Google Cloud's London zone went down 9 hours — GKE, Run, BigQuery
A 'power event' in europe-west2-b knocked ~38 services offline most of the business day; it's mitigated now — a blunt lesson in single-zone risk.
Copy markdown
A single zone, europe-west2-b, took the region down
Google blames a “power event” in one of London's three availability zones. Unplanned downtime began 06:08 UTC and Google marked it mitigated around 15:05 UTC — roughly nine hours. Existing workloads came back first; the ability to create new workloads lagged well behind.
What broke: GKE, Cloud Run, BigQuery and ~36 more
Confirmed casualties include Google Kubernetes Engine, Cloud Run, BigQuery (and its Data Transfer Service), App Engine, Cloud SQL, Looker, Filestore and Managed Service for Kafka — one estimate put it near 38 services. Google didn't name Vertex AI or the Gemini API, but if your inference, agents or eval pipelines live in europe-west2, assume elevated errors for the day.
The fix Google itself recommended: leave the zone
Mid-incident Google told customers to shift workloads to another zone or region if feasible — the tell that this was zone-scoped, not region-wide. If one AZ going dark takes your product down, that's your architecture, not Google's: spread serving across europe-west2-a/b/c or a second region before the next power event, not after.
A small number still affected; a report is coming
Google says a few customers may still see impact and to open a support case if you're one of them. A full incident write-up is promised but not yet out, so the live status incident is the canonical source — bookmark it if you run anything in London.