Happy SysAdmin Day — the last Friday in July. If you work in operations, July 2026 probably didn’t feel like a holiday.
The Fortnight
Here’s a non-exhaustive list of what went wrong in the second half of the month:
- 14 July: a Google Cloud network configuration change broke connectivity for Google Cloud VMware Engine stretched clusters in Sydney, Melbourne and Frankfurt for more than ten hours. That’s the feature you buy for resilience.
- 16 July: AWS CloudFront’s VPC Origins feature failed globally, taking sites like Canvas and Hugging Face with it.
- 19 July: the second anniversary of CrowdStrike’s Blue Friday. Not an outage, but it felt like the right moment to look back.
- 23 July: a maintenance automation bug removed more routes than intended and cut Azure West US off from the outside world for nearly five hours.
- 24 July: AWS us-west-2 lost connectivity on the network path to the Seattle metro area, with a long recovery tail for the services downstream.
None of these was as large as CrowdStrike or last October’s us-east-1 outage, but together they show how often things fail in a world that depends on a small number of very large platforms.
Two Years On
Have we learned the lessons of 2024? Partly. Staged rollouts are more common, and the big providers now publish better post-incident reports. Regulations like DORA have made boards ask harder questions. But the fundamentals I keep returning to — configuration as code, blast-radius limits, rehearsed recovery and honest communication — are still not as widely practised as they should be.
Thank You, Again
In 2020 I wrote Silent Running, about the invisible work that keeps everything running. Six years later, it’s still invisible when it goes well and very visible when it doesn’t.
To the people who picked up the pager this month, held the bridge together, wrote the comms, did the rollback and stayed on to write the review: thank you. Take the cake. You earned it. ?
More Information
