bench(wa-sqlite): measure scheduler contention - #164
Closed
marcus-pousette wants to merge 2 commits into
Closed
Conversation
marcus-pousette
changed the base branch from
main
to
codex/shared-worker-materialized-fastpath
June 3, 2026 14:38
marcus-pousette
force-pushed
the
codex/runtime-perf-benchmark
branch
3 times, most recently
from
June 4, 2026 06:46
688bee3 to
f5f5a3a
Compare
marcus-pousette
force-pushed
the
codex/runtime-perf-benchmark
branch
from
June 4, 2026 10:43
f5f5a3a to
a551889
Compare
marcus-pousette
force-pushed
the
codex/runtime-perf-benchmark
branch
from
July 12, 2026 15:16
a551889 to
549c543
Compare
marcus-pousette
changed the base branch from
codex/shared-worker-materialized-fastpath
to
fix/shared-worker-priority-scheduler
July 12, 2026 15:16
marcus-pousette
force-pushed
the
codex/runtime-perf-benchmark
branch
from
July 12, 2026 16:40
549c543 to
eafa7e5
Compare
Contributor
BenchmarksTriggered by
Status: ✅ <5% change | 🟡 5–10% | 🟠 10–30% | 🔴 >30% Artifacts:
|
marcus-pousette
force-pushed
the
codex/runtime-perf-benchmark
branch
from
July 12, 2026 18:09
9bf95c7 to
33ebd8d
Compare
marcus-pousette
force-pushed
the
fix/shared-worker-priority-scheduler
branch
2 times, most recently
from
July 12, 2026 19:05
18736d0 to
440acd5
Compare
Collaborator
Author
|
The experiment is complete and its decision-relevant results are preserved in this description and #167. Closing instead of merging avoids adding 443 lines and roughly 66% to every web benchmark checkout; the full artifact remains linked for future reference. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
Experimental raw-browser harness used to evaluate #167's worker RPC scheduler. It measures background remote ingest while foreground reads and normal local writes share the same wa-sqlite client.
The experiment has finished. Its useful evidence is preserved here and in #167; this PR is being closed rather than merged as permanent infrastructure.
What is measured
Normal omits the priority option and preserves FIFO barriers. Background exercises #167's foreground bypass. Direct memory is the no-scheduler control.
Avoiding misleading fast paths
The first run exposed the independent OPFS path-capacity bug fixed by #213: wa-sqlite reported 64 bytes, while one filename plus SQLite's
-journalsuffix was 66 bytes. It was deterministicSQLITE_CANTOPEN, not a teardown race. The false cooldown/workaround was removed.Results
Focused run 29203415286 completed all 64/64 runtime rows. Values below are background-vs-normal deltas between four-trial medians; negative latency is better.
The useful result is narrow and repeatable: every empty-tree trial improved dedicated-worker read p95 (memory: 11.3–14.6%; OPFS: 14.4–17.9%). Populated cases stayed near neutral. The largest median latency regression was 4.0% remote-batch p95.
The shared-worker case used one client/port, so it did not exercise the distinct cross-tab/global scheduler. That finding led #167 to remove shared-worker scheduling and retain only the validated dedicated-worker path. Four trials are directional evidence, not a claim of statistical certainty.
durationMswas captured after the final integrity read, so it dilutes the derived throughput comparison. The latency samples and exact final-count validation are unaffected; #167 deliberately does not use the throughput field.Scope and validation
Why close instead of merge