GitHub publishes a root cause for the 2026-08-21 SSH outage: a load-balancer defect that only fired through multiple layers, so canary testing missed it

Sources: GitHub incident page, GitHub incidents feed
Summary: GitHub states that successful Git operations over SSH fell by more than 95 percent at peak between 14:00 and 14:07 UTC on 2026-08-21, so clone, fetch, and push over SSH were effectively unavailable to most users for about four minutes. HTTPS was unaffected. GitHub attributes the failure to a defect in load-balancing infrastructure triggered by a configuration change, and states that the defect only manifested when connections crossed multiple layers of load balancers running the new configuration, which is why canary testing did not catch it. GitHub mitigated by rolling the configuration back, and committed to regression coverage for multi-layer load-balancer configurations and to better monitoring and alerting for Git over SSH.
Why it matters: Any team that canaries a configuration change on one tier of a multi-tier proxy path carries the same blind spot, because a defect that needs two tiers on the new configuration is invisible to a single-tier canary.