Skip to content

feat(observability): distinguish HTTP response selection from delivery failure - #13

Draft
afourniernv wants to merge 1 commit into
feat/5397-mcp-correlation/afourniernvfrom
feat/5398-terminal-delivery/afourniernv
Draft

afourniernv wants to merge 1 commit into
feat/5397-mcp-correlation/afourniernvfrom
feat/5398-terminal-delivery/afourniernv

Conversation

@afourniernv

Copy link
Copy Markdown
Owner

Summary

Track response delivery separately from policy and response selection. Keep one primary HTTP Activity record and emit an additional correlated record only for a failure after that record. Preserve the distinction between OpenShell-observed delivery and any downstream application result.

Stacked review 6/6. This draft targets the mcp-correlation draft in this fork; its diff contains only this layer, not the whole stack. The HTTP core is reviewed upstream in NVIDIA#5214; this dependent draft is on the author fork because upstream base branches are not available.

Related Issue

Closes NVIDIA#5398.
Tracking: NVIDIA#5210 (remains open).

Changes

  • This is the final and largest layer. Much of its raw diff moves existing response tests into a separate tests module; delivery-state and streaming changes still warrant separate review.
  • Complete, truncated, cancelled, client-disconnected and late middleware-denied deliveries are distinguishable with bounded byte/outcome metadata.
  • Routine successful requests still produce only one primary HTTP Activity record; post-header failures produce a correlated exception record.

Testing

  • Exact terminal-delivery layer: mise run pre-commit passed; 1,684 network tests passed.
  • Behavior tests added or retained for this layer; DCO-signed commit 1ad8639.
  • Cumulative native Hermes/Relay live validation on source 0764c1c4c: 13 selected workflows, 1,552 spans and 433 OCSF records; protected synthetic APIs and authenticated GitHub. The agent executed actual tools; no span parents were rewritten. This is cumulative evidence, not isolated-layer E2E qualification.
  • Cumulative mise run test rerun passed with four test threads. Initial unchanged VM cleanup and gateway fixture flakes passed isolated reruns and the final full run. Full mise run ci is not claimed.

Checklist

  • Follows Conventional Commits.
  • Commits are signed off (DCO).
  • Relevant crate and observability documentation updated.
  • Maintainer acceptance and upstream promotion, after the parent layer lands.

Tested upstream snapshot: 638974575 (merged NVIDIA#4789). Newer main has advanced; this draft does not claim validation against today's main. It is for incremental code review, not merge readiness. No Relay dependency or reverse workload-reporting endpoint is introduced.

…ivery

Signed-off-by: Alex Fournier <afournier@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant