Skip to content

release: 3.13.156 — Kafka offset-reset (#263) + raw producer (#264) - #269

Merged
andrevanzuydam merged 5 commits into
v3from
feature/release3.13.156
Nov 1, 2026
Merged

andrevanzuydam merged 5 commits into
v3from
feature/release3.13.156

Conversation

@andrevanzuydam

Copy link
Copy Markdown
Collaborator

Release 3.13.156 — Kafka queue parity fixes.

Fixes

  • Kafka Offset Hardcoded #263 Kafka consumer auto.offset.reset was hardcoded to earliest; now reads TINA4_KAFKA_AUTO_OFFSET_RESET (then bare KAFKA_AUTO_OFFSET_RESET), default earliest (unchanged behaviour).
  • Feature Request: No queue wrapper #264 push(..., raw=true) / produce(..., raw=true) emit the value with no Tina4 envelope and no injected id, for a non-Tina4 consumer. Kafka-only; file/rabbitmq/mongodb refuse raw=true by name.

Ported at parity across all four frameworks with equivalent tests (pure offset resolution + offline raw-refusal that run everywhere; a live Kafka round-trip proving no envelope/id on the wire, gated on a broker). Version bumped to 3.13.156 + CHANGELOG.

🤖 Generated with Claude Code

andrevanzuydam and others added 5 commits November 1, 2026 17:43
…264); release 3.13.156

#263: the confluent consumer hardcoded auto.offset.reset=earliest, so a new
consumer group replayed the whole retained backlog on first start. It now reads
TINA4_KAFKA_AUTO_OFFSET_RESET (then bare KAFKA_AUTO_OFFSET_RESET), default
earliest (unchanged behaviour).

#264: Queue.push(data, raw=True) / produce(..., raw=True) produce the value with
no Tina4 envelope and no injected id, for a non-Tina4 consumer. Kafka-only; the
file/rabbitmq/mongodb backends refuse raw=True by name. Default raw=False path
unchanged; pop() already unwraps both shapes.

Tests: pure offset-reset resolution + offline raw-refusal (run everywhere) and
live Kafka round-trips proving a raw produce carries no envelope/id (gated on a
real broker). Version bump to 3.13.156 + CHANGELOG.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Andre van Zuydam <andrevanzuydam@gmail.com>
…igrations in status/startup (#268)

The migrate:create next steps printed `tina4 migrate --rollback` and
`tina4 migrate --status` — flags `migrate` refuses — so a person or model
following them errored on the first command. They now name the real commands,
`tina4 migrate:rollback` and `tina4 migrate:status`.

Also closed the two consistency gaps the report flagged: `migrate` applies both
.sql AND .py migrations, but `migrate:status` listed only .sql as pending and
startup auto-migration only triggered on a .sql file, so a pending .py migration
was invisible and never auto-applied at boot. migrate(), get_files(), status()
and the startup gate now share one _migration_source_files() helper that globs
.sql + .py (excluding .down.sql and dunder .py).

Python-only: the wrong hint is python's and .py migrations are a python feature;
the other three frameworks carry neither. Tests use a real SQLite db (no mocks):
the hint names the existing commands, a pending .py migration shows in status()
and get_files(), and startup auto-migrate runs for a .py-only folder.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Andre van Zuydam <andrevanzuydam@gmail.com>
… changes

- core/server.py: the #268 startup-migrate gate now reuses the runner's
  _migration_source_files() helper instead of an inline generator, so it adds no
  cyclomatic complexity (server.py offenders stay at the baseline 17).
- queue_backends/kafka_backend.py: trimmed the _offset_reset docstring.
- .tina4-metrics.json: surgically bump ONLY the kafka_backend.py entry
  (loc 499->523, offenders 3->4 — the file was at the 500-LOC edge and the #263
  offset helper + #264 raw branch are a deliberate, documented feature). Every
  other file's baseline entry is left untouched (no 9-day-stale wholesale regen).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Andre van Zuydam <andrevanzuydam@gmail.com>
…it can't wedge the suite

This real-PostgreSQL two-thread test can leave a DB call blocked inside libpq (a C
call). The suite's default --timeout-method=signal delivers SIGALRM to the main
thread only and cannot interrupt a blocking C call, so such a block does not time
out — it silently wedged the whole pytest-split group for ~7 min (no output, no
traceback) until CI killed the step. The 3.13.156 branch's new test files shifted
pytest-split's even-by-count grouping onto this test, exposing the hazard.

A per-test @pytest.mark.timeout(120, method="thread") overrides the CLI method for
this test only: a watchdog thread dumps every thread's stack and fails fast
instead of hanging. It changes nothing the test asserts. Fix-on-discovery of a
suite-wedge hazard found while releasing 3.13.156.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Andre van Zuydam <andrevanzuydam@gmail.com>
…ommands (#268)

test_migration_create_returns_generate_v1_1_envelope asserted the next[] steps
contained "--rollback"/"--status" — the exact broken flags #268 fixed (migrate
refuses them). Updated it to assert the commands that exist, `tina4 migrate:rollback`
and `tina4 migrate:status`, plus a guard that the broken `migrate --rollback` /
`migrate --status` forms never return. Parity-safe: the cross-framework envelope
parity test does not pin this text, and the hint is python-only.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Andre van Zuydam <andrevanzuydam@gmail.com>
@andrevanzuydam
andrevanzuydam merged commit 8fbc225 into v3 Nov 1, 2026
13 of 14 checks passed
@andrevanzuydam
andrevanzuydam deleted the feature/release3.13.156 branch November 1, 2026 19:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant