Self-improving AI Employees for everyone
Python ai-agentsai-employeeai-employeesassistantautonomous-agentsco-employeescompanygrok-botinstinctloop-agentsmuseown-your-datapersonalself-hosted
fix(deployment): a running generation keeps the slug it started with
trohitg3c4cf815a0f0
diff --git a/CLAUDE.md b/CLAUDE.mdindex 568d395a..3c4f7592 100644--- a/CLAUDE.md+++ b/CLAUDE.md@@ -1092,7 +1092,7 @@ Single source of truth: [`server/services/workflow_naming.py`](./server/services`node_label` is `node.data.label` (the F2-renamed canvas label) when set, falling back to `node.type` (`chatTrigger` / `telegramReceive` / `aiAgent` / etc.). Computed once at deploy time via `node_label_slug(node)` and passed into Temporal workflows as `trigger_label` in `listener_data`; agent child ids use the node id, not the label. Stable `TriggerNodeId` / `EventWorkflowId` Search Attributes still use the immutable node_id / UUID so admin queries don't break across rename.-**Rename path** — there is NO dedicated rename endpoint. The frontend's auto-save chain (`TopToolbar` inline edit → `updateWorkflow({name})` → debounced save → REST `POST /api/database/workflows` → `services.workflow_storage.handlers.handle_save_workflow`) IS the rename path. When `name` changes between saves, the handler (1) allocates a fresh slug via `next_available_slug`, (2) `database.rename_workflow` updates name + slug atomically (id UUID stays put), (3) renames the on-disk workspace dir via `Path.rename()`, (4) broadcasts a CloudEvents `workflow.renamed` envelope (`broadcaster.broadcast_workflow_lifecycle("renamed", workflow_id=..., name=..., slug=..., old_slug=...)`) so other tabs invalidate their workflows query.+**Rename path** — the editor has NO dedicated rename endpoint: its auto-save chain (`TopToolbar` inline edit → `updateWorkflow({name})` → debounced save → REST `POST /api/database/workflows` → `services.workflow_storage.handlers.handle_save_workflow`) IS its rename path. Home's `rename_employee` renames through `rename_saved_workflow` in the same module, without touching the graph. When `name` changes, either path (1) allocates a fresh slug via `next_available_slug`, (2) `database.rename_workflow` updates name + slug atomically (id UUID stays put), (3) renames the on-disk workspace dir via `Path.rename()`, (4) broadcasts a CloudEvents `workflow.renamed` envelope (`broadcaster.broadcast_workflow_lifecycle("renamed", workflow_id=..., name=..., slug=..., old_slug=...)`) so other tabs invalidate their workflows query. A rename leaves a running generation's Temporal ids alone: its listeners and cron Schedules keep the slug its Start recorded in the control row's `graph_snapshot` (`workflow_slug`, read by `DeploymentManager._generation_slug`), so the re-arm after a restart registers the same ids instead of a second set beside them, which made every event start two runs. The next Start names them by the new slug.**Invariants** (locked by `tests/services/test_workflow_naming.py` + `test_workflow_rename.py` — 39 tests):- First creation always gets `_1` suffix (no bare-base slugs).diff --git a/docs-internal/temporal-workflow-control.md b/docs-internal/temporal-workflow-control.mdindex 435debc4..dd3b2ca5 100644--- a/docs-internal/temporal-workflow-control.md+++ b/docs-internal/temporal-workflow-control.md@@ -199,7 +199,9 @@ the control plane is hardened against Temporal's per-run event-history ceiling(admission, trigger pause flags, cron schedule pause). Re-arm is idempotent:controller `register_trigger` is keyed by listener id, legacy listenerstarts use `USE_EXISTING`, and cron schedule creation preserves server-owned- pause state.+ pause state. Listener and Schedule ids use the slug the generation's Start+ recorded in its snapshot (`workflow_slug`), not the saved one, so a rename+ while it runs cannot give the re-arm a second set of ids.- **No lifetime caps.** Spawned graph runs, agent children, and delegated-taskrunners no longer carry 1-2h execution/run timeouts (Temporal's timers keepticking through a pause, so the caps silently terminated paused work).diff --git a/docs-internal/workflow-schema.md b/docs-internal/workflow-schema.mdindex c64eaad5..af7a77ba 100644--- a/docs-internal/workflow-schema.md+++ b/docs-internal/workflow-schema.md@@ -39,7 +39,7 @@ A workflow JSON document contains:},"slug": {"type": "string",- "description": "Human-readable identifier (e.g. AI_Assistant_1). Derived from name via services.workflow_naming.next_available_slug. Mutable on rename — recomputed + propagated to workspace dir + Temporal Web UI prefix + cron Schedule IDs. UNIQUE across active workflows.",+ "description": "Human-readable identifier (e.g. AI_Assistant_1). Derived from name via services.workflow_naming.next_available_slug. Mutable on rename — recomputed + propagated to the workspace dir; Temporal ids and cron Schedule IDs follow at the next Start (a running generation keeps the slug it started with). UNIQUE across active workflows.","pattern": "^[A-Za-z0-9]+(_[0-9]+)?$"},"nodes": {diff --git a/server/services/deployment/control.py b/server/services/deployment/control.pyindex bffc4eac..d763a8d9 100644--- a/server/services/deployment/control.py+++ b/server/services/deployment/control.py@@ -130,6 +130,8 @@ class WorkflowControlService:raise ValueError("workflow_already_started")generation = (latest.generation + 1) if latest else 1execution_id = await self.database.allocate_execution_id(workflow_id)+ saved = await self.database.get_workflow(workflow_id)+ slug = getattr(saved, "slug", None)control = WorkflowControlExecution(id=f"workflow-control:{workflow_id}:{generation}", workflow_id=workflow_id,generation=generation, execution_id=execution_id, root_execution_id=execution_id,@@ -141,6 +143,9 @@ class WorkflowControlService:"owner_id": safe_graph["owner_id"],"nodes": nodes,"edges": edges,+ # What this generation's Temporal ids (listeners, Schedules)+ # are named by, fixed at Start: see DeploymentManager.deploy.+ "workflow_slug": slug if isinstance(slug, str) and slug else None,},idempotency_key=idempotency_key,# Revisions are monotonic across generations so a delayed requestdiff --git a/server/services/deployment/manager.py b/server/services/deployment/manager.pyindex 3ade1a95..e33da6ad 100644--- a/server/services/deployment/manager.py+++ b/server/services/deployment/manager.py@@ -196,11 +196,12 @@ class DeploymentManager:"deployment_id": self._deployments[workflow_id].deployment_id,}- # Human-readable slug for Temporal Web UI prefixing. Falls back- # to workflow_id when the DB row is missing OR slug is empty+ # Human-readable slug for Temporal Web UI prefixing. A generation+ # keeps the slug it started with; otherwise the saved one, falling+ # back to workflow_id when the DB row is missing OR slug is empty# (one-off deploys / tests without a saved record).wf = await self.database.get_workflow(workflow_id)- workflow_slug = (wf and wf.slug) or workflow_id+ workflow_slug = await self._generation_slug(workflow_id, generation) or (wf and wf.slug) or workflow_id# Setupdeployment_id = f"deploy_{workflow_id}_{int(time.time() * 1000)}"@@ -668,6 +669,23 @@ class DeploymentManager:"""return f"{workflow_slug}-{trigger_label}"+ async def _generation_slug(self, workflow_id: str, generation: int) -> Optional[str]:+ """The slug a generation's listeners and Schedules were named by at Start.++ A rename moves the saved slug while those keep their ids. Re-armed+ after a restart under the new slug, the generation would register a+ second listener (and Schedule) beside the first, and every event+ would start two runs. ``None`` for a generation that recorded none.+ """+ if generation <= 0:+ return None+ lookup = self.database.get_latest_workflow_control(workflow_id)+ control = await lookup if inspect.isawaitable(lookup) else None+ if control is None or control.generation != generation:+ return None+ slug = (control.graph_snapshot or {}).get("workflow_slug")+ return slug if isinstance(slug, str) and slug else None+@staticmethoddef _trigger_kind_for(node_type: str) -> str:"""Coarse-grained classification for the ``EventTriggerKind``diff --git a/server/tests/services/test_workflow_control.py b/server/tests/services/test_workflow_control.pyindex 86e01986..461ecc94 100644--- a/server/tests/services/test_workflow_control.py+++ b/server/tests/services/test_workflow_control.py@@ -58,6 +58,28 @@ async def test_begin_generation_is_idempotent():assert serialize_control(control)["can_pause"] is False+async def test_generation_records_the_slug_its_temporal_ids_are_named_by():+ db = AsyncMock()+ db.get_workflow_control_by_idempotency_key.return_value = None+ db.get_latest_workflow_control.return_value = None+ db.get_workflow.return_value = SimpleNamespace(slug="Ana_1")+ db.create_workflow_control.side_effect = lambda control: control+ service = WorkflowControlService(db)++ control, _ = await service.begin_generation(+ workflow_id="wf", nodes=[], edges=[], session_id="s", idempotency_key="start-1"+ )+ assert control.graph_snapshot["workflow_slug"] == "Ana_1"++ # A workflow that was never saved names its ids by its id instead.+ db.get_workflow.return_value = None+ control, _ = await service.begin_generation(+ workflow_id="wf", nodes=[], edges=[], session_id="s", idempotency_key="start-2"+ )+ assert control.graph_snapshot["workflow_slug"] is None++@pytest.mark.asyncioasync def test_generation_atomically_creates_and_archives_isolated_data_scope(control_database):service = WorkflowControlService(control_database)diff --git a/server/tests/test_deployment_canary_listener.py b/server/tests/test_deployment_canary_listener.pyindex d4c8a13e..c6ec6a44 100644--- a/server/tests/test_deployment_canary_listener.py+++ b/server/tests/test_deployment_canary_listener.py@@ -135,6 +135,68 @@ class TestDeterministicListenerId:assert a != b+class TestGenerationSlug:+ """A generation keeps the slug its Start recorded. Re-armed after a+ rename under the new slug, it registered a second listener beside the+ first with its controller, and every event started two runs."""++ @staticmethod+ def _manager(saved_slug: str, control):+ from types import SimpleNamespace++ from services.deployment.manager import DeploymentManager++ database = MagicMock()+ database.get_workflow = AsyncMock(return_value=SimpleNamespace(slug=saved_slug))+ database.get_latest_workflow_control = AsyncMock(return_value=control)+ broadcaster = MagicMock()+ broadcaster.update_node_status = AsyncMock()+ mgr = DeploymentManager(+ database=database,+ execute_workflow_fn=AsyncMock(),+ store_output_fn=AsyncMock(),+ broadcaster=broadcaster,+ )+ mgr._load_settings = AsyncMock()+ mgr._notify = AsyncMock()+ return mgr++ @staticmethod+ def _control(generation: int, slug=None):+ from types import SimpleNamespace++ snapshot: Dict[str, Any] = {"nodes": [], "edges": []}+ if slug is not None:+ snapshot["workflow_slug"] = slug+ return SimpleNamespace(generation=generation, graph_snapshot=snapshot)++ @pytest.mark.asyncio+ async def test_rearm_after_a_rename_keeps_the_generation_slug(self):+ mgr = self._manager("Ana_Ortiz_1", self._control(2, "Ana_1"))++ result = await mgr.deploy(nodes=[], edges=[], workflow_id="wf", graph_version=2, generation=2)++ assert result["success"] is True+ assert mgr._deployments["wf"].workflow_slug == "Ana_1"++ @pytest.mark.asyncio+ @pytest.mark.parametrize(+ "generation, control",+ [+ (0, None), # not a durable generation+ (2, None), # no control row+ (2, (3, "Ana_1")), # the latest row is another generation+ (2, (2, None)), # a snapshot from before the slug was recorded+ ],+ )+ async def test_otherwise_the_saved_slug(self, generation, control):+ mgr = self._manager("Ana_Ortiz_1", self._control(*control) if control else None)++ await mgr.deploy(nodes=[], edges=[], workflow_id="wf", graph_version=2, generation=generation)++ assert mgr._deployments["wf"].workflow_slug == "Ana_Ortiz_1"++# ---------------------------------------------------------------------------# C1d.2 — canary scope: flag + type# ---------------------------------------------------------------------------