Restore 55 Pulsar tests and skip only the four behaviours that are missing (GH-3763)

[jeremydmiller]

The Pulsar pass of #3763, after #3791 (AWS), #3794 (Kafka) and #3795 (RabbitMQ).

An exclusion with no recorded reason, and far too broad

Commit 02f531842 on 2026-06-08 — message "Disable Pulsar compliance tests", no other rationale — tagged all three Pulsar compliance classes Category=Flaky. That is 66 of the project's 224 tests; CIPulsar has been running 158.

Untagged: 213 pass, 11 fail — and the 11 are the same four behaviours repeated across the three fixtures.

Test	Failure
will_requeue_and_increment_attempts	Expected ending activity was not detected
can_schedule_retry	Expected ending activity was not detected
will_move_to_dead_letter_queue_without_any_exception_match	No ending activity detected
will_move_to_dead_letter_queue_with_exception_match	No ending activity detected

They are not flaky — they fail deterministically, every run, alone or in a suite, and they look like genuinely unimplemented transport behaviour: native requeue with attempt tracking, scheduled retry, and dead-letter routing. Filed as #3797.

So: all three classes come off the ledger, only those four are skipped

Each skip carries the reason and the issue number, so what's missing is greppable instead of buried under three blanket exclusions. can_schedule_retry had to become virtual in TransportCompliance to be overridable; the other three already were.

with_cloud_events already opted out of will_move_to_dead_letter_queue_with_exception_match by overriding it with a bare return Task.CompletedTask — which reports as a PASS, so the suite was counting a test that ran nothing. That's now a real Skip carrying its original explanation.

Net: CIPulsar goes from 158 executing tests to 213, plus 11 named skips.

Two things worth knowing that came out of this

• CIPulsar runs Pulsar in Testcontainers, not docker-compose (Global partitioning: dedicated end-to-end tests for Azure Service Bus, GCP Pub/Sub, NATS, Redis, and Pulsar #3467). A local run hangs silently at 224 batched if Docker is short on memory — nine stale containers were holding 4.7 of 7.7 GiB here and no Pulsar container could start. Stopping them fixed it; nothing in the output said so.
• The supervisor's summary counts SKIPPED tests inside its "passed" number. Run 2 below reports "224 passed" for 213 executed + 11 skipped. The per-assembly runner reports them correctly (18 succeeded / 4 skipped for one fixture). Worth a look given CI retry counts are invisible: a job burning 22 of 25 retries is indistinguishable from a clean green #3787's honest-reporting goal — not changed here.

Verification

	Result
CIPulsar, run 1	223 passed, 1 retry, exit 0
CIPulsar, run 2	224 passed, 0 retries, exit 0

Repo flaky tags 16 → 13. wolverine.slnx builds clean in Release.

🤖 Generated with Claude Code

https://claude.ai/code/session_0116vfBcKwcjWn8msM4ZjkuA

[jeremydmiller]

Follow-ups filed out of this work, so the PR itself stays scoped to the tag burn-down:

• Pulsar: requeue, scheduled retry and dead-letter compliance behaviours are unimplemented, and three whole compliance classes were excluded for them #3797 — the four compliance behaviours that are actually missing (will_requeue_and_increment_attempts, can_schedule_retry, both will_move_to_dead_letter_queue_*). These are the 11 skips this PR adds; the issue is what un-skips them.
• Pulsar tests start one heavy broker container per worker process, on an unpinned :latest, and hang silently when Docker runs out of memory #3799 — the test containers: one heavy Pulsar broker per worker process, an unpinned apachepulsar/pulsar:latest that disagrees with the 4.0.3 pin in docker-compose.yml, and the silent hang at "224 batched" when Docker is out of memory. Includes a cheap fix: the compose pulsar service already exists and is unused, and the fixture already honours WOLVERINE_PULSAR.
• CloudEvents compliance cannot test dead-lettering by exception type: ErrorCausingMessage does not round-trip through System.Text.Json #3800 — will_move_to_dead_letter_queue_with_exception_match under CloudEvents, split out because it is a shared-harness gap rather than a Pulsar defect: ErrorCausingMessage carries a Dictionary<int, Exception> that does not round-trip through System.Text.Json.