WEBDIENSTE/openswarm
1
0
Fork 0
mirror of https://github.com/openswarm-ai/openswarm.git synced 2026-09-28 20:44:50 +02:00
Code Issues Packages Projects Releases Wiki Activity

arena: bound 429 backoff to fit the step watchdog (flat 12s; unbounded waits booked as step_hang)

Browse Source 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WsbS5x2rYsMDxP2kW3qqmQ
This commit is contained in:
ciregenz
2026-08-14 22:23:30 -07:00
co-authored by Claude Fable 5
parent 61af236e2a
commit e6ecb83744
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
e2e/browser-v3/arena/llm_policy.py
+3 -1
View File
@@ -276,7 +276,9 @@ class LlmPolicy:
# Rolling TPM 429s reset in ~15-60s; a 2-6s backoff converts a slow lane into
# booked infra episodes (measured: 14 in one chore half-hour). Wait the window out.
is_rl = "429" in str(exc) or "rate limit" in str(exc).lower()
time.sleep((20.0 if is_rl else 2.0) * (attempt + 1))
# Bounded: act() runs under a ~150s step watchdog -- unbounded 429 waits turned
# every rate-limited step into a booked step_hang (measured: 10 in 30 min).
time.sleep(12.0 if is_rl else 2.0 * (attempt + 1))
d.think_ms = (time.time() - t0) * 1000
return str(text).strip(), d