fix: store asyncio task references to prevent GC mid-execution by teknium1 · Pull Request #3267 · NousResearch/hermes-agent

teknium1 · 2026-03-26T21:36:16Z

Summary

Python's asyncio event loop holds only weak references to tasks. Without a strong reference, the GC can destroy a task while it's awaiting I/O — silently dropping messages. Python 3.12+ made this more aggressive (docs, SO discussion).

Full audit of all 12 gateway platform adapters found 6 untracked create_task calls across 6 files:

Per-message tasks (tracked via `_background_tasks` set from base class)

webhook.py — handle_message task
sms.py — handle_message task
signal.py — SSE response aclose task

Long-running infrastructure tasks (stored in named instance vars)

slack.py — Socket Mode handler → self._socket_mode_task
discord.py — bot client → self._bot_task
whatsapp.py — message poll loop → self._poll_task (2 call sites)

All other adapters (telegram, mattermost, matrix, email, homeassistant, dingtalk) already tracked their tasks correctly.

Salvaged from #3160 by @memosr — expanded from 1 file to 6.

Test plan

Gateway tests: 1461 passed
Full suite: 6226 passed (1 pre-existing failure deselected)

Python's asyncio event loop holds only weak references to tasks. Without a strong reference, the garbage collector can destroy a task while it's awaiting I/O — silently dropping messages. Python 3.12+ made this more aggressive. Audit of all gateway platform adapters found 6 untracked create_task calls across 6 files: Per-message tasks (tracked via _background_tasks set from base class): - gateway/platforms/webhook.py: handle_message task - gateway/platforms/sms.py: handle_message task - gateway/platforms/signal.py: SSE response aclose task Long-running infrastructure tasks (stored in named instance vars): - gateway/platforms/slack.py: Socket Mode handler (_socket_mode_task) - gateway/platforms/discord.py: bot client (_bot_task) - gateway/platforms/whatsapp.py: message poll loop (_poll_task, 2 sites) All other adapters (telegram, mattermost, matrix, email, homeassistant, dingtalk) already tracked their tasks correctly. Salvaged from PR #3160 by memosr — expanded from 1 file to 6.

Replace per-request aiohttp.ClientSession() in every WhatsApp adapter method with a single persistent self._http_session, matching the pattern used by Mattermost, HomeAssistant, and SMS adapters. Changes: - Create self._http_session in connect(), close in disconnect() - All bridge HTTP calls (send, edit, send-media, typing, get_chat_info, poll_messages) now use the shared session - Explicitly cancel _poll_task on disconnect() instead of relying solely on self._running = False - Health-check sessions in connect() remain ephemeral (persistent session not yet created at that point) - Remove per-method ImportError guards for aiohttp (always available when gateway runs via [messaging] extras) Salvaged from PR #1851 by Himess. The _poll_task storage was already on main from PR #3267; this adds the disconnect cancellation and the persistent session. Tests: 4 new tests for session close, already-closed skip, poll task cancellation, and done-task skip.

…esearch#3267) Python's asyncio event loop holds only weak references to tasks. Without a strong reference, the garbage collector can destroy a task while it's awaiting I/O — silently dropping messages. Python 3.12+ made this more aggressive. Audit of all gateway platform adapters found 6 untracked create_task calls across 6 files: Per-message tasks (tracked via _background_tasks set from base class): - gateway/platforms/webhook.py: handle_message task - gateway/platforms/sms.py: handle_message task - gateway/platforms/signal.py: SSE response aclose task Long-running infrastructure tasks (stored in named instance vars): - gateway/platforms/slack.py: Socket Mode handler (_socket_mode_task) - gateway/platforms/discord.py: bot client (_bot_task) - gateway/platforms/whatsapp.py: message poll loop (_poll_task, 2 sites) All other adapters (telegram, mattermost, matrix, email, homeassistant, dingtalk) already tracked their tasks correctly. Salvaged from PR NousResearch#3160 by memosr — expanded from 1 file to 6.

…Research#3818) Replace per-request aiohttp.ClientSession() in every WhatsApp adapter method with a single persistent self._http_session, matching the pattern used by Mattermost, HomeAssistant, and SMS adapters. Changes: - Create self._http_session in connect(), close in disconnect() - All bridge HTTP calls (send, edit, send-media, typing, get_chat_info, poll_messages) now use the shared session - Explicitly cancel _poll_task on disconnect() instead of relying solely on self._running = False - Health-check sessions in connect() remain ephemeral (persistent session not yet created at that point) - Remove per-method ImportError guards for aiohttp (always available when gateway runs via [messaging] extras) Salvaged from PR NousResearch#1851 by Himess. The _poll_task storage was already on main from PR NousResearch#3267; this adds the disconnect cancellation and the persistent session. Tests: 4 new tests for session close, already-closed skip, poll task cancellation, and done-task skip.