Updated 23 October 2026. On 20 October, an electrical failure at a facility supporting Google Cloud's London zone europe-west2-b interrupted cloud services and affected websites and applications relying on the region. Google's preliminary incident report puts the main disruption at 07:00–13:10 BST (6 hours, 10 minutes). Banks and other UK services reported customer-facing problems. This wasn't a cyberattack, and the complete final incident review is still pending.
What actually went wrong in London?
- At approximately 06:53 BST, a power interruption affected the datacentre hosting europe-west2-b.
- UPS batteries temporarily sustained the facility, but both primary and backup programmable logic controllers failed to start or transfer load to emergency generators. As battery reserves ran down, servers, network equipment and local storage lost power from about 07:00 BST.
- Automatic failover shifted traffic and workloads towards other zones. Some regional control-plane operations still depended on unreachable components in the offline zone, causing timeouts.
- Sudden extra demand exhausted or contended for capacity in surviving zones. Some users saw HTTP 500/503 errors, delayed jobs, database failover delays and latency even outside the failed zone.
- Google reported recovery for the vast majority of workloads by 13:10 BST and posted a closure update at 15:03 BST. Its report published on 22 October is explicitly preliminary.
This is a failure of multiple layers, not simply 'a server went offline'. Google named services including Compute Engine, GKE, Cloud SQL, Cloud Run and BigQuery among those affected. Reports from Reuters and other outlets linked the wider incident with disruption to UK banks and online services. Do not infer that every service using Google Cloud went down.
Why did a multi-zone cloud region still struggle?
Multi-zone deployment helps when a single zone fails, but it is not a guarantee of uninterrupted service. The 20 October incident exposed shared regional dependencies: control-plane requests to an unavailable zone, sudden demand on remaining capacity and storage or database failovers. A working backup location is useful only if it has the capacity, data and independent dependencies needed to serve your users.
Seven changes that reduce the impact of the next outage
- Define your recovery targets. Set an RTO (acceptable recovery time) and RPO (acceptable data loss) for each critical feature. A brochure site, login service and payment system need different plans.
- Map the whole dependency chain. Inventory hosting zones, databases, queues, object storage, authentication, DNS, CDN, payment providers and email. Look for shared single points of failure.
- Distribute critical services across independent zones. Use regional or multi-zone compute and highly available data services where supported; do not leave all replicas on one zonal disk or network dependency.
- Add multi-region recovery for genuinely critical journeys. Keep a warm or active secondary region, choose a suitable database replication and write strategy, and verify any data residency or compliance requirements.
- Make traffic failover real. Configure health-checked load balancing or a tested DNS/traffic-routing strategy. Probe meaningful application readiness, not only whether a server responds to TCP connections.
- Plan for failover surges. Pre-provision or reserve enough secondary capacity; bound retries with exponential backoff and jitter, use circuit breakers and load shedding, and keep queued operations idempotent.
- Test the plan. Simulate loss of a zone and, where required, a region. Verify login, payments, database consistency, background jobs, notifications, rollback and the real time to recover.
A simple resilience design
Visitors
|
Health-checked traffic routing
| |
Primary region Secondary region
App in 2+ zones Warm/active app
HA database Replicated/restorable data
| |
+-- Monitors test real user journeys --+
If primary is unhealthy: route to a tested healthy destination.
If data is not safe for writes: degrade gracefully, not silently.A high-availability database in one region does not automatically provide safe cross-region writes. Decide in advance whether your fallback is read-only, uses controlled promotion of a replica, or supports active-active writes. Test for split-brain and duplicate processing. Google Cloud's architecture documentation recommends distributing resources across zones and, for region-level resilience, across regions; its 2026 Cloud Run guidance also describes health-based failover for suitable configurations.
What a smaller website can do this week
- Check your host's region, zone placement and backup policy; confirm where the backups physically reside and test a restore.
- Write down who can change DNS or hosting settings during an incident, plus the steps to put up a lightweight status or maintenance page.
- Monitor your homepage and one critical transaction from locations outside your hosting provider. Test alerts and distinguish genuine user-impacting errors from noisy probes.
- Reduce brittle dependencies: cache safe public content, add sensible timeouts and bounded retries, and make nonessential features fail gracefully.
- Decide what to do if payments, login or your database are unavailable. Never treat a payment request as successful unless confirmed by the payment provider.
How to know your recovery plan works
- Check external availability and error rates before, during and after a controlled exercise.
- Trigger a staging failover using supported mechanisms. Record time to detect, time to reroute and time to restore critical functionality.
- Verify replica lag, outstanding writes and queues before reopening traffic. Audit transactions for duplicates or missing records.
- Document the result, owners and gaps. Repeat after infrastructure, database or routing changes.
Where Zenith helps — and where it doesn't
Zenith can help reveal externally visible symptoms such as a homepage becoming unavailable, server errors and slower responses. Those observations can support detection and post-incident review. They cannot prove the root cause, test internal database recovery, guarantee that failover is configured or prevent an upstream outage. Pair outside-in monitoring with infrastructure metrics, provider alerts and rehearsed recovery procedures.
Sources and further reading
- Google Cloud Service Health — preliminary incident report, updated 22 October 2026: https://status.cloud.google.com/incidents/Xs2XTHzAVC9YnafnPyzj
- Google Cloud Architecture Center — design reliable infrastructure: https://docs.cloud.google.com/architecture/infra-reliability-guide/design
- Google Cloud — Cloud Run multi-region resilience updates, July 2026: https://cloud.google.com/blog/products/serverless/cloud-run-multi-region-services-enhanced-for-high-availability
- Google Cloud — disaster recovery architecture: https://docs.cloud.google.com/architecture/disaster-recovery
- Reuters — Google Cloud disruption of UK services, 20 October 2026: https://www.reuters.com/world/europe/google-cloud-services-experience-outage-europe-2026-10-20/
Glossary
- Availability zone
- An independently provisioned location within a cloud region; a zone can fail while other zones remain available, but some services are still shared regionally.
- RTO
- Recovery Time Objective: the maximum targeted duration before a service is restored after an incident.
- RPO
- Recovery Point Objective: the maximum targeted age of recoverable data, often described as acceptable data loss.
- Failover
- Switching traffic or workloads to healthy backup infrastructure when a primary component becomes unavailable.
- Control plane
- The systems and APIs used to configure, schedule and manage cloud resources, distinct from the application data path.
- Graceful degradation
- Keeping safe, useful portions of a service working when a dependent component is unavailable.