Fix ECONNREFUSED driver_exit failures in Copilot SDK engine by gating on api-proxy listener readiness by Copilot · Pull Request #52994 · github/gh-aw

Copilot · 2026-08-16T00:14:13Z

Daily Ambient Context Optimizer (and other docker-sbx + Copilot SDK workflows) failed repeatedly because the harness sent its first chat-completion request to the api-proxy sidecar before its per-provider listener was actually accepting connections, exhausting all retries on ECONNREFUSED and ending the run with zero agent turns (driver_exit).

Root cause

api-proxy reports /reflect healthy and passes key validation before the per-provider listener (e.g. http://api-proxy:10002) is actually accepting TCP connections.
The Copilot SDK driver's 3 retry attempts all land inside this startup window and fail identically, so the run dies within ~1 minute with no agent turns recorded.
This exact defect was previously diagnosed and fixed in Harden Copilot api-proxy startup: verify listener accept readiness and absorb first-request ECONNREFUSED #52619, but that PR was closed without merging, leaving the underlying race unresolved.

Fix

awf_reflect.cjs: added waitForProviderListenerReady(), which probes a real TCP connect (TLS handshake for https:// baseUrls) against a provider listener with a bounded timeout/retry cadence.
copilot_harness.cjs: in SDK multi-provider mode, gates on listener readiness for every unique provider baseUrl before the first request, and classifies a first-attempt ECONNREFUSED as a one-shot short-backoff retry instead of burning the full retry budget on a known-transient race.
Re-applied the prior fix's test coverage in awf_reflect.test.cjs / copilot_harness.test.cjs, resolving conflicts with unrelated test additions merged since.

// copilot_harness.cjs (SDK mode): require real listener readiness before first request
const readiness = await waitForProviderListenerReady({
  baseUrl: providerBaseUrlToProbe,
  timeoutMs: AWF_PROVIDER_LISTENER_READY_TIMEOUT_MS,
  logger: log,
});
if (!readiness.ok) {
  emitInfrastructureIncomplete(
    `api-proxy provider listener was not ready at ${providerBaseUrlToProbe} before first Copilot SDK request (${readiness.error}).`
  );
  process.exit(1);
}

Fixes [aw] Daily Ambient Context Optimizer failed #52959

Generated by 👨‍🍳 PR Sous Chef · gpt54 · 8.49 AIC · ⌖ 5.67 AIC · ⊞ 6.3K · ◷
_{Comment /souschef to run again}

Run: https://github.com/github/gh-aw/actions/runs/31921676473> Generated by 👨‍🍳 PR Sous Chef · gpt54 · 3.82 AIC · ⌖ 5.65 AIC · ⊞ 8.7K · ◷

_{Comment /souschef to run again}

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

github-actions · 2026-08-16T00:31:48Z

Triage

Category: bug (WIP self-debug)
Risk: low
Score: 20/100 (impact 5, urgency 10, quality 5)
Recommendation: defer

Zero-diff WIP PR auto-created by an agent to debug a failed Daily Ambient Context Optimizer run. No files changed yet. Revisit once a diff lands or close if stale.

Generated by 🔧 PR Triage Agent · auto · 40.8 AIC · ⌖ 2.37 AIC · ⊞ 8.1K · ◷

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot

Pull request overview

Adds provider-listener readiness checks to prevent Copilot SDK startup races.

Changes:

Probes provider TCP/TLS listeners before SDK requests.
Adds one-shot ECONNREFUSED retry handling.
Adds readiness and retry tests.

Show a summary per file

File	Description
`actions/setup/js/awf_reflect.cjs`	Implements listener readiness probing.
`actions/setup/js/awf_reflect.test.cjs`	Tests probe behavior and timeouts.
`actions/setup/js/copilot_harness.cjs`	Gates SDK startup and handles refused connections.
`actions/setup/js/copilot_harness.test.cjs`	Adds detection and retry tests.

Review details

Files reviewed: 4/4 changed files
Comments generated: 3
Review effort level: Balanced

+          // provider died in the narrow window between the probe and the first request — retry
+          // once as a fresh run. Later attempts (attempt > 0) fall through to the generic retry
+          // handling below instead of taking this one-shot path.
+          if (attempt === 0 && isConnectionRefused && maxRetries > 0) {


+      socket.once(readyEvent, () => {
+        clear();
+        // Remove the error listener before tearing down the socket: a successful handshake has
+        // already settled this probe as ready, and a late/trailing error (e.g. an abrupt RST)
+        // must not overwrite lastError or otherwise affect the already-settled result.
+        socket.removeAllListeners("error");
+        socket.destroy();
+        settle(true);
+      });
+      socket.once("error", err => {
+        clear();
+        lastError = getErrorMessage(err);
+        socket.destroy();
+        settle(false);
+      });


+    function decideConnectionRefusedRetry(isConnectionRefused, attempt, maxRetries) {
+      if (attempt === 0 && isConnectionRefused && maxRetries > 0) {
+        return { action: "retry", nextDelayMs: FIRST_CONNECTION_REFUSED_RETRY_DELAY_MS };
+      }
+      return { action: "continue" };


gh-aw-bot · 2026-08-16T01:05:27Z

@copilot this PR looks ready for a finishing pass.

Please:

address any open inline review feedback from copilot-pull-request-reviewer
refresh the branch if needed
run the pr-finisher skill before returning it for maintainer review

Run: https://github.com/github/gh-aw/actions/runs/31918518969

Generated by 👨‍🍳 PR Sous Chef · gpt54 · 8.49 AIC · ⌖ 5.67 AIC · ⊞ 6.3K · ◷
_{Comment /souschef to run again}

…or listener, test production helper Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Copilot · 2026-08-16T01:20:04Z

@copilot this PR looks ready for a finishing pass.

Please:...

Finishing pass done in 047e2a8:

SDK-mode scoping: the one-shot ECONNREFUSED retry now goes through a new exported shouldRetryFirstConnectionRefused({ copilotSDKMode, attempt, isConnectionRefused, maxRetries }), so CLI failures keep the generic retry policy (including --continue) instead of being forced into a fresh run.
Socket error listener: waitForProviderListenerReady() no longer calls removeAllListeners("error"). The probe settles as ready before destroy(), the error listener stays installed, and the handler ignores already-settled probes — so a late RST can't surface as an uncaught EventEmitter error. The timeout path settles before destroy() for the same reason.
Tests: the inlined retry-decision helper in copilot_harness.test.cjs was replaced with tests that call the exported production helper, including a CLI-mode guard case; awf_reflect.test.cjs now asserts the error listener is still installed at destroy() time.

Branch is up to date with main (0 commits behind). Prior CI on this branch had no failures — the completed CJS run on 839b739 was green, and the remaining runs are action_required (awaiting maintainer approval). Local validation (make agent-report-progress, incl. format, JS lint, typecheck, impacted tests, and make recompile lock-file sync) passes. CI on the new HEAD is unverified until a maintainer re-triggers it.

gh-aw-bot · 2026-08-16T02:26:06Z

@copilot this PR looks ready for a finishing pass.

Please:

address the unresolved inline review feedback from copilot-pull-request-reviewer:
refresh the branch if needed
run the pr-finisher skill before returning it for maintainer review

Run: https://github.com/github/gh-aw/actions/runs/31921676473

Generated by 👨‍🍳 PR Sous Chef · gpt54 · 3.82 AIC · ⌖ 5.65 AIC · ⊞ 8.7K · ◷
_{Comment /souschef to run again}

github-actions · 2026-08-18T22:26:13Z

🎉 This pull request is included in a new release.

Release: v0.87.1

Initial plan

9675800

Copilot AI assigned Copilot and pelikhan Aug 16, 2026

Copilot started work on behalf of pelikhan August 16, 2026 00:14 View session

Copilot AI linked an issue Aug 16, 2026 that may be closed by this pull request

[aw] Daily Ambient Context Optimizer failed #52959

Closed

Update; rm -rf /

906230d

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

github-actions Bot added pr-action:defer pr-agent:github-copilot pr-priority:low pr-risk:low pr-type:bug labels Aug 16, 2026

github-actions Bot mentioned this pull request Aug 16, 2026

[PR Triage Report] PR Triage Report — 2026-08-16 #53000

Closed

Fix clear() temporal-dead-zone ordering in listener readiness probe

839b739

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot AI changed the title ~~[WIP] Debug daily ambient context optimizer workflow failure~~ Fix ECONNREFUSED driver_exit failures in Copilot SDK engine by gating on api-proxy listener readiness Aug 16, 2026

Copilot finished work on behalf of pelikhan August 16, 2026 00:33

Copilot AI requested a review from pelikhan August 16, 2026 00:33

pelikhan marked this pull request as ready for review August 16, 2026 00:35

Copilot AI balanced review requested due to automatic review settings August 16, 2026 00:35

Copilot started reviewing on behalf of pelikhan August 16, 2026 00:35 View session

Copilot AI reviewed Aug 16, 2026

View reviewed changes

Merge branch 'main' into copilot/aw-debug-daily-ambient-workflow

d15e77d

github-actions Bot mentioned this pull request Aug 16, 2026

[pr-sous-chef] PR Sous Chef run report #52990

Closed

Copilot started work on behalf of gh-aw-bot August 16, 2026 01:05 View session

Address review: scope ECONNREFUSED retry to SDK mode, keep socket err…

047e2a8

…or listener, test production helper Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Copilot AI requested a review from gh-aw-bot August 16, 2026 01:20

Copilot finished work on behalf of gh-aw-bot August 16, 2026 01:20

Merge branch 'main' into copilot/aw-debug-daily-ambient-workflow

6dd4293

pelikhan merged commit 1f1a46a into main Aug 16, 2026
1 check failed

Copilot started work on behalf of gh-aw-bot August 16, 2026 02:26 View session

pelikhan deleted the copilot/aw-debug-daily-ambient-workflow branch August 16, 2026 02:26

Copilot stopped work on behalf of gh-aw-bot due to an error August 16, 2026 02:26
The session was cancelled by the user.

Navigation Menu

Uh oh!

Fix ECONNREFUSED driver_exit failures in Copilot SDK engine by gating on api-proxy listener readiness - #52994

Fix ECONNREFUSED driver_exit failures in Copilot SDK engine by gating on api-proxy listener readiness#52994
pelikhan merged 6 commits into
mainfrom
copilot/aw-debug-daily-ambient-workflow

Copilot AI commented Aug 16, 2026 •

edited by github-actions Bot

Loading

Uh oh!

github-actions Bot commented Aug 16, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

gh-aw-bot commented Aug 16, 2026

Uh oh!

Copilot AI commented Aug 16, 2026

Uh oh!

gh-aw-bot commented Aug 16, 2026

Uh oh!

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

Copilot AI commented Aug 16, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Root cause

Fix

Uh oh!

github-actions Bot commented Aug 16, 2026

Triage

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Review details

Uh oh!

gh-aw-bot commented Aug 16, 2026

Uh oh!

Copilot AI commented Aug 16, 2026

Uh oh!

gh-aw-bot commented Aug 16, 2026

Uh oh!

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Copilot AI commented Aug 16, 2026 •

edited by github-actions Bot

Loading