Donate to Postulo/postulo

• https://buymeacoffee.com/tiagoagueda

---

See the documentation for more details on donation links.

Cache what every CI job rebuilt, and fetch no map tiles in tests (#1033) #1083

Merged

tiagoagueda merged 1 commit from 1033-ci-workers into main 2026-10-31 02:10:03 +00:00

Conversation 3 Commits 1 Files changed 9 +166 -11

tiagoagueda commented 2026-10-28 18:20:13 +00:00

Owner

• Copy link
• Copy Markdown

Items 2, 3, 6 and 7 of the CI review, now that item 1 runs CI once per
push instead of three times and the host carries one browser and one unit
job at a time.

• The browser job runs on two xdist workers (-n 2 --dist worksteal), the
  unit job on three, in both ci.yml and every-python.yml: five workers on
  six cores. The comments say why two browser workers failed before (run
  723, under three CI runs and a dev image), and that #1024's file-backed
  live-server database was not on main when this changed. Three measured
  green runs decide whether it stays.
• The compiled catalogues are kept between runs, by an exact key over
  every .po, the compiler, the language table and the version; the browser,
  unit and PostgreSQL jobs compile only on a miss.
• Checks clones with filter: blob:none: the guard reads trees for the
  changed paths and fetches the few catalogues it reads. Forgejo
  advertises filter, and the repository is readable without a token.
• Chromium's libraries and fonts, as Playwright 1.62's --with-deps lists
  them for Debian 12 (bar xvfb), join WeasyPrint's in one apt run, and
  playwright install chromium drops --with-deps. Not --only-shell:
  test_web_app.py starts the full Chromium.
• The heavy step names tests/security/test_markdown.py instead of
  collecting the whole suite; a test holds every file with a heavy test to
  that list.
• The test settings draw the map's outline and fetch no tiles; the tests
  about the tiles ask for OpenStreetMap with a new openstreetmap
  fixture.

Refs #1033

Notes for review

Draft until three measured green CI runs, as #1033 asks for items 2 and 3. Each run's wall time and job times get recorded in this PR before it merges.

• Item 2: browser job -n 2 --dist worksteal. The earlier failures under two workers (run 723) came with three CI runs per push and the dev image on the same host, and that load is gone since #1072. #1024's file-backed live-server database (#1074) wasn't on main when this changed.
• Item 3: unit job -n 3, in both ci.yml and every-python.yml.
• Item 6:
  • the compiled catalogues are cached on an exact key over every .po and the tool;
  • the Checks job clones blobless (filter: blob:none; Forgejo advertises it);
  • Chromium's libraries install in the WeasyPrint apt step, with xvfb left out and the fonts kept.
• Item 7:
  • the heavy step names its file, and a test keeps that list honest;
  • tests draw the map with no tiles (POSTULO_MAP_TILES = "none"), with an openstreetmap fixture for the tests that are about tiles.
• Proven only by a CI run: the cache's hit rate and whether the checkout action accepts the blobless filter.

🤖 Generated with Claude Code

Items 2, 3, 6 and 7 of the CI review, now that item 1 runs CI once per push instead of three times and the host carries one browser and one unit job at a time. - The browser job runs on two xdist workers (`-n 2 --dist worksteal`), the unit job on three, in both ci.yml and every-python.yml: five workers on six cores. The comments say why two browser workers failed before (run 723, under three CI runs and a dev image), and that #1024's file-backed live-server database was not on main when this changed. Three measured green runs decide whether it stays. - The compiled catalogues are kept between runs, by an exact key over every .po, the compiler, the language table and the version; the browser, unit and PostgreSQL jobs compile only on a miss. - Checks clones with `filter: blob:none`: the guard reads trees for the changed paths and fetches the few catalogues it reads. Forgejo advertises `filter`, and the repository is readable without a token. - Chromium's libraries and fonts, as Playwright 1.62's `--with-deps` lists them for Debian 12 (bar xvfb), join WeasyPrint's in one apt run, and `playwright install chromium` drops `--with-deps`. Not `--only-shell`: test_web_app.py starts the full Chromium. - The heavy step names tests/security/test_markdown.py instead of collecting the whole suite; a test holds every file with a heavy test to that list. - The test settings draw the map's outline and fetch no tiles; the tests about the tiles ask for OpenStreetMap with a new `openstreetmap` fixture. Refs #1033 ## Notes for review **Draft until three measured green CI runs,** as #1033 asks for items 2 and 3. Each run's wall time and job times get recorded in this PR before it merges. - **Item 2:** browser job `-n 2 --dist worksteal`. The earlier failures under two workers (run 723) came with three CI runs per push and the dev image on the same host, and that load is gone since #1072. #1024's file-backed live-server database (#1074) wasn't on `main` when this changed. - **Item 3:** unit job `-n 3`, in both `ci.yml` and `every-python.yml`. - **Item 6:** - the compiled catalogues are cached on an exact key over every `.po` and the tool; - the Checks job clones blobless (`filter: blob:none`; Forgejo advertises it); - Chromium's libraries install in the WeasyPrint apt step, with xvfb left out and the fonts kept. - **Item 7:** - the heavy step names its file, and a test keeps that list honest; - tests draw the map with no tiles (`POSTULO_MAP_TILES = "none"`), with an `openstreetmap` fixture for the tests that are about tiles. - **Proven only by a CI run:** the cache's hit rate and whether the checkout action accepts the blobless filter. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

tiagoagueda added 1 commit 2026-10-28 18:20:14 +00:00

Run CI's browser tests on two workers and its unit tests on three (#1033) ... 2a6fa3b94c

Items 2, 3, 6 and 7 of the CI review, now that item 1 runs CI once per
push instead of three times and the host carries one browser and one unit
job at a time.

- The browser job runs on two xdist workers (`-n 2 --dist worksteal`), the
  unit job on three, in both ci.yml and every-python.yml: five workers on
  six cores. The comments say why two browser workers failed before (run
  723, under three CI runs and a dev image), and that #1024's file-backed
  live-server database was not on main when this changed. Three measured
  green runs decide whether it stays.
- The compiled catalogues are kept between runs, by an exact key over
  every .po, the compiler, the language table and the version; the browser,
  unit and PostgreSQL jobs compile only on a miss.
- Checks clones with `filter: blob:none`: the guard reads trees for the
  changed paths and fetches the few catalogues it reads. Forgejo
  advertises `filter`, and the repository is readable without a token.
- Chromium's libraries and fonts, as Playwright 1.62's `--with-deps` lists
  them for Debian 12 (bar xvfb), join WeasyPrint's in one apt run, and
  `playwright install chromium` drops `--with-deps`. Not `--only-shell`:
  test_web_app.py starts the full Chromium.
- The heavy step names tests/security/test_markdown.py instead of
  collecting the whole suite; a test holds every file with a heavy test to
  that list.
- The test settings draw the map's outline and fetch no tiles; the tests
  about the tiles ask for OpenStreetMap with a new `openstreetmap`
  fixture.

Refs #1033

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

tiagoagueda commented 2026-10-29 10:03:39 +00:00

Author

Owner

• Copy link
• Copy Markdown

Measured run 1 (run 1658): the browser job on two workers was green.

Unit job on three workers: failed, and the cause is memory, not the tests. The kernel killed two of the three workers at the same second ([gw1]/[gw2] node down: Not properly terminated, about 75 % through). xdist then reported whichever tests they were holding. The job container is capped at 2 GB and each worker grows towards 1 GB (#830), so three don't fit.

The unit job is now back on two workers. Item 3 stays open: it needs either a larger memory cap on the runner's job container (a host setting) or finding what makes each worker grow (for example the full-catalogue loads in test_translations.py). The same memory pressure likely explains #1079's browser job being killed with exit 137.

Still to go: two more measured green runs for item 2.

**Measured run 1 (run 1658):** the browser job on two workers was green. **Unit job on three workers: failed, and the cause is memory, not the tests.** The kernel killed two of the three workers at the same second (`[gw1]`/`[gw2] node down: Not properly terminated`, about 75 % through). xdist then reported whichever tests they were holding. The job container is capped at 2 GB and each worker grows towards 1 GB (#830), so three don't fit. The unit job is now back on **two** workers. **Item 3 stays open**: it needs either a larger memory cap on the runner's job container (a host setting) or finding what makes each worker grow (for example the full-catalogue loads in `test_translations.py`). The same memory pressure likely explains #1079's browser job being killed with exit 137. Still to go: two more measured green runs for item 2.

tiagoagueda force-pushed 1033-ci-workers from 2a6fa3b94c to 26d8bad31f 2026-10-29 10:03:39 +00:00 Compare

tiagoagueda force-pushed 1033-ci-workers from 26d8bad31f to d3579afbaa 2026-10-29 10:32:33 +00:00 Compare

tiagoagueda referenced this pull request 2026-10-29 13:18:15 +00:00

Run the unit tests on three workers, each letting go of the languages it met (#1033) #1085

tiagoagueda commented 2026-10-29 18:39:27 +00:00

Author

Owner

• Copy link
• Copy Markdown

Measured run 2 (run 1691, commit d3579afba): green.

Job	Time
Browser tests (two workers)	24m 0s
Unit tests and coverage (two workers)	14m 28s
Database tests on PostgreSQL	6m 35s
Checks	1m 58s
Security audit	2m 0s

For comparison, the browser suite took 33 to 42 minutes in one process on 27 and 28 October. One more measured green run to go.

**Measured run 2 (run 1691, commit d3579afba): green.** | Job | Time | |---|---| | Browser tests (two workers) | 24m 0s | | Unit tests and coverage (two workers) | 14m 28s | | Database tests on PostgreSQL | 6m 35s | | Checks | 1m 58s | | Security audit | 2m 0s | For comparison, the browser suite took 33 to 42 minutes in one process on 27 and 28 October. One more measured green run to go.

tiagoagueda referenced this pull request from a commit 2026-10-30 13:50:45 +00:00

Run the unit tests on three workers, each letting go of the languages it met (#1033) (#1085)

tiagoagueda force-pushed 1033-ci-workers from d3579afbaa to 4e5d547e13 2026-10-30 20:53:26 +00:00 Compare

tiagoagueda commented 2026-10-30 21:34:37 +00:00

Author

Owner

• Copy link
• Copy Markdown

Measured run 3 (commit 4e5d547e1, after #1085 put the unit job on three workers): red. The browser job on two workers failed four layout and timing tests: test_target_size, test_site_search scripts-off panel, test_board_fold long column at 640, and test_every_page phone dark. All four passed when run again alone. The job took 32 minutes against 24 on run 2. There were no database errors; this is CPU contention, with three unit workers, two browser workers and the PostgreSQL job on one six-core host.

Item 2 fails its measurement. The browser step goes back to one process. Items 6 and 7 stay: the translation cache, the blobless checkout, the Chromium libraries, the heavy step by path, and no map tiles in tests. This PR then stops being a draft.

**Measured run 3 (commit 4e5d547e1, after #1085 put the unit job on three workers): red.** The browser job on two workers failed four layout and timing tests: `test_target_size`, `test_site_search` scripts-off panel, `test_board_fold` long column at 640, and `test_every_page` phone dark. All four passed when run again alone. The job took 32 minutes against 24 on run 2. There were no database errors; this is CPU contention, with three unit workers, two browser workers and the PostgreSQL job on one six-core host. **Item 2 fails its measurement.** The browser step goes back to one process. Items 6 and 7 stay: the translation cache, the blobless checkout, the Chromium libraries, the heavy step by path, and no map tiles in tests. This PR then stops being a draft.

tiagoagueda changed title from WIP: Run the browser suite on two workers and the unit suite on three, and trim CI's setup (#1033) to Cache what every CI job rebuilt, and fetch no map tiles in tests (#1033) 2026-10-30 21:42:02 +00:00

tiagoagueda force-pushed 1033-ci-workers from 4e5d547e13 to 7f064bfd47 2026-10-30 21:42:03 +00:00 Compare

tiagoagueda merged commit 2f335b13bc into main 2026-10-31 02:10:03 +00:00

tiagoagueda deleted branch 1033-ci-workers 2026-10-31 02:10:03 +00:00

tiagoagueda referenced this pull request from a commit 2026-10-31 02:10:03 +00:00

Cache what every CI job rebuilt, and fetch no map tiles in tests (#1033) (#1083)

tiagoagueda referenced this pull request 2026-10-31 03:04:41 +00:00

Efficiency and reliability audit of 2026-10-30: index #1108

tiagoagueda referenced this pull request 2026-10-31 15:55:19 +00:00

CI: make a push report sooner and cost less, without new hardware #1033

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

accessibility

Usable by everyone, including people with disabilities: one of the project's stated missions

  

authentication

Signing in, and proving who you are

  

breaking change

Changes behaviour or data in a way operators must act on

  

bug

Something does not do what it says it does

  

documentation

The wiki, the docs, or the words in the interface

  

enhancement

An improvement or a new capability

  

interface

The shape of the pages: what is where, and how it reads

  

internationalisation

Languages, regions, and the formats and symbols that go with them

  

observability

What an operator can see about a running instance: metrics, logs, health

  

performance

Uses fewer resources, or is more reliable, with no loss of features or accessibility

  

security

Keeping the code and the data secure: one of the project's stated missions

  

tier

1

Easy: small, self-contained, no prerequisites

  

tier

2

High-severity security: account access, personal data, the boundaries

  

tier

3

Foundations: prerequisite work that unblocks other issues

  

tier/4

Later: large features and long-horizon work

No labels

accessibility

authentication

breaking change

bug

documentation

enhancement

interface

internationalisation

observability

performance

security

tier

1

tier

2

tier

3

tier/4

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

Postulo/postulo!1083

No description provided.