Issue Summary

The benchmarks_sep performance test is currently unstable and is blocking the commit queue for changes that do not impact performance-critical code paths. A recent example involved a unit test modification being blocked due to a regression detected in BM_FIND_ONE, likely caused by a previous commit pushing performance to the threshold. This can lead to false positives and unnecessary investigations.

Context

• The thresholds for blocking commits are intentionally set high (see performance_thresholds.yml) to avoid blocking developers.
• The comparison logic appears to be flawed: in the commit queue, changes are being compared as 'master + patch' vs. an older base commit (e.g., 08d10bb), rather than 'patch on base commit' vs. base commit. This is especially problematic for PRs based on older commits, resulting in misleading performance regressions.
• Recent pain points in the channel are attributed to this logic, and a previous change intended to compare against the PR's base commit does not seem to be working as expected.

Proposed Solution

• Investigate and correct the comparison logic so that performance tests compare the patch applied to its base commit, not to master or an unrelated commit.
• Verify if the intended logic (comparing to the PR's base commit) is implemented and functioning.
• Consider disabling commit blocking on benchmarks_sep until the logic is fixed to prevent further developer disruption.

Original Slack thread: https://mongodb.slack.com/archives/C08D8RX2K5H/p1766018451839409
This ticket was generated by AI from a Slack thread.