Verify typed startup failures across factory adapter and protocol

2026-11-04 11:47:05 +09:00
parent 326616d07e
commit 9e404b0021
1 changed files with 183 additions and 0 deletions
@@ -101,3 +101,186 @@ impl PendingRuntimeRestore {
        result
    }
 }
+
+#[cfg(test)]
+mod tests {
+    //! Connect the typed factory seam to the real Tokio adapter and protocol codec.
+    //! The external startup action is a deterministic failure, not a real domain
+    //! authorization decision. Resource disposition remains the existing backend result.
+    use super::*;
+    use crate::execution::WorkerExecutionContext;
+    use crate::identity::WorkerId;
+    use protocol::startup_failure::*;
+
+    struct TypedFailingFactory;
+    fn primary() -> StartupFailureAttachment {
+        worker::startup_failure::runtime_startup_attachment(RuntimeStartupCause {
+            stage: StartupStage::Feature,
+            reason: RuntimeStartupReason::FeatureFailed,
+        })
+        .unwrap()
+    }
+    fn failure() -> RuntimeRestoreFailure {
+        let mut failure = RuntimeRestoreFailure::unproven(
+            RuntimeRestoreStage::Controller,
+            "factory failed after receiving primary",
+        )
+        .with_startup_failure(primary());
+        failure.cleanup = RuntimeRestoreCleanup::Failed(RuntimeRestoreCleanupFailure::Workdir);
+        failure.cleanup_failure = Some(RuntimeRestoreCleanupFailure::Workdir);
+        failure.io_kind = Some(std::io::ErrorKind::PermissionDenied);
+        failure
+    }
+    #[async_trait]
+    impl RuntimeWorkerFactory for TypedFailingFactory {
+        async fn spawn_controller(
+            &self,
+            _: WorkerExecutionSpawnRequest,
+        ) -> Result<RuntimeWorkerController, RuntimeRestoreFailure> {
+            Err(failure())
+        }
+        async fn preflight_restore(
+            &self,
+            _: &WorkerExecutionRestoreRequest,
+        ) -> Result<(), RuntimeRestoreFailure> {
+            Err(failure())
+        }
+        async fn prepare_restore(
+            &self,
+            _: WorkerExecutionRestoreRequest,
+        ) -> Result<Box<dyn RuntimeRestoreCandidate>, RuntimeRestoreFailure> {
+            panic!("typed preflight rejection must not prepare live work")
+        }
+    }
+
+    fn request() -> WorkerExecutionSpawnRequest {
+        let worker_ref = WorkerRef::new(WorkerId::now_v7());
+        // No factory reads this archive: it is explicit launch input, not a retained approval.
+        let archive = crate::profile_archive::ProfileSourceArchive::build(
+            crate::profile_archive::ProfileSourceArchiveInput {
+                id: "profile-source-archive:typed-startup".into(),
+                entrypoints: BTreeMap::from([(
+                    "builtin:companion".into(),
+                    "profiles/default.dcdl".into(),
+                )]),
+                imports: BTreeMap::new(),
+                sources: BTreeMap::from([("profiles/default.dcdl".into(), "{}".into())]),
+            },
+        )
+        .unwrap();
+        let request = crate::catalog::CreateWorkerRequest {
+            worker_id: worker_ref.worker_id,
+            create_fingerprint: "typed-startup-failure".into(),
+            profile: crate::catalog::ProfileSelector::Builtin("builtin:companion".into()),
+            display_name: None,
+            profile_source: crate::catalog::ProfileSourceArchiveSource::Embedded { archive },
+            config_bundle: None,
+            initial_input: None,
+            workdir_attachment_requests: Vec::new(),
+            workdir_attachments: Vec::new(),
+            worker_observation_enabled: false,
+            worker_observation_grants: Vec::new(),
+            workspace_api: None,
+            memory_settings: None,
+            plugin_execution: None,
+            subjektiv_attached: false,
+            backend_job: None,
+        };
+        #[cfg(feature = "ws-server")]
+        let context = WorkerExecutionContext::new(
+            worker_ref.clone(),
+            Arc::new(|_, _| unreachable!("failed startup must not publish a Controller")),
+        );
+        #[cfg(not(feature = "ws-server"))]
+        let context = WorkerExecutionContext::new(worker_ref.clone());
+        WorkerExecutionSpawnRequest {
+            worker_ref,
+            request,
+            workspace_scope: None,
+            context,
+            workdir_attachments: BTreeMap::new(),
+            config_bundle: None,
+        }
+    }
+
+    fn assert_primary(result: &WorkerExecutionResult) {
+        assert_eq!(result.startup_failure.as_ref(), Some(&primary()));
+        let Some(error_contract::DecodeFailure::Known(received)) =
+            result.startup_failure::<RuntimeStartupCause>()
+        else {
+            panic!("typed primary missing")
+        };
+        assert_eq!(received.facts().cause().stage, StartupStage::Feature);
+        assert_eq!(
+            received.facts().cause().reason,
+            RuntimeStartupReason::FeatureFailed
+        );
+        assert_eq!(
+            received.facts().context().operations[0].operation,
+            "worker_startup"
+        );
+        assert!(
+            result
+                .secondary_failures
+                .contains(&StartupSecondaryFailure::Cleanup {
+                    failure: StartupCleanupFailure::Workdir
+                })
+        );
+        assert_eq!(
+            result.startup_io_kind,
+            Some(StartupIoKind::PermissionDenied)
+        );
+        assert!(!result.is_accepted());
+    }
+
+    #[test]
+    fn typed_factory_failure_survives_adapter_and_protocol_without_changing_disposition() {
+        let backend = WorkerRuntimeExecutionBackend::new(TypedFailingFactory).unwrap();
+        let WorkerExecutionSpawnResult::Errored(result) = backend.spawn_worker(request()) else {
+            panic!("cause attachment must not reclassify the existing spawn result")
+        };
+        assert_primary(&result);
+        let transferred: WorkerExecutionResult =
+            serde_json::from_slice(&serde_json::to_vec(&result).unwrap()).unwrap();
+        assert_eq!(transferred, result);
+        assert_primary(&transferred);
+        assert_eq!(
+            transferred.startup_resources.unwrap().cleanup,
+            StartupCleanup::Failed(StartupCleanupFailure::Workdir)
+        );
+    }
+
+    #[test]
+    fn typed_preflight_failure_survives_adapter_without_starting_live_preparation() {
+        let backend = WorkerRuntimeExecutionBackend::new(TypedFailingFactory).unwrap();
+        let spawn = request();
+        let request = WorkerExecutionRestoreRequest {
+            source_repair_possible: false,
+            operation_id: crate::execution::WorkerLifecycleOperationId::new(),
+            worker_ref: spawn.worker_ref,
+            request: spawn.request,
+            workspace_scope: None,
+            context: spawn.context,
+            previous_workdir_attachments: Vec::new(),
+            logical_workdir_attachments: Vec::new(),
+            workdir_attachments: BTreeMap::new(),
+            deferred_workdir_sessions: BTreeMap::new(),
+            config_bundle: None,
+        };
+        let result = backend.preflight_restore(&request).unwrap_err();
+        assert_eq!(
+            result.outcome,
+            crate::execution::WorkerExecutionOutcome::Rejected
+        );
+        assert_primary(&result);
+        let failure = backend
+            .run_on_adapter_runtime(async move {
+                Ok(prepare_restore_operation(&TypedFailingFactory, request)
+                    .await
+                    .err()
+                    .unwrap())
+            })
+            .unwrap();
+        assert_eq!(failure.startup_failure(), &primary());
+    }
+}