mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-11-06 05:17:15 +00:00
Teknium
c41de4d95f
Review findings on #80873: - _timeout_phase walked __context__ too, so an HTTP error raised inside an `except ReadTimeout:` block (e.g. a 401) was reported as a timeout. Walk only the explicit __cause__ chain (the SDK's `raise APITimeoutError from err`). - PoolTimeout / WriteTimeout mean the provider never got to answer; they now keep the existing summary instead of the "waiting for the provider" hint. - The read-timeout hint prints providers.<agent.provider>.request_timeout_seconds when called on an agent; class-level calls keep the <provider> placeholder.
64 lines
2.9 KiB
Python
64 lines
2.9 KiB
Python
"""Phase-specific one-liners for transport-timeout API errors.
|
|
|
|
A timeout that survives the retry loop otherwise surfaces as the SDK's bare
|
|
text: ``Request timed out.`` (openai ``APITimeoutError``) or ``timed out`` (a
|
|
raw httpx ``ReadTimeout`` mid-stream). Neither says whether a connection was
|
|
ever made, nor which setting to raise. The openai SDK wraps EVERY httpx
|
|
timeout, connect included, in ``APITimeoutError``, so the phase is read from
|
|
the most specific timeout in the explicit ``__cause__`` chain. ``__context__``
|
|
is not followed: an HTTP error raised while a timeout was being handled is a
|
|
different failure and keeps its own summary.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from typing import Optional
|
|
|
|
# No TCP/TLS connection was established: raising the read timeout will not help.
|
|
_CONNECT_TIMEOUT_TYPES = frozenset({"ConnectTimeout", "ConnectTimeoutError"})
|
|
# The request was sent (or a stream was open) but no bytes arrived in time.
|
|
# ``APITimeoutError`` without a transport cause also lands here.
|
|
_READ_TIMEOUT_TYPES = frozenset({
|
|
"ReadTimeout", "TimeoutException", "ReadTimeoutError", "APITimeoutError",
|
|
})
|
|
# Waiting for a pooled connection or while uploading the request: the provider
|
|
# never got to answer, so neither hint applies and the existing summary stays.
|
|
_OTHER_TIMEOUT_TYPES = frozenset({"PoolTimeout", "WriteTimeout"})
|
|
|
|
|
|
def _timeout_phase(error: BaseException) -> Optional[str]:
|
|
"""``"connect"``, ``"read"``, or ``None`` when no timeout is in the chain."""
|
|
phase = None
|
|
current: Optional[BaseException] = error
|
|
seen: set[int] = set()
|
|
while current is not None and id(current) not in seen:
|
|
seen.add(id(current))
|
|
name = type(current).__name__
|
|
if name in _CONNECT_TIMEOUT_TYPES:
|
|
return "connect"
|
|
if name in _OTHER_TIMEOUT_TYPES:
|
|
return None
|
|
if name not in _READ_TIMEOUT_TYPES:
|
|
break
|
|
phase = "read"
|
|
current = current.__cause__
|
|
return phase
|
|
|
|
|
|
def summarize_timeout_error(error: BaseException, provider: Optional[str] = None) -> Optional[str]:
|
|
"""Timeout summary naming the phase and the setting to change, or ``None``
|
|
for any other error (the caller's existing summary handling then applies).
|
|
|
|
Both lines keep the "timed out" wording: cron failure notices and the error
|
|
classifier re-read this summary as text and match on it.
|
|
"""
|
|
phase = _timeout_phase(error)
|
|
if phase == "connect":
|
|
return ("Connection to the provider timed out before it was established "
|
|
"(check the base_url and network path)")
|
|
if phase == "read":
|
|
key = provider.strip() if isinstance(provider, str) and provider.strip() else "<provider>"
|
|
return ("Request timed out waiting for the provider to respond (if the model needs "
|
|
f"longer, raise providers.{key}.request_timeout_seconds in config.yaml)")
|
|
return None
|