You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There is no code path change between the passing baseline and the current main that could affect insert_file latency.
Conclusion
The Bencher threshold was crossed because the baseline run happened to land on a faster runner, and this run hit scheduling jitter (evidenced by multiple high-severe outliers across measurements). The next nightly run should return to baseline levels.
Closing as false positive. The Bencher alerts can be dismissed.
Activity
adityamukho commentedon Apr 17, 2026
Analysis: False positive — runner variability, not a code regression
After investigation, this is not a genuine performance regression. The alert was triggered by GitHub Actions runner jitter, not by any code change.
Evidence
The benchmark numbers themselves are incoherent:
insert_file/single_fact/1k: confidence interval [426 µs … 1.6 ms] — a 4× spread with 1 high-severe outlier flagged by Criterioninsert_file/single_fact/1kmeasured slower thaninsert_file/single_fact/10k— physically impossible for the same operation at smaller scaleinsert_file/single_fact/1mmeasured faster thaninsert_file/single_fact/100k— same issueThis pattern is classic OS/runner scheduling noise, not a workload characteristic.
No code change affects insert performance:
The Bencher baseline was established from run 24553303618, which ran before PRs #144, #145, #146 were merged. The commits merged since then are:
positionsHashMap in LRU cache — caused panics, was reverted by fix(storage): remove stale positions HashMap from LRU cache #147touch()to its pre-perf: Track VecDeque positions in HashMap for O(1) LRU touch #146 O(N)iter().position()implementation — identical algorithm to the baselineThere is no code path change between the passing baseline and the current
mainthat could affectinsert_filelatency.Conclusion
The Bencher threshold was crossed because the baseline run happened to land on a faster runner, and this run hit scheduling jitter (evidenced by multiple high-severe outliers across measurements). The next nightly run should return to baseline levels.
Closing as false positive. The Bencher alerts can be dismissed.