How Distributed Leases Work — and Why They Can Still Cause Split-Brain Issues

Distributed systems use leases — time-limited ownership grants — to prevent abandoned locks from blocking resources indefinitely when an application crashes or becomes unresponsive. Unlike permanent locks, leases expire automatically, allowing another node to take over without manual intervention. However, leases introduce a subtle risk: if a process is paused, such as during a JVM garbage collection cycle longer than the lease duration, it may resume work believing it still holds ownership while another node has already acquired a new valid lease. This results in two applications simultaneously assuming exclusive control over the same resource, a condition known as split-brain. Crucially, split-brain can occur without any network partition, even when the coordination service is functioning correctly.