PDU fault in rack AMS-B06
Alimentación y refrigeración — AMS · Gestión fuera de banda — AMS
Cronología
-
Resuelto
Every machine in rack B06 has been up since 02:52 UTC. Duration: 16 minutes. The PDU is being replaced; the 8 machines that lost power receive the SLA extension automatically. Data on local disks is intact; the machines simply rebooted.
-
Causa identificada
The A-side PDU in rack B06 failed with a breaker fault. The 8 machines that rebooted are the ones whose B-side supply was not seated correctly after a recent hardware change. All are powered from the B-side now and are booting.
-
Investigando
8 machines in rack B06 of AMS lost power at 02:36 UTC. Machines in the rack are dual-fed, but these 8 rebooted when one PDU failed. Engineers are at the rack.
Revisión posterior al incidente ·
Qué ocurrió y qué cambió
Causa
The A-side PDU in AMS-B06 suffered a breaker fault. Every machine in the rack has two power supplies on separate feeds, but on 8 of them the B-side cord had been left partially seated during a hardware change the previous week, so they lost power when the A side went.
Impacto
8 machines rebooted and were unavailable for 16 minutes. No data was lost; the machines came back to the state they were installed in, with local disks intact.
Qué cambiamos
- Every hardware change now ends with a power-redundancy test: each feed is dropped in turn and the machine must stay up.
- The faulty PDU model is being phased out across the fleet.
- Term extensions were applied to the 8 machines the same day.
Las horas se muestran en su zona horaria ().
Páginas relacionadas
- Estado del servicioEstado en vivo de gpuserver.io: componentes y centros de datos sondeados cada 60 s desde 5 ciudades, con disponibilidad de 90 días e incidentes desde 2022.
- Historial de incidentesTodos los incidentes y ventanas de mantenimiento de gpuserver.io desde septiembre de 2022, mes a mes, con cronología y revisión posterior al incidente.
- Acuerdo de nivel de servicioEl compromiso del 99,9 %: qué cuenta como inactividad, cómo se mide desde fuera, cinco minutos de plazo por cada minuto perdido, y las exclusiones completas.
- RedPuertos sin contador hasta 25 Gbit/s, dos operadores y un IX por sede, filtrado DDoS permanente, IPv6 enrutado — y las cuatro cosas que no ofrecemos, dichas ya.