Per-command timeouts for ROS graph collection, tests, and integration script hardening

[RN0H]

Motivation

• Prevent long-running or stuck ros2 subprocesses from blocking overall graph collection by applying a per-command timeout.
• Surface and propagate failures from topic and service list commands instead of silently continuing on partial data.
• Make the integration test script more robust, debuggable, and tolerant of transient timing issues while removing the CI integration job from the sanity workflow.

Description

• Added a graphCommandTimeout variable and a runGraphCommand helper that runs each ros2 invocation with a per-command timeout via WithTimeout.
• Updated CollectSystemGraph to call runGraphCommand for node/topic/service commands and to return errors when topic list or service list fail.
• Added unit tests graph_errors_test.go and graph_timeout_test.go that validate error propagation and that per-command timeouts are used by CollectSystemGraph using mapGraphRunner and delayedGraphRunner test runners.
• Hardened tests/integration.sh with: dynamic container naming, stricter cleanup, debug logging on failure, explicit waits for nodes and topics, retries and JSON validation for gim graph, and retries for gim analyze output checks.
• Removed the integration job from .github/workflows/sanity.yml so CI only runs the sanity job.

Testing

• Ran the new package unit tests with go test ./gimble-ros/internal/ros -v which executed graph_errors_test.go and graph_timeout_test.go and they passed.
• The updated tests/integration.sh improvements were exercised manually during development to validate behavior under transient container startup timing.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Gimble/.github/workflows/sanity.yml

Line 22 in 73ae316

run: ./tests/sanity.sh

Keep integration test job in CI workflows

Removing the integration job here leaves ./tests/integration.sh unreferenced by CI; I checked the current workflow set (.github/workflows/sanity.yml and .github/workflows/publish-apt.yml) and neither runs it now. That drops automated coverage of the Docker/ROS path (gim ... --container), so container-specific regressions can merge undetected even though this commit hardens that script.

---

Gimble/gimble-ros/internal/ros/graph.go

Line 162 in 73ae316

tctx, cancel := WithTimeout(ctx, graphCommandTimeout)

Reintroduce an overall timeout for graph collection

This per-command timeout wrapper creates a fresh deadline for every ROS call, so CollectSystemGraph no longer has a single end-to-end cap and can run for many sequential 10s waits (for example, one hung ros2 node info per node). With callers like the CLI using a broad parent timeout, this can turn graph/analyze operations into long stalls instead of failing fast as before the shared timeout was removed.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".