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

Merged

littlefox merged 12 commits from issue-287-perf-budgets into main 2026-10-23 17:10:24 +00:00

Conversation 0 Commits 12 Files changed 9 +248 -8

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

Owner

Copy link

Closes #287

CI fails when the gains of 0.6 slip back:

• tools/perf-budget.py RESULTS.json [BUDGET.json] checks a perf run against e2e/perf/budget.json ("scenario.metric": {max, why}); over a budget, a failed scenario or a missing metric fails. Its tests (tools/test_perf_budget.py: passes within, fails over/failed/missing, the budget file parses) run in the check job.
• Budgets from main today (tools/perf.sh main-07): channels listed at login, calls at login, chain p95/slowest step and calls, one-offs p95, frames over 50 ms reading by scrolling and scrolling a thread, idle calls, calls paging back a channel, the giant thread (#285), wasm memory, DOM nodes and all calls at the end. Counts ~20 % headroom; times ~2.5x until CI's own numbers are in.
• .forgejo/workflows/perf.yml: nightly, by hand, and on pull requests labelled perf (this one is). tools/perf.sh runs the check locally too.

The time budgets will likely be tightened once a few CI runs show the shared runners' numbers.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

Closes #287 CI fails when the gains of 0.6 slip back: - `tools/perf-budget.py RESULTS.json [BUDGET.json]` checks a perf run against `e2e/perf/budget.json` (`"scenario.metric": {max, why}`); over a budget, a failed scenario or a missing metric fails. Its tests (`tools/test_perf_budget.py`: passes within, fails over/failed/missing, the budget file parses) run in the `check` job. - Budgets from main today (`tools/perf.sh main-07`): channels listed at login, calls at login, chain p95/slowest step and calls, one-offs p95, frames over 50 ms reading by scrolling and scrolling a thread, idle calls, calls paging back a channel, the giant thread (#285), wasm memory, DOM nodes and all calls at the end. Counts ~20 % headroom; times ~2.5x until CI's own numbers are in. - `.forgejo/workflows/perf.yml`: nightly, by hand, and on pull requests labelled `perf` (this one is). `tools/perf.sh` runs the check locally too. The time budgets will likely be tightened once a few CI runs show the shared runners' numbers. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added 1 commit 2026-10-23 00:49:21 +00:00

CI: speed and memory budgets from the perf run (#287) ... 5fb513b157

0.7 rewrites the channel and thread views; CI should notice when what
0.6 gained slips back. tools/perf-budget.py checks a perf run against
e2e/perf/budget.json (server calls, DOM nodes, wasm memory, slow frames,
step times; main's numbers with headroom). The perf run is too slow for
every push, so a perf workflow runs it nightly, by hand, and on pull
requests labelled perf. The check's tests run in the check job.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added the

perf

label 2026-10-23 00:49:26 +00:00

littlefox added 1 commit 2026-10-23 01:10:50 +00:00

Perf budgets: times from CI's first run ... a0cdaebd3c

The first perf run on CI (#499) was within every budget but the calls
while idle: a slower runner still finishes loads during those 20 s (4
calls, not 1). Times are now about 1.3x CI's, not 2.5x the dev
machine's. The results upload as logs-perf, which tools/forge.py logs
fetches.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added 1 commit 2026-10-23 01:24:24 +00:00

Perf: the end snapshot before the giant thread step ... 9e73f22573

The end numbers (DOM nodes, memory, calls) came right after opening the
giant thread as a chat, so they counted that chat, as far as it had
loaded: 11 000 nodes in one CI run, 16 800 in the next. Taken before it,
they stay comparable with 0.6's runs.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added 1 commit 2026-10-23 01:25:56 +00:00

Merge main into issue-287-perf-budgets a98051d9bf

littlefox added 1 commit 2026-10-23 01:48:32 +00:00

Perf: close the search before the end snapshot ... fccee8b68c

The end's DOM nodes counted the search hits left open, and how many
there are depends on what the search index happened to gather during
the run: 11 000, 16 800 and 15 100 nodes in three CI runs of the same
code. Closed again, end measures the app itself.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added 1 commit 2026-10-23 02:01:54 +00:00

Perf budgets: room for how CI's login time varies ... ec40967a66

Four CI runs of the same code listed all channels at login in 8.9 to
13.9 s, as busy as the shared runner was; the DOM at the end was 11 000
to 13 200 nodes. The budgets now allow that, still far below what a
real regression (a list that stops windowing) would show.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added 1 commit 2026-10-23 10:06:08 +00:00

Merge main into issue-287-perf-budgets e2870ba61a

littlefox added 2 commits 2026-10-23 11:55:54 +00:00

Merge remote-tracking branch 'origin/main' into issue-287-perf-budgets f2fb9bfafe

Merge main into issue-287-perf-budgets; chain budgets after #305 ... ccdb2316a2

#305 fetches the next 200 texts of a thread walk at once: fewer calls a
step afterwards, but one burst near step 180, so 300 steps of the long
chain take 1092 calls here instead of 925. The slowest step stays 139 ms
here; on CI one run (#536) had a 434 ms step with the p95 at 207, so the
max, a single step, gets more room and the p95 stays the steady guard.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox added 1 commit 2026-10-23 14:01:32 +00:00

Merge remote-tracking branch 'origin/main' into issue-287-perf-budgets ... 7ac76bd8bb

# Conflicts:
#	docs/DESIGN.md

littlefox referenced this pull request 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 pull request 2026-10-23 14:33:25 +00:00

Thread view: fetch only the stats it will use when walking down #312

littlefox referenced this pull request 2026-10-23 16:21:46 +00:00

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

littlefox added 2 commits 2026-10-23 16:22:01 +00:00

Merge remote-tracking branch 'origin/main' into issue-287-perf-budgets 7305bb6fa4

Merge main into issue-287-perf-budgets; chain budgets back down ... a951ca96ba

With #311 the walk down no longer fetches 400 texts at once: the long
chain's slowest step and its calls are back near where the budgets were
set, so they get their old room back.

Refs #287

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XJW2ocwLunyaLiUwCBHEP8

littlefox merged commit b23ca6ebba into main 2026-10-23 17:10:24 +00:00

littlefox referenced this pull request from a commit 2026-10-23 17:10:26 +00:00

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

littlefox referenced this pull request 2026-10-23 17:48:09 +00:00

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

littlefox referenced this pull request 2026-10-24 13:23:54 +00:00

Release 0.7.0 #328

littlefox referenced this pull request from a commit 2026-10-24 13:23:56 +00:00

Release 0.7.0

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

area/ci

Forgejo Actions, runner, tests

  

area/desktop

Desktop build

  

area/packaging

Arch, Debian, Fedora, NixOS packages

  

area/protocol

kom-proto, wire format, charset

  

area/reading

Unread tracking, reading order, review

  

area/ui

Dioxus app, layout, interaction

  

area/web

Web build and relay

  

area/writing

Composing and sending texts

  

kind/bug

Something is broken

  

kind/feature

New functionality

  

kind/infra

Tooling, tests, CI, dev environment

  

kind/request

requests for features or changes from users

  

needs/user

Waiting for a decision or action from littlefox

  

parked

Kept for reference; no work planned until the user says so

  

perf

Run the perf workflow and its budgets on this pull request (#287)

  

security

Security-relevant

  

upstream

A proposal for Protocol A or lyskomd, for the LysKOM development group

No labels

area/ci

area/desktop

area/packaging

area/protocol

area/reading

area/ui

area/web

area/writing

kind/bug

kind/feature

kind/infra

kind/request

needs/user

parked

perf

security

upstream

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

littlefox/neokom!303

No description provided.