Benchmark regression: insert (main) #678

Activity

1. github-actions

   added

   performanceBenchmark performance regressionsBenchmark performance regressions

   on Oct 21, 2026
2. adityamukho commented on Oct 21, 2026

   

   adityamukho

   on Oct 21, 2026

   Collaborator

   More actions

   Not a regression: runner noise.

   A/B on an idle dedicated machine, insert/batch_100, dcbc19e (last fast nightly) vs e434148 (v2.0.9), run in parallel and swapped each round, 4 rounds:

   scale	dcbc19e	e434148	Δ
   1k	525 µs	523 µs	−0.4%
   10k	542 µs	543 µs	+0.1%
   100k	579 µs	580 µs	+0.1%
   1m	575 µs	573 µs	−0.2%

   Criterion found no change in 30 of 32 comparisons; the other two were single 1m rounds on e434148, +8.8% and −7.0%, which cancel.

   Bencher history agrees:

   • The 10-18 nightly ran the same commit as 10-17 (228a087) and was ~1.3× slower; the 10-20 and 10-21 runs sit at that same level.
   • insert_file/batch_100 runs the same parse + transact path and was flat or faster (0.71–0.95×) on those runs.
   • The commits between the runs don't touch the insert path apart from one length check in parse_transact (PRS-108, fix(parser): reject a fact with more than four elements, PRS-108 (#649) [v2.x backport] #656).

   🤖 Generated with Claude Code

   https://claude.ai/code/session_01PtCdzbJMSS1V8qEPaJ4jXz
3. adityamukho

   closed this as not plannedon Oct 21, 2026