[finger563]

Summary

Extends the DDS interop matrix (components/rtps/interop) with the defect classes the recent CycloneDDS work turned up. Every ROS 2 pair in the matrix proved a single sample within a few seconds, which is why none of them caught a peer that is discovered and then silently lost.

case	guards	FastDDS	CycloneDDS
ros2_pub->espp_sub_liveliness	samples keep arriving 22 s in, past a remote participant lease (CycloneDDS: 10 s, re-announced with readerId ENTITYID_UNKNOWN). Only samples after 22 s count - discovery, the lease and the liveliness check cadence all fit before that, so a peer whose SPDP re-announcements were not delivered - lease expired, every matched endpoint dropped - fails. The #871 class.	control	yes
espp_pub->ros2_sub_liveliness	the reverse leg, counted on the ROS 2 side by a small rclpy subscriber (ros2 topic echo --once cannot count)	control	yes
many_participants_late_joiner	twelve unrelated ROS 2 participants are on the domain before espp starts, then the publisher espp needs joins last and must still be matched. (CycloneDDS allows ten participants per host per domain by default; the CycloneDDS pass raises MaxAutoParticipantIndex through CYCLONEDDS_URI.) A participant beyond the remote-participant pool (SPDP_MAX_NUMBER_FOUND_PARTICIPANTS) is ignored silently - the #870 class. The harness builds the host profile (32 slots), so this passes; the embedded profile's 10 slots need #870's knob for such a domain, which the case note and README say.	yes	yes

• rtps_interop_sub gains a backward-compatible ninth argument, count_after_s: samples that arrive earlier are reported but not counted. ros2_count_subscriber.py is the ROS 2-side counterpart.
• The cases are shell functions called from both the FastDDS and the CycloneDDS pass (same convention as the rest: timeout on every process, background publishers killed and waited after the subscriber exits, log tails on the way out). They add about three minutes; the workflow timeout goes from 30 to 40 minutes.

Observed on CI (run 41378439624, 53 pass / 1 fail): the one failure is cyclone: espp_pub->ros2_sub_liveliness (FAIL received=0 required=10 (not counted, before 22s: 75)): espp stops sending to the CycloneDDS peer once its 10 s lease expires, the #871 signature; it fails until #871 merges. The other CycloneDDS leg, cyclone: ros2_pub->espp_sub_liveliness, passes on main: a HEARTBEAT that reaches a matched espp reader also refreshes the sending participant's liveliness (MessageReceiver::processHeartbeat), so a CycloneDDS writer keeps its participant alive through its heartbeats even with its SPDP re-announcements lost, while an espp writer only gets ACKNACKs back, which do not refresh it. Both FastDDS controls and both many_participants_late_joiner cases pass. (The README records the asymmetry.)

Left out: a live multi-homed peer (the #865 / #866 / #867 class, whose selection logic locator_subnet already checks on canned announcements). It needs a peer address espp cannot reach, which one shared network namespace cannot provide - every local address is reachable there. The README describes the design (a second Docker network, the ROS 2 side on both, espp on one, i.e. a two-container run.sh) as follow-up work rather than shipping a half-test.

Test plan

• Host build of the interop binaries; count_after_s verified both ways on a local cross-process pair (counts only after the window opens; fails with the early samples reported when the publisher stops before it)
• CI interop matrix, run 41378439624: 53 pass / 1 fail, the one expected failure above
• CI interop matrix after the review follow-ups (bf0b16e)

🤖 Generated with Claude Code

https://claude.ai/code/session_01DFjjnq3XCTRAXENSxsCxJU

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

8 open issues

• rospub is the PID of the timeout wrapper, so kill $rospub may not terminate the underlying… · New
• Same cleanup issue as the liveliness case: the stored PIDs are for timeout, not necessarily the… · New
• If an exception occurs (e.g., ROS 2 initialization/spin errors), the node may not be destroyed and… · New
• If an exception occurs (e.g., ROS 2 initialization/spin errors), the node may not be destroyed and… · New
• If an exception occurs (e.g., ROS 2 initialization/spin errors), the node may not be destroyed and… · New
• count_after_s is a new user-facing argument but is not validated. Negative values effectively… · New
• count_after_s is a new user-facing argument but is not validated. Negative values effectively… · New
• This prints + flushes once per early sample. In the liveliness cases, this can be tens/hundreds of… · New

What changed in this PR

Extends the RTPS/DDS interop matrix to cover liveliness past participant lease expiry and high-participant-count discovery edge cases (notably those found during recent CycloneDDS addressing work).

Changes:

• Add count_after_s gating to rtps_interop_sub so early samples are reported but not counted (used for liveliness/lease tests).
• Introduce new interop harness cases for lease-liveliness and “many participants + late joiner”, and add a ROS 2 counting subscriber helper.
• Increase CI workflow timeout to accommodate the expanded matrix runtime.

File	Description
pc/​tests/​rtps_interop_sub.cpp	Adds count_after_s argument and gating logic for counted samples.
components/​rtps/​interop/​run_interop.sh	Adds new liveliness/lease and many-participants cases to the interop run matrix.
components/​rtps/​interop/​ros2_count_subscriber.py	New ROS 2 helper to count messages after a time gate for liveliness validation.
components/​rtps/​interop/​README.md	Documents the new interop cases and notes a not-yet-covered multi-homed scenario.
.github/​workflows/​rtps_interop.yml	Increases workflow timeout from 30 to 40 minutes.

_{Review effort: Lite}

[github-actions]

✅Static analysis result - no issues found! ✅

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

13 open issues

• Same cleanup issue as the liveliness case: the stored PIDs are for timeout, not necessarily the…
• rospub is the PID of the timeout wrapper, so kill $rospub may not terminate the underlying…
• wait can return a non-zero status (e.g., if a PID has already exited/reaped), which can… · New
• count_after_s accepts negative values (via atoi), which contradicts the documented semantics… · New
• count_after_s accepts negative values (via atoi), which contradicts the documented semantics… · New
• With count_after_s > 0, early samples skip the large-payload byte-exact verification entirely due… · New
• This prints + flushes once per early sample. In the liveliness cases, this can be tens/hundreds of…
• count_after_s is a new user-facing argument but is not validated. Negative values effectively…
• count_after_s is a new user-facing argument but is not validated. Negative values effectively…
• If an exception occurs (e.g., ROS 2 initialization/spin errors), the node may not be destroyed and…
• If an exception occurs (e.g., ROS 2 initialization/spin errors), the node may not be destroyed and…
• If an exception occurs (e.g., ROS 2 initialization/spin errors), the node may not be destroyed and…
• Invalid CLI values (e.g., non-numeric SECONDS / COUNT_AFTER_S) will currently raise a Python… · New

_{Review effort: Lite}