ci: retry transient registry failures

.gitea/workflows/deploy.yml

@@ -62,7 +62,23 @@ jobs:
           printf '%s\n' "$DEPLOY_KNOWN_HOSTS" > "$HOME/.ssh/known_hosts"
           chmod 600 "$HOME/.ssh/known_hosts"
           SSH_ARGS=(-i "$KEY_FILE" -o BatchMode=yes -o StrictHostKeyChecking=yes)
-          printf '%s' "$REGISTRY_TOKEN" | ssh "${SSH_ARGS[@]}" "$DEPLOY_USER@$DEPLOY_HOST" \
+          for attempt in 1 2 3; do
-            "docker login git.olns.shop -u admin --password-stdin"
+            if printf '%s' "$REGISTRY_TOKEN" | ssh "${SSH_ARGS[@]}" "$DEPLOY_USER@$DEPLOY_HOST" \
+              "docker login git.olns.shop -u admin --password-stdin"; then
+              break
+            fi
+            if [[ "$attempt" -eq 3 ]]; then
+              exit 1
+            fi
+            sleep "$((attempt * 5))"
+          done
           ssh "${SSH_ARGS[@]}" "$DEPLOY_USER@$DEPLOY_HOST" \
-            "cd /opt/olns && docker compose pull olons-admin && docker compose up -d --no-deps --wait olons-admin"
+            'cd /opt/olns && for attempt in 1 2 3; do
+              if timeout 180 docker compose pull olons-admin; then
+                break
+              fi
+              if [ "$attempt" -eq 3 ]; then
+                exit 1
+              fi
+              sleep $((attempt * 5))
+            done && docker compose up -d --no-deps --wait olons-admin'

tests/test_actions_workflow.py

@@ -29,6 +29,10 @@ class DeployWorkflowContractTest(unittest.TestCase):
         self.assertIn("docker compose pull olons-admin", self.workflow)
         self.assertIn("docker compose up -d --no-deps --wait olons-admin", self.workflow)
 
+    def test_retries_transient_remote_registry_failures(self):
+        self.assertGreaterEqual(self.workflow.count("for attempt in 1 2 3"), 2)
+        self.assertIn("timeout 180 docker compose pull olons-admin", self.workflow)
+
     def test_removes_old_kubernetes_and_registry_targets(self):
         self.assertNotIn("kubectl", self.workflow)
         self.assertNotIn("KUBECONFIG", self.workflow)