mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-07-20 15:33:54 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 7

fix(codex): raise hard-ceiling default above the max stale floor

Browse source 
With the TTFB watchdog now scaled (not disabled) for large requests on
main, the hard ceiling is a backstop against mid-stream wedges, not the
primary stall detector. The original 600s default clamped BELOW the
intentional 1200s stale floor for >100k-token requests, partially
reverting the floor that keeps healthy gateway-scale payloads alive.
Raise the default to 1500s so the ceiling only catches unbounded
growth, never healthy slow turns.
This commit is contained in:
Teknium 2026-07-15 04:28:12 -07:00
parent bcd7e2ce89
commit 7af59f474d
1 changed files with 12 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

23
agent/chat_completion_helpers.py
View file

@ -505,20 +505,21 @@ def interruptible_api_call(agent, api_kwargs: dict):
_stale_timeout = max(_stale_timeout, _codex_floor)
# ── Codex absolute hard ceiling (#64507) ──────────────────────────
# For large Codex requests the no-byte TTFB watchdog is intentionally
# disabled (so legitimate backend admission / prefill isn't killed), and
# ``openai_codex_stale_timeout_floor`` *raises* the stale timeout (up to
# 1200s at >100k tokens) so healthy gateway-scale payloads aren't aborted.
# That combination means a request that is genuinely stalled at the
# backend — no first byte AND no events, exactly the issue-64507 symptom —
# is only reclaimed at the (high) stale floor, so a session can hang for
# many minutes with an idle worker and no ended_at. Add a flat, finite
# hard ceiling on total request time that ALWAYS applies to openai-codex
# requests regardless of the TTFB/stale interaction, so a stalled large
# request is recovered (retry loop / visible failure) instead of hanging
# indefinitely. Tunable via HERMES_CODEX_HARD_TIMEOUT_SECONDS (set to 0 to
# The scaled no-byte TTFB watchdog catches dead streams that never emit a
# first byte, but a request that emits SOME bytes and then wedges (the
# issue-64507 symptom: vision-inflated request, worker idle, no ended_at)
# is only reclaimed at the (high) stale floor. Add a flat, finite hard
# ceiling on total request time that ALWAYS applies to openai-codex
# requests regardless of the TTFB/stale interaction, so a stalled request
# is recovered (retry loop / visible failure) instead of hanging
# indefinitely. The default sits ABOVE the maximum stale floor (1200s) so
# it never clamps an intentionally-raised timeout for healthy large
# requests — it is a backstop against unbounded growth, not a tighter
# limit. Tunable via HERMES_CODEX_HARD_TIMEOUT_SECONDS (set to 0 to
# disable the ceiling entirely; that restores the pre-fix behavior).
_codex_hard_timeout = _env_float("HERMES_CODEX_HARD_TIMEOUT_SECONDS", 600.0)
_codex_hard_timeout = _env_float("HERMES_CODEX_HARD_TIMEOUT_SECONDS", 1500.0)
if (
_codex_watchdog_enabled
and _openai_codex_backend