GitHub blames the 2026-08-24 Actions start delays on a node that failed its disk but kept reporting healthy

Sources: incident, status API
Summary: GitHub attributes the 2026-08-24 Actions job start delays to a node whose disk failed while its health checks continued to pass, so the usual eviction path did not remove it. GitHub states the remediation is application-level removal of stalled consumers.
Why it matters: A node that loses one subsystem while still passing health checks is invisible to the usual eviction path, and GitHub's stated fix is application-level removal of stalled consumers rather than faster node replacement.