mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-15 14:22:43 +00:00
105 lines
3.9 KiB
Python
105 lines
3.9 KiB
Python
"""Regression guard for #62151 — gateway cron must not wedge on the 2nd+ call.
|
|
|
|
Gateway-fired cron jobs hung forever on the 2nd+ API call because both the
|
|
non-streaming (``interruptible_api_call``) and the default streaming
|
|
(``interruptible_streaming_api_call``) paths run the request on a spawned
|
|
daemon worker thread. Inside the gateway's nested cron thread pools that extra
|
|
worker wedged before the socket opened; the same job succeeded via ``hermes
|
|
cron tick`` (foreground, no nested pools). Cron has no interactive interrupt
|
|
surface, so both paths now run inline on the conversation thread for the
|
|
``cron`` platform.
|
|
|
|
These tests pin: (1) the inline gate is cron-only, (2) the inline call runs on
|
|
the *calling* thread — no worker is spawned — for both entry points, and (3)
|
|
the shared dispatch closes the per-request client.
|
|
"""
|
|
|
|
import threading
|
|
from types import SimpleNamespace
|
|
from unittest.mock import MagicMock
|
|
|
|
from agent.chat_completion_helpers import (
|
|
direct_api_call,
|
|
interruptible_api_call,
|
|
interruptible_streaming_api_call,
|
|
should_use_direct_api_call,
|
|
)
|
|
|
|
|
|
def _make_agent(*, platform="cron"):
|
|
agent = MagicMock()
|
|
agent.platform = platform
|
|
agent.api_mode = "chat_completions"
|
|
agent.provider = "openrouter"
|
|
agent._interrupt_requested = False
|
|
agent._consecutive_stale_streams = 0
|
|
agent._touch_activity = MagicMock()
|
|
agent._close_request_openai_client = MagicMock()
|
|
return agent
|
|
|
|
|
|
def test_should_use_direct_api_call_only_for_cron_platform():
|
|
assert should_use_direct_api_call(_make_agent(platform="cron")) is True
|
|
assert should_use_direct_api_call(_make_agent(platform="cli")) is False
|
|
assert should_use_direct_api_call(_make_agent(platform="telegram")) is False
|
|
assert should_use_direct_api_call(_make_agent(platform=None)) is False
|
|
|
|
|
|
def test_direct_api_call_runs_inline_and_closes_client():
|
|
agent = _make_agent()
|
|
caller_tid = threading.get_ident()
|
|
ran_on = {}
|
|
fake_client = MagicMock()
|
|
|
|
def _create(**_kwargs):
|
|
ran_on["tid"] = threading.get_ident()
|
|
return fake_client
|
|
|
|
fake_client.chat.completions.create.return_value = SimpleNamespace(id="resp")
|
|
agent._create_request_openai_client.side_effect = _create
|
|
|
|
resp = direct_api_call(agent, {"model": "m", "messages": []})
|
|
|
|
assert resp.id == "resp"
|
|
# Inline: the request ran on the calling thread, no worker was spawned.
|
|
assert ran_on["tid"] == caller_tid
|
|
assert agent._close_request_openai_client.call_count == 1
|
|
|
|
|
|
def test_interruptible_api_call_routes_cron_inline_no_worker_thread():
|
|
agent = _make_agent()
|
|
caller_tid = threading.get_ident()
|
|
fake_client = MagicMock()
|
|
ran_on = {}
|
|
|
|
def _create(**_kwargs):
|
|
ran_on["tid"] = threading.get_ident()
|
|
return fake_client
|
|
|
|
fake_client.chat.completions.create.return_value = SimpleNamespace(id="first")
|
|
agent._create_request_openai_client.side_effect = _create
|
|
|
|
resp = interruptible_api_call(agent, {"model": "m", "messages": []})
|
|
|
|
assert resp.id == "first"
|
|
assert ran_on["tid"] == caller_tid # no daemon worker thread
|
|
|
|
|
|
def test_interruptible_streaming_api_call_routes_cron_via_nonstream_method():
|
|
"""Streaming is the default even for cron — the gate must catch it too.
|
|
|
|
It delegates to the ``_interruptible_api_call`` method (which itself runs
|
|
inline for cron) rather than calling ``direct_api_call`` directly, so the
|
|
outer loop's per-request retry/refresh seam — which patches that method —
|
|
stays intact (regression from the codex 401-refresh path).
|
|
"""
|
|
agent = _make_agent()
|
|
sentinel = SimpleNamespace(id="via-nonstream")
|
|
agent._interruptible_api_call = MagicMock(return_value=sentinel)
|
|
|
|
resp = interruptible_streaming_api_call(
|
|
agent, {"model": "m", "messages": []}, on_first_delta=lambda: None
|
|
)
|
|
|
|
assert resp is sentinel
|
|
agent._interruptible_api_call.assert_called_once()
|