Repository navigation
Benchmark regression: open (main) #561
Description
Bencher flagged these benchmarks in the nightly run (open, main) on two runs in a row:
open/wal_replay/100k
Branch: main
Run: https://github.com/project-minigraf/minigraf/actions/runs/38225035471
Previous run: https://github.com/project-minigraf/minigraf/actions/runs/38124422311
Please review the Bencher dashboard to confirm the regression.
Activity
[-]Benchmark regression: open[/-][+]Benchmark regression: open (main)[/+]adityamukho commentedon Oct 13, 2026
Not a code regression. Main stepped at
0da66eb3..a861e551; bisected locally to #540, the v2.x backport of #476. Before it, dropping awal_checkpoint_threshold(usize::MAX)handle still saved the.graph, sopopulate_file_no_checkpointleft a fully checkpointed file.checkpointthen took the nothing-to-do early return, andwal_replayreplayed 0 entries. Local opt-z bench at0edfbe38→99410b1a:checkpoint/1k109 µs → 11.2 ms,open/wal_replay/1k5.09 ms → 0.39 ms. The new numbers are the real cost of the work these benches are named for (v2.x checkpoint is a v7 rewrite; v3'sopen/wal_replay/100kis 104 ms). Bencher kept alerting because its t-test used all history; #568 limits it to the last 30 runs. Closing once #568 is merged.🤖 Generated with Claude Code
ci(bench): compare each run with the last 30 runs, not all history (#568
adityamukho commentedon Oct 13, 2026
Closing: not a code regression (see above). #568 merged; Bencher now compares each run with the last 30 runs of its branch.
🤖 Generated with Claude Code