[Kayvan-Zahiri]

Problem and change

Django's Postmark backend leaves an automatically opened client attached after send_batch raises or returns a failed message. The close call is only reached after every batch succeeds.

Put the batch loop inside try/finally so clients opened by send_messages are closed on failure too. Keep this cleanup inside the existing exception-handling boundary so fail_silently, error signals, partial success counts, and caller-owned connection reuse remain intact. Add an Unreleased changelog entry.

Fixes #38.

Validation

• Before: 4 failed, 4 passed new lifecycle cases. Cover transport exceptions and per-message failures, both fail_silently values, and automatically opened versus caller-owned clients.
• After: full pytest --cov=postmark --cov-report=term --cov-fail-under=85 — 462 passed, 96.32% coverage.
• Full Ruff lint and format checks pass; MyPy passes across 64 source files.
• Required pre-commit run --all-files — all applicable hooks pass, including Poetry metadata/lock consistency, pinned Ruff/MyPy, codespell, Bandit, and secret scanning.
• git diff --check passes. No unrelated file changes.
• Python 3.12 / Django 5.2; full Python/Django matrix not run. The existing fork-in-multithreaded-process test emits a deprecation warning.
• Synthetic messages and mocked clients only. No hosted Postmark requests, actual email delivery, credentials, paid resources, or UI automation.

AI assistance: OpenAI Codex investigated, implemented, tested, and prepared this draft. Human verification is not claimed.

[gregsvo]

Excellent. Good catch. I'm pulling/testing now. Will merge if no issue.

[gregsvo]

Excellent work here, @Kayvan-Zahiri. I checked it out all tests pass. lint and mypy are clean, all good.

I did find one small edge case: if client.close() itself raises while we're cleaning up after a failed send, the close error replaces the original send error, both for the caller and for the on_exception signal. Obviously I'd like the original error to win. I want to get this out fast, and since you've allowed maintainer edits, I'm pushing one small follow-up commit to your branch (guard, and 2 tests). Your commit stays exactly as it is and remains the one that fixes the issue.

Could you also mark the PR Ready for review? Our CI needs a maintainer to approve the first run on a fork, and I'll do that as soon as it's ready. This should go out in the next release, and I'll credit you in the release notes. Thanks again!

[Kayvan-Zahiri]

Marked ready. Thanks for testing it, and good call on letting the original send error win over a close() failure.

[gregsvo]

Thanks again! Merged.