Dynamic workflow deadline leaks an httpx pool per LLM call and raises "Event loop is closed"

[FacaHD]

Summary

Long scans run with a workflow deadline (SKILLSPECTOR_MAX_WORKFLOW_SECONDS) emit repeated unretrieved-task errors and steadily leak HTTP connection pools:

Task exception was never retrieved
future: <Task finished coro=<AsyncClient.aclose() done, defined at ...> exception=RuntimeError('Event loop is closed')>
RuntimeError: Event loop is closed

Steps to reproduce

1. Set SKILLSPECTOR_MAX_WORKFLOW_SECONDS so a dynamic deadline is active.
2. Run a scan large enough to issue more than 128 LLM calls.
3. Watch stderr: Event loop is closed appears repeatedly once the scan passes the cache limit.

Root cause

Under a workflow deadline every batch is sent with whatever time is left, so each call carries a slightly different timeout float. _model_for_call() expressed that by building a fresh chat model per call.

LangChain caches the underlying httpx client in an lru_cache keyed on (base_url, timeout, socket_options). A deadline that shrinks by a few milliseconds per call is therefore a new cache key every call — a long scan opens one connection pool per LLM request and starts evicting the oldest once it passes 128.

Eviction is what turns the leak into a crash. Each analyzer node is synchronous and reaches run_async() → asyncio.run(), so it owns its event loop. A pool evicted while a later analyzer is running belongs to an earlier, already-closed loop, and the OpenAI SDK's client finaliser reacts to garbage collection with asyncio.get_running_loop().create_task(self.aclose()) — closing the earlier analyzer's sockets from the current analyzer's loop. Nothing awaits that task, so the RuntimeError surfaces only as "Task exception was never retrieved", once per eviction.

Impact

• Noisy, misleading stderr output on every large scan run under a deadline.
• One httpx connection pool per LLM request instead of one per analyzer.

A 200-call reproduction against the openai 3.11 / httpx 2 combination that reported the failure produced 75 unretrieved Event loop is closed tasks and 220 httpx clients.

Expected behaviour

A bounded scan should reuse one connection pool per analyzer and complete without unretrieved-task errors, while still honouring the dynamic deadline.

Environment

• openai 3.11 with httpx 2.x
• Scan run with SKILLSPECTOR_MAX_WORKFLOW_SECONDS set