Repository navigation
chore: bump fastapi from 0.163.0 to 0.163.1 - #451
Conversation
Brief: the third Dependabot
|
| PR | Change | Release notes, against Watcher |
|---|---|---|
| #453 | sqlalchemy 2.1.8 → 2.1.9 (floor >=2.1.9) |
ORM bug fixes only: post_update deletes, and_() criteria in selectinload/lazyload against aliased subqueries, and loader options under selectin_polymorphic and subqueryload on aliased entities. Watcher's models use none of those. Alembic is its only other dependent in the lock. |
| #452 | python-ulid 4.1.0 → 4.2.0 | Adds ulid.sqlalchemy.ULIDType behind a new extra. The fixes: the ULID(...) constructor now raises TypeError for a 16-character string or an int; ULID.parse reads integers differently; from_datetime/from_int raise ValueError rather than OverflowError; Pydantic validation errors change. Watcher calls only ULID() and ULID.from_str [traced: src/core/models/base.py:27,32; src/core/ulids.py; the str(ULID()) calls in fetch_commands.py and process_commands.py]. Its own src.core.models.base.ULIDType is unrelated to the new ulid.sqlalchemy.ULIDType, since nothing installs the extra. tests/core/test_ulids.py holds the canonical-uppercase contract and is green on the head. |
| #451 | fastapi 0.163.0 → 0.163.1 | One fix, OpenAPI for a response model with a nested computed field. Watcher serves only /health and /ready, with the docs endpoints off. |
No migrations: none of the three is procrastinate or ships schema, so there's no pre step this time (cf. #443).
Hazards (weigh them; overrule with evidence)
-
One PR at a time, by the recipe. For each one:
- the co-core grep on
gh pr diff <N>, which must print nothing; gh pr checks <N> --watch --required;gh pr merge <N> --squash;- on the host,
git pull --ff-only, thenbash scripts/co-pypi.sh uv sync --locked; - then the tree check against
pull/<N>/head.
After each merge, the next PR is behind. Request a rebase by commenting
dependabot rebasewith the@prefix. Never use GitHub's Update branch on these: it adds a non-Dependabot commit, and Dependabot stops rebasing. Re-post after about 6 minutes of silence. Suggested order: chore: update sqlalchemy[asyncio] requirement from <3,>=2.1.8 to >=2.1.9,<3 #453, chore: bump python-ulid from 4.1.0 to 4.2.0 #452, chore: bump fastapi from 0.163.0 to 0.163.1 #451. - the co-core grep on
-
Merge and sync in the same sitting, and restart once after the last. All three are runtime packages, so the batch needs one
sudo systemctl restart watcher, with the operator's OK at the time.- Monday's patch reboot, 2026-11-23 01:00Z, deploys
mainas checked out. The unit'suv sync --locked --offlinestart check refuses a locked wheel that's in neither the venv nor uv's cache. So a merge left unsynced breaks that boot. - Land well before Monday's window, and keep clear of the quiet ranges, 03:15–03:45Z and 05:15–06:00Z.
- Monday's patch reboot, 2026-11-23 01:00Z, deploys
-
The squash title is Dependabot's PR title (
chore: bump …), which already fits the repo's form. No edit is needed. -
Rollback for any one PR: a revert PR, squash-merged after green, then pull, sync and restart. There's no schema to undo.
Read-back
/healthreports the finalmainSHA, andNRestarts=0.- The venv reads sqlalchemy 2.1.9, python-ulid 4.2.0 and fastapi 0.163.1, with co-core still 0.48.2 and procrastinate 3.11.0.
- The embedded worker runs its periodic jobs (
drain_pending_*,publish_pending_fetch_commands), each endingSuccess. That covers the ULID load path (ULIDType→ULID.from_str) on real rows. - The four consumers are polling, and nothing is at WARNING or higher since the restart.
- Each PR's tree check matches, and the push runs on
mainare green by job.
Done when
All three are merged as squash PRs with CI green by job, synced, live after one restart, and the read-back is posted here. Each PR shows as merged.
Out of scope
Later Dependabot PRs; redis (#441's ignore); any sibling repo.
Brief re-check before landing (2026-11-20)Re-ran against Held Corrections:
|
|
@dependabot rebase |
662e29d to
afdd106
Compare
Bumps [fastapi](https://github.com/fastapi/fastapi) from 0.163.0 to 0.163.1. - [Release notes](https://github.com/fastapi/fastapi/releases) - [Commits](fastapi/fastapi@0.163.0...0.163.1) --- updated-dependencies: - dependency-name: fastapi dependency-version: 0.163.1 dependency-type: direct:production update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com>
afdd106 to
2cd5e76
Compare
Read-back: third Dependabot
|
| PR | Squash on main |
Head merged | Tree check (pull/<N>/head vs HEAD) |
main push CI, by job |
|---|---|---|---|---|
| #453 sqlalchemy 2.1.9 | d674ed77 |
1b648914 |
044712cc… = |
lint, test, integration, migrations ✅ (run 44784505147) |
| #452 python-ulid 4.2.0 | 54c8a787 |
dd4f3054 (rebased) |
efb5452b… = |
all four ✅ (44784892863) |
| #451 fastapi 0.163.1 | a61cf8dc |
2cd5e764 (rebased) |
18703a40… = |
all four ✅ (44785297625) |
Before each merge, the co-core grep on gh pr diff printed nothing, and the four required checks were green on the head being merged. Each landing was followed by co-pypi.sh uv sync --locked (each swapped exactly its one package) and scripts/pre-ship.sh (green). The audit push runs are green with no annotations.
Live [measured]:
/health→{"status":"ok","build":"a61cf8dc"},/ready200.NRestarts=0, active.- Venv: sqlalchemy 2.1.9, python-ulid 4.2.0, fastapi 0.163.1. co-core and co-core-aio are still 0.48.2, procrastinate still 3.11.0.
- The start check printed
Checked 79 packages, then: broker reachable, the four consumers started (info.registry,info.watch-check,content.blobs,content.derived), andinfo.registryreplay complete → tailing (1003 announcements processed). - In the first ~3½ min, every periodic job that ran ended
Success:- ×3 each:
drain_pending_watch_events,drain_pending_archiver_sync,publish_pending_fetch_commands,publish_pending_process_commands,schedule_tick; - ×1 each:
publish_watch_status,reap_fetch_commands,reap_process_commands,recover_stalled_jobs.
- ×3 each:
- The ULID load path ran on real rows:
publish_watch_statuspublished 22, i.e. everyWatchedItem, whoseid/archiver_info_item_idareULIDType→ULID.from_str. - The journal since the restart has 2122 JSON records, all
INFO, and nothing atWARNINGor higher.
Known limits:
- The drains had nothing to drain.
drain_pending_*andpublish_pending_*each published 0, so their own row loads, andULID.from_intin_mint_event_id(which only runs on a sub-millisecond collision), weren't exercised live. Their ULID evidence is the suite, pluspublish_watch_statusabove. - "Consumers polling" rests on their start lines and the absence of errors. Only
info.registrylogs per message at INFO. The other three consumers saw no traffic in the window, and they don't log idle polls. - Recipe, didn't hold as written: a head change isn't proof a rebase request was answered. chore: update sqlalchemy[asyncio] requirement from <3,>=2.1.8 to >=2.1.9,<3 #453's merge made Dependabot rebase chore: bump fastapi from 0.163.0 to 0.163.1 #451 on its own (
afdd106c, ond674ed77), unprompted. After chore: bump python-ulid from 4.1.0 to 4.2.0 #452 merged, that put chore: bump fastapi from 0.163.0 to 0.163.1 #451's head ahead of my request while the PR was stillBEHIND. The answering push (2cd5e764, on54c8a787) came about 80s after the comment. Check the new head's parent (git merge-base --is-ancestor <main> FETCH_HEAD), ormergeStateStatus=CLEAN, beforegh pr checks --watch. - Recipe, held: both
@dependabot rebaserequests were answered on the first try (about 90s and 80s), with no re-posts. Strict checks blocked nothing, and Update branch was never used. - Recipe, gap:
gh pr checks --watch --requiredrun right after a force-push can report before the new head's checks register. I waited 15s first. With the four required checks the ruleset enforces,gh pr mergewould refuse anyway, so this is cosmetic.
Epic #394 is untouched. Monday's 2026-11-23 01:00Z patch reboot deploys a61cf8dc with a matching venv.
…h, reply rule - Ancestry check fetches the PR ref first in one call (measured on #451: 1, #455: 0, bad ref: 128) and names its exit codes. - mergeStateStatus: only CLEAN reads as on main; BLOCKED (#455, up to date with checks pending) defers to the ancestry form. - The cause sentence names the uv.lock conflict that triggers the auto-rebase. - --watch also waits for the new head's runs to be listed. - Re-post rule: a Dependabot reply is an answer, not silence. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Bumps fastapi from 0.163.0 to 0.163.1.
Release notes
Sourced from fastapi's releases.
Commits
a3260d4🔖 Release version 0.163.1 (#16775)b111374📝 Update release notesa1a05b7🐛 Fix OpenAPI for response model with nested computed field (#14490)