Skip to content

pusher-degraded.md

Latest commit

 

History

History
16 lines (10 loc) · 1023 Bytes

pusher-degraded.md

File metadata and controls

16 lines (10 loc) · 1023 Bytes
Raw

Pusher — degraded session ratio high

What it means: More than 5% of active listener WebSocket sessions are in degraded mode because the listener cannot use Pusher and routes audio elsewhere. This is a likely user-impact signal, not proof that every listener lost data.

PromQL: sum(pusher_sessions_degraded{job="backend-listen-metrics"}) / clamp_min(sum(backend_listen_active_ws_connections{job="backend-listen-metrics"}), 1)

The degraded-session gauge is emitted by the backend-listen reconnect loop. Do not query the similarly named Pusher-process connection metric: it does not own this outcome.

Owner: listen/pusher team.

First checks:

  1. Pusher pod health and circuit breaker state (pusher_circuit_breaker_state, rejections).
  2. Recent pusher deploys or upstream STT/diarizer outages.
  3. Grafana pusher dashboard — error rates and active connection trends.

Note: Degraded sessions are a silent heal path; sustained elevation means pusher is unhealthy for a large share of listeners.