Google has added health checking and automatic failover for multi-region services on Cloud Run. The update is available in all Cloud Run regions at no extra charge beyond the CPU and memory used for readiness probes.
The changes are intended to reduce downtime for applications running in more than one region. Cloud Run users can already deploy the same service configuration across multiple regions and, when those services sit behind a global external application load balancer, direct traffic from different locations.
The new features are designed to detect service problems in a region and shift traffic to a healthy region within seconds. They include readiness probes, which provide instance-level health checks for Cloud Run services and determine when containers are ready to receive traffic.
The probes can also monitor the number of healthy and unhealthy instances for a service in each region. Google has also introduced service health, which aggregates instance-level readiness probe checks to calculate a service's health in each region.
That aggregate health is exposed through serverless network endpoint groups in each region. When a service is connected to a global application load balancer, traffic can automatically move away from regions where services are unhealthy.
Service health can be used with both single-region and multi-region services, though the announcement focuses on improving resilience in multi-region deployments. The approach is meant to reduce the manual intervention needed when a regional outage or disruption affects application availability.
Public and private
Google drew a distinction between internet-facing and private applications. For public websites and APIs, it recommends using Cloud Run with a global external application load balancer to enable automated detection and failover.
For applications that handle only internal traffic inside a private network, customers should use Cloud Run with a cross-regional internal application load balancer. In both cases, the goal is to shift traffic automatically when a regional service issue is detected.
The update reflects broader demand for simpler ways to build resilient cloud systems without managing complex failover logic. Multi-region deployments are often used by organisations that want to avoid a single regional outage taking down customer-facing services, but they can be difficult to configure and monitor.
Google said the service health feature is best suited to active-active setups, where two or more regions are already configured to serve traffic. In that architecture, fast detection and recovery depend not only on application health checks but also on whether the rest of the stack is designed to survive a regional failure.
Design choices
Google also highlighted the risk of single points of failure elsewhere in an application. Customers should ensure every layer, including databases, has regional redundancy if they want a multi-region deployment to keep working during an outage.
For three-tier applications running on Cloud Run, users should consider separate multi-region architectures for the web tier and the application tier. That reflects the different networking demands of public internet traffic and private internal traffic.
Data replication is another central consideration. Organisations need to assess their recovery point objective and decide whether they require zero data loss when synchronising information across regions, particularly for applications with heavy read and write activity.
Google said the new service health model works best with applications that actively synchronise data across regions. It also pointed to several managed database and storage products that can be used in multi-region setups with Cloud Run, including Firestore, Spanner, Cloud Storage, and Cloud SQL.
That matters for customers with strict data sovereignty requirements, because data location and replication policies can shape how a multi-region architecture is built. In practice, application-layer resilience solves only part of the problem if data stores remain tied to a single region or have inconsistent replication behaviour.
The enhanced multi-region high-availability features are available now in all Cloud Run regions. Customers pay only for the standard CPU and memory consumed by readiness probes.