Distributed Coordination Failure

A fleet of workers pulls batches of work items through a Coordination Service that tracks ownership across the entire item population. The backing ownership table has started serving stale reads. Every work item must eventually be processed exactly once — a batch processed twice can duplicate a real-world side effect, and a batch nobody believes it owns simply never gets processed, growing a backlog that the business is accountable for clearing. Diagnose whether ownership conflicts, missed work, or stalled workers (or some mix of all three) are the dominant failure right now, and bring the ownership table back to a state where every item has exactly one confirmed owner.

Difficulty

Staff

Expected Time

~18 min

Category

Distributed Systems