#472)

# Issue
It's been described well in #464 :

> When the client-side connection is terminated, the EventConsumer stops
processing. As a result, any changes to the task state after the
disconnection are not persisted to the TaskStore. The task itself
continues running in the background, but its updated state is no longer
reflected in the TaskStore.

This has been addressed in this PR by simply adding a catch for
`(asyncio.CancelledError, GeneratorExit)` in the
`on_message_send_stream` method.

However, adding that revealed a difference in semantics between Python
3.13+ and <3.13 for `EventQueue.close()`. I have also addressed that.

# How it's reproduced

[@azyobuzin](https://github.com/azyobuzin) provided a detailed guide on
this in #464 . My only addition would be to add loggers for
`a2a.server.events.event_queue` and `a2a.server.events.event_consumer`
to get a better understanding of what's happening under the hood.

# Fix

## Code

- Ensure streaming continues persisting events after client disconnect
via background consumption by adding a catch for
`(asyncio.CancelledError, GeneratorExit)` in the
`on_message_send_stream` method.
- Align EventQueue.close() behavior on Python ≥3.13 and ≤3.12 (graceful
vs. immediate).

## Tests

### Event queue tests (`tests/server/events/test_event_queue.py`)
Added/updated tests to verify:

- Graceful close on ≥3.13 waits for drain and children.
- Immediate close clears queues and propagates.
- To support Python 3.10, when simulating ≥3.13 using sys.version_info,
inject a dummy queue.shutdown on asyncio.Queue so tests don’t fail on
runtimes without it. I've seen this pattern used in existing tests too.

### Request handler tests
(`tests/server/request_handlers/test_default_request_handler.py`)
Added `test_disconnect_persists_final_task_to_store` which tests the
flow described in issue #464 :

- Starts streaming, yields first event, then simulates client
disconnect.
- Background consumer persists the final Task to `InMemoryTaskStore`.
- Uses `wait_until` to await disappearance of the specific
`background_consume:{task_id}` task, then asserts `TaskState.completed`.

### General
Added a cleanup for lingering background tasks, I think it's an
improvement for my earlier PR where I've tracked background tasks.


# Misc
Ruff `0.13.0` now fails the check for unused variables, made some
minimal changes as suggested by the linter, irrelevant to the issue:
underscores for `_payload` and `_task_manager`.


Fixes #464

…475)

The underlying `aconnect_sse` library will overwrite the header and
timeout of what's set in the `httpx_client`.

From a user using adk `a2a_remote_agent` library, they would pass in these
information in the `httpx_client`. I have tried this fix in my local
setup.

🤖 I have created a release *beep* *boop*
---


##
[0.3.7](v0.3.6...v0.3.7)
(2025-09-22)


### Bug Fixes

* jsonrpc client send streaming request header and timeout field
([#475](#475))
([675354a](675354a))
* Task state is not persisted to task store after client disconnect
([#472](#472))
([5342ca4](5342ca4)),
closes [#464](#464)

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).