GitHub suffered two major incidents in late April. On April 23rd, a bug in the merge queue's squash merge caused incorrect rollbacks of previously merged commits when a merge group contained multiple PRs, affecting 658 repositories and 2,092 PRs. Git data wasn't lost, but default branch states were wrong. On April 27th, an Elasticsearch subsystem overload (possibly from a botnet attack) took down search, leaving PR, issue, and Projects interfaces showing empty results.
Right after the April 27th outage, Ghostty founder Mitchell Hashimoto announced his project would leave GitHub. Hashimoto, GitHub user #1299 for over 18 years and creator of Vagrant and Terraform, said in a blog footnote that the migration had been discussed for months and the post was written a week before — the timing was coincidental. But the real reason: "For the past month, almost every day my work has been affected by GitHub outages. I can't do PR reviews properly. This place is no longer suitable for serious work." He's now talking to multiple code hosting services; existing repo URLs will remain as read-only mirrors.
https://twitter.com/mitchellh/status/2049213597419774026
The same day, GitHub CTO Vlad Fedorov apologized for both incidents, blaming an explosion in agent development workflows. Since mid-December 2025, repo creation, PR activity, and API calls have surged dramatically: peak monthly merged PRs hit 90 million, monthly commits 1.4 billion, and monthly new repos 20 million. A single PR can touch over a dozen subsystems (Git storage, merge checks, branch protection, Actions, search, notifications, permissions, webhooks, etc.), so small efficiency issues cascade under high concurrency. GitHub started a 10x scaling plan last October, upgraded to a 30x design in February, and is now working on isolating critical services, migrating performance-sensitive code from Ruby monolith to Go, and pursuing multi-cloud deployment.