test(gateway): de-flake concurrent-compression lock test with a barrier

[teknium1]

Summary

tests/gateway/test_compression_concurrent_sessions.py::test_concurrent_compressions_same_session_serialize no longer flakes under CI load. It was failing on shard test (1) with "Expected exactly one agent to compress, got 2" — including on main HEAD, independent of any feature PR.

Root cause (test-only, not a lock bug): the test relied on a time.sleep(0.25) inside the stubbed compressor to make the two threads overlap inside the per-session lock window. Under CI CPU starvation that sleep is insufficient — one thread can acquire → compress → rotate → release the lock before the other thread even reaches try_acquire. Both then acquire on the shared session_id and both compress. The real per-session lock (#34351) is correct; the test's overlap was a timing accident.

Changes

• tests/gateway/test_compression_concurrent_sessions.py: wrap the shared db's try_acquire_compression_lock in a threading.Barrier(2) so both threads rendezvous immediately before the real atomic acquire — guaranteeing genuine simultaneous contention regardless of scheduling. The real lock logic is untouched; it still picks exactly one winner. The wrapper is restored after join() so the post-join lock-leak assertion hits the unwrapped method.

Validation

	Old (sleep-overlap)	New (barrier)
20× plain runs	flaked in CI	20/20 pass
15× under all-core CPU stress (load avg ~4.6)	reproduced the race	15/15 pass
sibling test (5-session aliasing)	pass	pass
canonical isolated runner	2/2	2/2

CPU-stress reproduction is the key check: the old version's failure mode (one thread finishing before the other starts) is now structurally impossible.

Infographic

[github-actions]

🔎 Lint report: fix/compression-concurrency-test-flake vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 10394 on HEAD, 10392 on base (🆕 +2)

🆕 New issues (2):

Rule	Count
invalid-assignment	2

First entries

tests/gateway/test_compression_concurrent_sessions.py:192: [invalid-assignment] invalid-assignment: Object of type `bound method SessionDB.try_acquire_compression_lock(session_id: str, holder: str, ttl_seconds: int | float = ...) -> bool` is not assignable to attribute `try_acquire_compression_lock` of type `def try_acquire_compression_lock(self, session_id: str, holder: str, ttl_seconds: int | float = ...) -> bool`
tests/gateway/test_compression_concurrent_sessions.py:171: [invalid-assignment] invalid-assignment: Object of type `def _barriered_acquire(...) -> Unknown` is not assignable to attribute `try_acquire_compression_lock` of type `def try_acquire_compression_lock(self, session_id: str, holder: str, ttl_seconds: int | float = ...) -> bool`

✅ Fixed issues: none

Unchanged: 5425 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.