Benchmark regression - 2026-06-30

Bencher detected a performance regression in the nightly benchmark run (query).

Run: https://github.com/project-minigraf/minigraf/actions/runs/28424237003

Please review the Bencher dashboard for details on which benchmarks regressed.

Result: runner noise, not a real regression

Bencher raised one alert in this run: query/predicate_pushdown/100k. All other query benchmarks stayed under the threshold.

The flagged commit (cbf899ea) was the head of main for every nightly run from Jun 29 to Aug 6. So the nights around Jun 30 ran the same code:

Night	Commit	predicate_pushdown/100k	Alert?
Jun 29	`cbf899e`	238 ms	no
Jun 30 (flagged)	`cbf899e`	293 ms	yes
Jul 1	`cbf899e`	214 ms	no
Sep 25 (v2.0.1)	`ab65350`	235 ms	no

Why this is noise:

The same code ran faster the night before and the night after. A real regression does not go away without a code change.
The 10k size of the same benchmark moved much less on Jun 30 (77 ms, vs 73 ms and 66 ms on the nights around it). Only the longest benchmark in the job spiked. That fits a slow or busy shared runner.
The latest nightly on v2.0.1 is back at 235 ms with no alert.

No code change needed. Closing.

One side note: the bench workflow does not open a new regression issue while any Benchmark regression issue is open. This issue stayed open for about three months, so alerts in that window did not create new issues. Closing it turns that back on.

Generated by Claude Code

ci: report nightly bench alerts only when they repeat (#401)

Navigation Menu

Repository navigation

Benchmark regression - 2026-06-30 #301

Result: runner noise, not a real regression

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Benchmark regression - 2026-06-30 #301

Description

Activity

adityamukho commented on Sep 26, 2026

Result: runner noise, not a real regression

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions