Entity-lifecycle stress tests; fix a discovery/teardown UAF

[sqt]

Motivation

zzdds has no concurrency- or lifecycle-under-load tests — the roadmap and DCPS coverage audit both call for an OpenDDS-EntityLifecycleStress-shaped stress tier. Standing up the first scenario immediately exposed a real use-after-free: rapid concurrent create/delete of DataWriters and Publishers on one participant crashes the SEDP receive thread mid-on_publication_matched dispatch.

Description of changes

• New stress-tests/ (mirrors examples/ layout). The README carries a survey of OpenDDS's EntityLifecycleStress and the port design. Python run_all.py orchestrator + two Zig apps under zig/:
  • entity_lifecycle_stress — faithful multi-process port: N pub + M sub processes, interleaved, on a per-run-unique domain, keyed Messenger type, teardown-duration monitor, deterministic --cleanup {explicit,cascade}, structured SUMMARY: line, optional --churn.
  • lifecycle_churn — in-process N-thread churn under DebugAllocator, --scenario {entities,reentrant} (frame ready for waitset/listener/instance/cft).
• New ci.yml stress job (needs: test-linux): builds native + -Ddebug-allocator, runs run_all.py --strict, plus a ThreadSanitizer reentrant lane and a heavier schedule-only matrix.
• UAF fix (src/dcps/{participant,writer,reader,publisher,subscriber}.zig), two parts:
  a. onReaderDiscovered / onWriterDiscovered / the two announceMatched* paths released participant.mu before firing matched_notify while pinning only the RTPS proto, not the target DataWriterImpl/DataReaderImpl. MatchedNotify now carries is_reader + quiesce hooks, acquired while mu is held and held across the notify.
  b. PublisherImpl / SubscriberImpl gained an EntityQuiesce; each child DataWriter/DataReader holds a lifetime reference on its parent, so the DDS §2.2.4.1.5 listener-fallback chain (writer → publisher → participant) can never read a freed enclosing entity. Owned-child teardown moved from reallyDeinit into deinit to break the resulting ref cycle; a parent_pinned flag keeps hand-built test impls (which skip init()) balanced.
• New test/dcps/lifecycle_churn_test.zig — regression (UDP loopback, stable reader + writer-churn thread), wired into dcps_test_files.

[greptile-apps]

Greptile Summary

The PR adds lifecycle stress coverage and strengthens entity ownership during concurrent discovery callbacks and teardown.

• Pins mlugg/setup-zig to an immutable commit across CI and release workflows, resolving the prior supply-chain finding.
• Holds quiescence references across match notifications and pins writers/readers to their enclosing publisher/subscriber.
• Adds multi-process and in-process lifecycle stress applications, orchestration, DebugAllocator coverage, and a TSan lane.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported mutable third-party action reference is now pinned to a full commit SHA throughout the changed workflows.

Important Files Changed

Filename	Overview
.github/workflows/ci.yml	Adds the gated stress and sanitizer lanes and replaces mutable third-party Zig setup tags with an immutable commit.
.github/workflows/release.yml	Pins the third-party Zig setup action to the same immutable commit used by CI.
src/dcps/participant.zig	Extends matched-notification handling so endpoint quiescence remains pinned while callbacks run outside the participant lock.
src/dcps/publisher.zig	Adds publisher quiescence and moves owned-writer teardown earlier to support child-to-parent lifetime references.
src/dcps/subscriber.zig	Adds the corresponding subscriber quiescence and owned-reader teardown ordering.
src/dcps/writer.zig	Pins initialized writers to their parent publisher and releases that reference during final destruction.
src/dcps/reader.zig	Pins initialized readers to their parent subscriber and releases that reference during final destruction.
stress-tests/run_all.py	Orchestrates the new lifecycle stress scenarios with strict result handling.
stress-tests/zig/entity_lifecycle_stress/main.zig	Implements multi-process publisher/subscriber lifecycle stress with explicit and cascading cleanup modes.
stress-tests/zig/lifecycle_churn/main.zig	Implements threaded in-process entity churn and reentrant lifecycle scenarios.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    D[Discovery callback] --> L[Lock participant state]
    L --> Q[Acquire endpoint quiesce reference]
    Q --> U[Release participant lock]
    U --> N[Dispatch matched notification]
    N --> R[Release quiesce reference]
    T[Writer or reader teardown] --> B[Block new quiesce entries]
    B --> W[Wait for active references]
    W --> F[Destroy endpoint and release parent reference]

Loading

_{Reviews (4): Last reviewed commit: "resolving CI issues" | Re-trigger Greptile}

[codecov]

Codecov Report

❌ Patch coverage is 91.66667% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/dcps/participant.zig	82.35%	6 Missing ⚠️

📢 Thoughts on this report? Let us know!