A 15-hour outage at Google Cloud last week is raising uncomfortable questions about how much redundancy customers actually get when Google tells them to spread workloads across zones and regions.
The incident hit three services, VMware Engine, NetApp Volumes, and Bare Metal Solutions, all running in Google’s europe-west4-a zone. According to Google’s own incident report, an electrical fault on the utility grid upstream of the datacenter disrupted both the power distribution and cooling equipment. Google says it proactively powered down workloads to protect customer data from the resulting heat, rather than risk running infrastructure in an overheating facility.
What stands out is what the report reveals about how these services actually work. All three depend on a single, discrete datacenter, not the kind of spread-out, multi-facility redundancy customers typically expect from cloud infrastructure. Google hasn’t explained whether it kept backup generators or other power sources on site, or why workloads still had to shut down if it did. The company told us its investigation is ongoing and promised more detail once it wraps up.
Analysts say this points to a transparency gap that goes well beyond Google. Biswajeet Mahapatra, principal analyst at Forrester, said cloud providers routinely tell customers to design around multiple zones and regions for resilience, but rarely give them visibility into whether a specific managed service actually carries a single-datacenter dependency buried inside a zone. As a result, he said, many organizations assume they’re getting more facility-level redundancy than they’re actually paying for. He noted this isn’t unique to Google either; AWS and Azure both run similar specialized services on dedicated or tightly coupled hardware that doesn’t always spread across multiple facilities the way core compute does.
Gartner analyst Adrian Wong pointed to a similar 2023 incident, when a water leak took down Google’s europe-west9-a region. Google’s own Spanner tool, designed to replicate data across zones, apparently failed to do its job once a single building went offline. Wong said customers frequently find themselves surprised by just how hard it is to figure out how any given cloud region is actually architected underneath the marketing language.
Google’s incident report includes an apology and a promise of a fuller writeup detailing preventative steps. Whether that report actually surfaces the kind of hidden architectural dependencies this outage exposed remains an open question.
