fix(release): ride out transient npm registry failures

.github/workflows/release.yml

@@ -99,17 +99,25 @@ jobs:
       # OIDC migration replaces this isolated auth step only. Candidate and
       # promotion behavior deliberately know nothing about credential source.
       - name: verify npm publisher identity
-        run: npm whoami --registry=https://registry.npmjs.org
+        shell: bash
+        run: |
+          for attempt in $(seq 1 10); do
+            if npm whoami --registry=https://registry.npmjs.org; then exit 0; fi
+            echo "npm whoami failed (attempt ${attempt}/10); retrying after the registry settles"
+            sleep 30
+          done
+          exit 1
         env:
           NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
       - run: node script/release-packages.mjs publish
         env:
           NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
       # Staged publishing can accept a publish without making it installable.
-      # Surface the maintainer approval queue in the failing run's log.
+      # Surface the maintainer approval queue in the failing run's log. Bound
+      # the npx download so a registry outage cannot hang the job for hours.
       - name: list npm staged releases awaiting approval
         if: failure()
-        run: npx --yes npm@12 stage list || true
+        run: timeout 180 npx --yes npm@12 stage list || true
         env:
           NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
 

script/release-packages.mjs

@@ -27,9 +27,24 @@ const problems = [
 if (problems.length > 0) throw new Error(`release preflight failed: ${problems.join('; ')}`)
 for (const pkg of packages) pkg.filename = join(ROOT, '.artifacts', 'pack', pkg.filename)
 
+/** npm registry failures that are worth retrying inside one release step. */
+const RETRYABLE_NPM = /\b(?:E503|EAI_AGAIN|ETIMEDOUT|ECONNRESET)\b|503 Service Unavailable|network/u
+
+/** Block one release step between registry attempts without an async refactor. */
+function sleepSync(ms) {
+  Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, ms)
+}
+
 function npmView(spec, field) {
-  const result = spawnSync('npm', ['view', spec, field, '--json'], { cwd: ROOT, encoding: 'utf8' })
-  return readNpmView(result, spec, field)
+  let last
+  for (let attempt = 1; attempt <= 4; attempt += 1) {
+    const result = spawnSync('npm', ['view', spec, field, '--json'], { cwd: ROOT, encoding: 'utf8' })
+    const failure = `${result.stderr ?? ''}${result.error?.message ?? ''}`
+    if (result.status === 0 || !RETRYABLE_NPM.test(failure)) return readNpmView(result, spec, field)
+    last = new Error(`npm view ${spec} ${field} failed during a registry outage (attempt ${attempt}/4)`)
+    if (attempt < 4) sleepSync(attempt * 3_000)
+  }
+  throw last
 }
 
 /**