London Cloud Outage: Thousands of machines write the same error. The answer is in the ones that went silent.
On 19 October at 23:08 PDT, 07:08 in London, Google Cloud reported “an issue with multiple services in europe-west2”, its London region. According to the BBC, cited by DCD, Lloyds Bank, Revolut, Halifax, DPD and Royal Mail were affected. Google later said it had “suffered an outage within Europe-west2-b zone” “due to a power event”, and most services were back roughly six hours after the start.
07:08
London time, 20 October: first report
1 zone
europe-west2-b, “a power event”
~6 h
Until most services recovered
Sources: DCD, ITV News, Khaleej Times, 20 October 2026. Google’s full incident report may add detail.
The Hour After
I won’t guess beyond what Google has said. What interests me is the hour after. Thousands of machines will have written the same error. The answer usually sits in one or two lines that look nothing like the rest, or in the machines that went silent and wrote nothing at all. When a zone loses power, silence is often the first and loudest symptom.
At fleet scale, root-cause work is not about reading more logs. It is about collapsing identical noise into a single count, so the rare signal becomes visible. Silence is data too.
Built for that hour. Vitali de Mille built Glyph Fleet RCA for exactly this situation. It turns fleet logs into semantic glyphs, collapses repeated errors into counts, and surfaces the survivors and the silent nodes. Tested so far on synthetic fleets; the core is open on GitHub.
Operators: when a region fails, how long does it take you to find the first line that matters?