From 0030292de198f1aa8079b393e007d843767efabc Mon Sep 17 00:00:00 2001
From: Chisanan232 <chi10211201@cycu.org.tw>
Date: Mon, 1 Jun 2026 21:21:34 +0800
Subject: [PATCH] =?UTF-8?q?=F0=9F=90=9B=20(ci):=20Retry=20gh=20release=20d?=
 =?UTF-8?q?ownload=20on=20cross-repo=20race=20(release=20not=20found)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

release-node.yml fires on the same tag-push event as agent-assembly's
Release workflow, in parallel. agent-assembly takes ~10 min to build
+ publish the binaries that release-node depends on, so the first
gh release download attempt consistently fails with:

  release not found

(observed on v0.0.1-alpha.2 and v0.0.1-alpha.3 dry-runs; would have
failed every prior alpha too if not blocked at earlier steps).

Fix: wrap gh release download in a retry loop. Up to 20 attempts
with 60-second sleeps = 20 min ceiling waiting for the Release.
Distinguish 'release not found' (race condition → retry) from
other errors (auth failure, rate limit, etc. → fail-fast).

Verified locally:

  * retry-on-race: 3-attempt simulation with first 2 returning
    'release not found' → third succeeds → loop exits cleanly
  * fail-fast on other error: 'HTTP 403 rate limit' → grep miss
    on 'release not found' → exit 1 without retry

Trade-off vs alternatives considered:

  - workflow_run cross-repo trigger: NOT possible; workflow_run
    only fires for same-repo workflows
  - repository_dispatch from agent-assembly: cleaner architecturally
    but needs PAT + extra cross-repo wiring. Defer until needed.
  - Manual workflow_dispatch only: loses automation entirely.

Retry-with-backoff is the smallest change that solves the race.
Migrate to repository_dispatch later if polling cost becomes real.

Tracked: AAASM-2328
---
 .github/workflows/release-node.yml | 30 ++++++++++++++++++++++++++----
 1 file changed, 26 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/release-node.yml b/.github/workflows/release-node.yml
index 28286a6..d688bb3 100644
--- a/.github/workflows/release-node.yml
+++ b/.github/workflows/release-node.yml
@@ -77,10 +77,32 @@ jobs:
         run: |
           set -euo pipefail
           mkdir -p bin-staging
-          gh release download "$RELEASE_TAG" \
-            --repo AI-agent-assembly/agent-assembly \
-            --pattern "aasm-*.tar.gz" \
-            --dir bin-staging
+          # Cross-repo race: release-node fires on the same tag-push event as
+          # agent-assembly's Release workflow, in parallel. agent-assembly takes
+          # ~10 min to build + publish the binaries we depend on. Retry up to
+          # 20 times × 60s = 20 min ceiling waiting for the Release object.
+          # AAASM-2328.
+          MAX_ATTEMPTS=20
+          for attempt in $(seq 1 $MAX_ATTEMPTS); do
+            if gh release download "$RELEASE_TAG" \
+                 --repo AI-agent-assembly/agent-assembly \
+                 --pattern "aasm-*.tar.gz" \
+                 --dir bin-staging 2>&1 | tee /tmp/gh-rd.log; then
+              echo "✓ Downloaded on attempt ${attempt}/${MAX_ATTEMPTS}"
+              break
+            fi
+            # Distinguish 'release not found' (race — retry) from other errors (fail-fast)
+            if ! grep -q 'release not found' /tmp/gh-rd.log; then
+              echo "::error::gh release download failed with non-race error — aborting retry"
+              exit 1
+            fi
+            if [ "$attempt" -eq "$MAX_ATTEMPTS" ]; then
+              echo "::error::Release v${RELEASE_TAG} never appeared after $((MAX_ATTEMPTS * 60))s — agent-assembly Release pipeline likely failed"
+              exit 1
+            fi
+            echo "Attempt ${attempt}/${MAX_ATTEMPTS}: release not yet published, sleeping 60s..."
+            sleep 60
+          done
           ls -la bin-staging/
 
       - name: Stage runtime binaries into runtime-* sub-packages