrong)
    after #445 fix (6739a94) 709 ms 9,999 (correct)

    1k and 10k sizes (under 65,535 facts) are unchanged. ~710 ms for 10k rows is in line with 10k at ~56 ms for 1k rows. v3 measures ~533 ms. The new number is the correct baseline; Bencher should settle once its history catches up.

    Closing as not a regression.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    performanceBenchmark performance regressions

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions