littlefox/neokom

Watch

Fork 1

CI: speed and memory budgets from the perf run #287

New issue

Closed

opened 2026-10-22 14:41:28 +00:00 by littlefox · 1 comment

littlefox commented

2026-10-22 14:41:28 +00:00

Owner

Approved for 0.7 (D-063).

CI already fails when the web download outgrows its budget (tools/wasm-size.sh). 0.7 rewrites the channel and thread views, so CI should also fail when the gains measured in 0.6 slip back. Those gains are in the 0.6.0 changelog, measured with tools/perf.sh on the big seeded meeting.

Budgets, a little above main-6:

all channels listed at login;
the p95 step and the slowest step when reading in a chain;
wasm memory and DOM nodes at the end;
frames over 50 ms while scrolling;
server calls at login and in the chain.

How:

tools/perf-budget.py checks a results file against e2e/perf/budget.json.
The perf spec runs in CI. If that's too slow per push, it runs nightly next to the soak test and on PRs labelled perf.
A budget is raised in a commit that says why.

Tests: the check itself, on a results file that passes and on one that fails.

Approved for 0.7 (D-063). CI already fails when the web download outgrows its budget (`tools/wasm-size.sh`). 0.7 rewrites the channel and thread views, so CI should also fail when the gains measured in 0.6 slip back. Those gains are in the 0.6.0 changelog, measured with `tools/perf.sh` on the big seeded meeting. **Budgets**, a little above main-6: - all channels listed at login; - the p95 step and the slowest step when reading in a chain; - wasm memory and DOM nodes at the end; - frames over 50 ms while scrolling; - server calls at login and in the chain. **How:** - `tools/perf-budget.py` checks a results file against `e2e/perf/budget.json`. - The perf spec runs in CI. If that's too slow per push, it runs nightly next to the soak test and on PRs labelled `perf`. - A budget is raised in a commit that says why. Tests: the check itself, on a results file that passes and on one that fails.

littlefox added this to the 0.7: Threads milestone

2026-10-22 14:41:28 +00:00

littlefox added the

kind/infra

area/ci

labels

2026-10-22 14:41:28 +00:00

littlefox referenced this issue

2026-10-22 14:41:45 +00:00

Threads 1d: giant threads open at the first unread, with a timeline #285

littlefox referenced this issue

2026-10-22 14:41:45 +00:00

Remember each text's thread root on the device #288

littlefox referenced this issue

2026-10-22 14:41:47 +00:00

Channels as lists of threads, threads as chats (Teams-style) #242

littlefox referenced this issue

2026-10-22 23:41:47 +00:00

Giant threads: first unread, where in it, a timeline (#242 step 1d) #301

littlefox referenced this issue

2026-10-23 00:43:57 +00:00

Remember each text's thread root on the device #288

littlefox referenced this issue from a commit

2026-10-23 00:49:06 +00:00

CI: speed and memory budgets from the perf run (#287)

littlefox referenced this issue from a pull request that will close it,

2026-10-23 00:49:21 +00:00

CI: speed and memory budgets from the perf run (#287) #303

littlefox referenced this issue from a commit

2026-10-23 01:10:50 +00:00

Perf budgets: times from CI's first run

littlefox referenced this issue from a commit

2026-10-23 01:24:24 +00:00

Perf: the end snapshot before the giant thread step

littlefox referenced this issue from a commit

2026-10-23 01:48:32 +00:00

Perf: close the search before the end snapshot

littlefox referenced this issue from a commit

2026-10-23 02:01:54 +00:00

Perf budgets: room for how CI's login time varies

littlefox referenced this issue

2026-10-23 07:39:15 +00:00

Remember each text's thread root on the device #288

littlefox referenced this issue from a commit

2026-10-23 11:55:54 +00:00

Merge main into issue-287-perf-budgets; chain budgets after #305

littlefox referenced this issue

2026-10-23 14:31:53 +00:00

Thread view: a burst of 400 calls stalls a step on a slow CPU #311

littlefox referenced this issue from a commit

2026-10-23 14:33:11 +00:00

Thread view: fetch only the stats it will use when walking down

littlefox referenced this issue from a commit

2026-10-23 16:22:01 +00:00

Merge main into issue-287-perf-budgets; chain budgets back down

littlefox closed this issue

2026-10-23 17:10:24 +00:00

littlefox referenced this issue from a commit

2026-10-23 17:10:25 +00:00

Merge pull request 'CI: speed and memory budgets from the perf run (#287)' (#303) from issue-287-perf-budgets into main

littlefox commented

2026-10-23 17:48:09 +00:00

Author

Owner

Verified, merged in #303; main green (run #561), perf run #560 green.

.forgejo/workflows/perf.yml: nightly, by hand, and on pull requests labelled perf. It runs the perf spec (big seeded meeting, 30 ms each way), then tools/perf-budget.py checks the results against e2e/perf/budget.json. Results are uploaded as logs-perf.
tools/test_perf_budget.py (run in check): a value over budget fails, a failed scenario fails, a missing metric is reported.
The budgets did their job before merging: they caught a slow step on CI (434, then 633 ms) that came from #305. That was fixed in #311; on CI the slowest step is now 271 ms (budget 320), with 972 calls (budget 1170).
New for finding such things locally: PERF_CPU=4 tools/perf.sh LABEL slows the browser's CPU like CI's runners.

Verified, merged in #303; main green (run #561), perf run #560 green. - `.forgejo/workflows/perf.yml`: nightly, by hand, and on pull requests labelled `perf`. It runs the perf spec (big seeded meeting, 30 ms each way), then `tools/perf-budget.py` checks the results against `e2e/perf/budget.json`. Results are uploaded as `logs-perf`. - `tools/test_perf_budget.py` (run in `check`): a value over budget fails, a failed scenario fails, a missing metric is reported. - The budgets did their job before merging: they caught a slow step on CI (434, then 633 ms) that came from #305. That was fixed in #311; on CI the slowest step is now 271 ms (budget 320), with 972 calls (budget 1170). - New for finding such things locally: `PERF_CPU=4 tools/perf.sh LABEL` slows the browser's CPU like CI's runners.

littlefox referenced this issue

2026-10-23 17:48:23 +00:00

Channels as lists of threads, threads as chats (Teams-style) #242

littlefox referenced this issue

2026-10-23 19:59:48 +00:00

forge.py pr-merge: a skipped run doesn't block a merge #319

littlefox referenced this issue from a commit

2026-10-23 19:59:48 +00:00

forge.py pr-merge: a skipped run doesn't block a merge

littlefox referenced this issue

2026-10-24 13:23:54 +00:00

Release 0.7.0 #328