馃煚 LOW SEV (5 servers) 路 Inference Consumer Unit Inactive
馃敟 ozone-inference-consumer systemd unit is not active
<|馃搱 graph> 路 <|馃搳 service dashboard> 路
馃毃 alarm definition 路
馃敃 silence
馃敟 ozone-inference-consumer systemd unit is not active
<|馃搱 graph> 路 <|馃搳 service dashboard> 路
馃毃 alarm definition 路
馃敃 silence
馃敟 ozone-inference-consumer systemd unit is not active
<|馃搱 graph> 路 <|馃搳 service dashboard> 路
馃毃 alarm definition 路
馃敃 silence
馃敟 ozone-inference-consumer systemd unit is not active
<|馃搱 graph> 路 <|馃搳 service dashboard> 路
馃毃 alarm definition 路
馃敃 silence
馃敟 ozone-inference-consumer systemd unit is not active
<|馃搱 graph> 路 <|馃搳 service dashboard> 路
馃毃 alarm definition 路
馃敃 silence
LOW SEV - cause-detector. The process is down (crashed, failed to restart, or the host is gone) as opposed to a wedged metrics endpoint. Normally the consumer group rebalances onto the surviving box and screen time backfills from the log, so the customer sees a gap of seconds-to-minutes in enforcement and nothing permanent. Escalates to CRITICAL via 'Inference Consumer Unit Inactive (sustained)' at 30m, because a box death riding unnoticed into the morning DOES poison the parent digest. Prod runs ozone-inference-consumer@1.service on 2 boxes. Check: journalctl -u ozone-inference-consumer@1 -n 200.
Grafana vmain-b1e6358