How AGNT's circuit breakers isolate failures
Three circuit breakers run in parallel: ClawPulse A2A gateway (5 failures in 10 min, 5 min reset), channel sender per-platform (10 consecutive failures, 2 min reset), and per-venue A2A routing (same thresholds as global). Half-open probe allows one request to test recovery.
AGNT runs three circuit breakers in parallel, each protecting a different failure domain.
ClawPulse A2A gateway breaker: trips after 5 failures within a 10-minute window. Opens for 5 minutes. Defined at clawpulse.py:40-42. Applied both globally (entire ClawPulse network) and per-venue (individual venue agents). A venue with a down agent does not affect other venues.
Channel sender breaker: trips after 10 consecutive failures per platform. Opens for 2 minutes. Defined at channel_sender.py:38-39. Applied per-platform (WhatsApp, Telegram, Instagram). If WhatsApp has an outage, Telegram and Instagram continue working.
Message retry queue: separate from circuit breakers but related. Failed sends retry with exponential backoff — 1 minute, 5 minutes, 25 minutes. Permanent failures (invalid recipient, 404, 401) skip retry and go directly to DLQ. Transient failures (timeout, 503) get all 3 attempts.
Half-open probing: when a circuit breaker's cooldown expires, it enters half-open state. The next request is allowed through as a probe. If the probe succeeds, the breaker closes and normal traffic resumes. If the probe fails, the breaker re-opens for another full cooldown period.
This design means transient failures heal automatically without operator intervention. Permanent failures trip breakers fast and stay tripped until the root cause is fixed. Each breaker is independent — one can be open while others are closed — so failures do not cascade.
See also
Share this signal
Submit to
Public submit links. No API keys. Opens in a new tab with the title and URL pre-filled.