[tatoalo]

Problem

cloud task stream relays could mark the Redis stream complete when the sandbox /events SSE connection closed normally, even if the task run was still non-terminal.
ph-code then saw the stream end, fetched an in_progress run state, and reconnected, which can produce repeated stream churn.

Changes

• mark the Redis task-run stream complete after a normal sandbox stream close when the task run is already terminal
• reconnect on normal sandbox stream closure while the run is still non-terminal, and keep the existing retry budget behavior for exhausted reconnect attempts

[greptile-apps]

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
products/tasks/backend/temporal/process_task/activities/tests/test_relay_sandbox_events.py:337-364
**Missing unit tests for `_is_keepalive_event`**

`_is_keepalive_event` is a new exported helper analogous to `_is_session_update` and `_is_end_of_turn`, both of which have dedicated parameterized test classes. There are no direct unit tests for `_is_keepalive_event`, leaving the new reset-skipping logic (`if not _is_keepalive_event(event_data): reconnect_count = 0`) covered only implicitly by the integration test. A small parameterized suite (e.g. `{"type": "keepalive"} → True`, `{"type": "notification"} → False`, `{} → False`) would make the contract explicit and consistent with the rest of the file.

### Issue 2 of 2
products/tasks/backend/temporal/process_task/activities/tests/test_relay_sandbox_events.py:337-364
**`_mark_complete_if_run_is_terminal` `DoesNotExist` branch is untested**

The existing `_mark_error_unless_run_is_terminal` has a dedicated test for the `DoesNotExist` case (see `test_missing_run_marks_stream_error_on_relay_error`). The new `_mark_complete_if_run_is_terminal` has the analogous branch — `mark_error("Task run not found")` + `return True` — but no test covering it. If `TaskRunModel.DoesNotExist` is raised during a normal stream closure, the relay currently stops (returns `True`) and marks the stream errored, which may be surprising. A test would pin that behaviour and catch future regressions.

_{Reviews (1): Last reviewed commit: "fix(cloud-agent): handle stream churn" | Re-trigger Greptile}

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-04-30 10:32 UTC	Run
prod-us	✅ Deployed	2026-04-30 10:45 UTC	Run
prod-eu	✅ Deployed	2026-04-30 10:50 UTC	Run