The Google Cloud outage in Paris on 25–26 April 2023 was caused by a cooling-system water leak that reached a UPS room, triggered a fire response and forced a data-center building to shut down. The main disruption was in Google Cloud’s europe-west9 region, not across every Google Cloud region. The outage lasted into a second day as services recovered in stages.
How the fire and flood disrupted Google Cloud
Google’s incident report says a cooling-system water pipe leaked in a part of a data-center facility not operated by Google. Water entered an associated uninterruptible power supply (UPS) room and led to a fire. The facility was evacuated, local firefighters responded and power to the entire building was shut off for several hours.
Google said europe-west9 had three buildings with independent cooling, power and networking. However, Regional Spanner replicas—which supported the regional control plane for most Google Cloud services in the region—were not correctly distributed across those buildings to preserve quorum when the affected building went offline. With quorum lost, control-plane services were interrupted, contributing to unavailability or elevated error rates across many services.
The incident report recorded a peak 100% drop in traffic flow for resources hosted in the affected area. That figure describes the affected resources at the incident’s peak; it is not a platform-wide Google Cloud error rate or a general reliability measure.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
Outage timeline: why recovery took into a second day
Google’s incident record lists the outage start as 16:46 US/Pacific on 25 April 2023 and its end as 20:00 US/Pacific on 26 April. Recovery was staggered rather than simultaneous:
| Milestone | Time, US/Pacific |
|---|---|
| Battery-room fire extinguished | 04:11, 26 April |
| Regional Spanner quorum restored | 12:47, 26 April |
| Regional IAM restored | 14:42, 26 April |
| Google Cloud Storage regional services restored | 15:14, 26 April |
| Pub/Sub restored | 15:30, 26 April |
| BigQuery restored | 17:10, 26 April |
| Dataflow restored | 18:40, 26 April |
| Compute Engine control-plane services across all zones restored | 19:00, 26 April |
| Incident record lists outage end | 20:00, 26 April |
Google described the region as primarily recovered and operational after the 19:00 Compute Engine milestone. The separate 20:00 entry is the incident record’s listed end time.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
Which services and customers were affected?
Google listed Compute Engine, Kubernetes Engine, Persistent Disk, Cloud SQL, AlloyDB, Cloud Spanner, Bigtable, Filestore, Apigee, Artifact Registry, Cloud Run and related services, BigQuery, Dataflow, Dataproc, Cloud Storage, Pub/Sub, IAM and Cloud Console among the services affected during some phase. IT Pro reported that more than 100 services were affected in the initial outage; that is the publication’s figure, distinct from Google’s service-by-service account.
The principal disruption was regional, in europe-west9. Google also reported temporary problems with Cloud Console globally and failures in some global operations, but this does not mean all Google Cloud regions were down.
Google said 79% of Regional Persistent Disk volumes in europe-west9 had one replica in the impacted building. It also said no such volume had both replicas in that building. The 79% figure describes replica placement, not data loss.
Was customer data lost?
No. Google’s incident report says: “Customers experienced no data loss from the incident.” The affected cluster in europe-west9-a sustained water intrusion and smoke damage. Google said it decommissioned the cluster only after cleaning it and recovering all zonal data.
Rank #4
What Google said it would change
Google’s incident report described physical-risk and service-resilience follow-ups. It said it planned detailed water-intrusion risk analysis across data centers, evaluation of drip trays in relevant multi-story facilities, and refinement of its fire-and-flood recovery playbook.
For service resilience, the report listed improved zonal and regional failure testing, validation of resource placement across physical failure domains, better graceful degradation for APIs that require all regions and zones, and keeping Cloud Console operational when a control-plane scope is unavailable. These are actions or plans described in the incident report; it does not establish their later completion.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
How Google Cloud customers can check for incidents now
For project-specific disruptions, Google identifies Personalized Service Health as the primary source and recommends incorporating it into incident response. It supports project-aware dashboard views, configurable alerts, API access to active disruptive events and Cloud Logging integration, according to Google’s configuration documentation.
The public Google Cloud Service Health dashboard provides a broad, no-login view for severe incidents and can serve as a fallback. Google’s incident communication guidance also points to manual checks of the public dashboard if Personalized Service Health is unavailable, and to a public RSS feed for automated fallback.
When checked on 4 October 2026, the public dashboard said “No broad severe incidents” and showed an update time of 4 October 2026, 13:57 PDT. That is a point-in-time public status, not confirmation that every customer project was unaffected; use Personalized Service Health for events relevant to a particular project.
Practical resilience lessons for cloud workloads
Google’s incident report described failover to other regions as a workaround while europe-west9 was affected. It also recommended zonal Internal DNS namespaces in the context of reducing a class of global-scope management impact. These incident-specific recommendations are not a complete disaster-recovery design for every workload.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteQuick Recap
- Check whether critical data and service replicas are placed across independent physical failure domains, not merely across logical zones.
- Plan and test how workloads behave when a regional control plane or management interface is unavailable, not only when an individual compute zone fails.
- Decide in advance whether and how critical workloads can fail over to another region, including dependencies such as identity, DNS and data replication.
- Use project-relevant incident notifications and test the fallback channel your team would rely on if its primary status service were unavailable.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
