Repair Distributed Lock Fencing Bugs

The problem

Distributed locking bugs often lead to silent data corruption, split-brain scenarios, or leaked leases that stall system progress. Manually tracing fencing tokens and stale-owner write races across distributed repository logic is error-prone and difficult to reproduce.

What it does

• Identifies race conditions in lease acquisition, renewal, and release cycles.
• Implements monotonic fencing tokens to reject stale-owner writes at the storage layer.
• Hardens trust boundaries between owner identity, generation binding, and write guards.
• Refactors validation logic to ensure rejection occurs before irreversible state mutations.
• Generates deterministic local fixtures and regression tests using time-shifts and network barriers.

Frameworks & tools

Works with repository-level code involving distributed locks (Redis/Redlock, Etcd, Zookeeper) and storage authorities (PostgreSQL, DynamoDB, S3) using Read, Edit, and Bash tools.

Why this beats prompting it yourself

This skill follows a formal method to map every concurrent timeline and state transition before touching code. It prevents common "quick fixes" like adding sleeps or widening retry policies, ensuring the fix addresses the underlying clock or partition logic defect.

Use cases

• Fixing split-brain issues where two nodes believe they hold the same leader lease.
• Preventing "zombie" processes from writing to a database after their lease has expired.
• Hardening monotonic counter increments used for fencing in distributed environments.
• Adding observability to track lease TTL drift and fencing rejection rates.