andrewmusselman opened a new issue, #21:
URL: https://github.com/apache/tooling-gofannon/issues/21
### Summary
Today every agent run executes inside the request thread, with the SSE
response holding the connection open for the run's full duration. Closing the
browser tab orphans the run from the user's view (the asyncio.Task keeps
executing but they have no way to reconnect). Restarting the api container
kills every in-flight run. This blocks true multi-tenancy and basic operational
visibility.
### Details
**Problem:**
- The browser tab *is* the run. Refresh / navigate / close drops the SSE;
the run keeps executing on the backend but the user can't see results when they
return.
- `uvicorn --reload` kills in-flight runs in dev. Felt repeatedly.
- No way to list "what's running right now" — operators cannot answer
"should I kill my first run before I start another?" without tailing logs.
- All downstream features (stop button, home-page running-jobs module, runs
page subscribe-by-id, headless invocation) require a run identity that exists
independently of the request that started it.
### Proposed solution
Run registry at the user-service level — a dict keyed by `run_id`, value is
a `RunRecord` dataclass:
```python
@dataclass
class RunRecord:
run_id: str
user_id: str
agent_name: str
started_at: datetime
status: Literal["running", "success", "error", "stopped"]
trace: Trace
queue: asyncio.Queue
task: asyncio.Task
cancel_token: CancelToken
result: Optional[dict]
error: Optional[str]
completed_at: Optional[datetime]
schema_warnings: Optional[list]
ops_log: Optional[list]
```
Two storage tiers: in-memory dict for active and recently-completed runs
(last hour); persisted to CouchDB on completion when the in-memory entry is
evicted.
New endpoints:
```
POST /agents/run-code/start # Returns {run_id} immediately, run
continues in background
GET /runs # List current user's runs
GET /runs/{run_id} # Fetch run state
GET /runs/{run_id}/stream # SSE subscribe (live + replay)
POST /runs/{run_id}/stop # Set cancel token; respond 202
DELETE /runs/{run_id} # Clear in-memory entry
POST /agents/run-code/stream # Existing — kept compatible; first frame
includes run_id
```
Each run is owned by the user who started it. `GET /runs/{run_id}` returns
403 if `request.user.uid != run.user_id` (and once ISSUE-002 lands, also widens
to `run.workspace_id IN session.workspace_ids`).
### Acceptance Criteria
- [ ] Fixed: `RunRecord` dataclass and in-memory registry implemented
- [ ] Fixed: `POST /agents/run-code/start` returns `run_id` and dispatches
asyncio.Task
- [ ] Fixed: `GET /runs` lists runs for current user
- [ ] Fixed: `GET /runs/{run_id}/stream` subscribes via SSE with
replay-then-live
- [ ] Fixed: completed runs persisted to CouchDB after eviction
- [ ] Fixed: existing `POST /agents/run-code/stream` endpoint emits `run_id`
in first frame (backward-compatible)
- [ ] Test added: closing the SSE connection does not kill the underlying run
- [ ] Test added: reconnecting to a running run replays past events before
going live
- [ ] Test added: user A cannot read user B's run
- [ ] Test added: completed runs are queryable for at least 1 hour
- [ ] Documentation: API reference for new endpoints
### References
- File: `webapp/packages/api/user-service/routes.py:659` (existing stream
endpoint)
- File: `webapp/packages/api/user-service/dependencies.py:148-200`
(_execute_agent_code)
- Tracker: FIXES.md item #2
### Priority
**High** - Foundation for ISSUE-005 (runs page subscribes), ISSUE-006 (home
page Running Jobs), ISSUE-007 (stop button), and ISSUE-015 (headless invocation
auth). Single-process assumption acceptable for now; scaling to replicas
requires sticky sessions or shared store (future work).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
