How did you handle deadlocks in a high-load system? What did you do when they were detected?

In high-load systems, deadlocks are situations where two or more processes wait for each other to release resources, leading to mutual blocking.

When deadlocks are detected, the following steps are usually taken:

1. Log analysis and monitoring: Using database monitoring tools and logs to identify locking points.
2. Identifying causes: Determining the transactions and resources involved in the deadlock.
3. Optimizing lock order: Changing the order of resource acquisition in the code to avoid cyclic dependencies.
4. Reducing lock duration: Breaking large transactions into smaller ones to shorten lock times.
5. Using timeouts: Setting timeouts to automatically interrupt transactions causing deadlocks.
6. Refactoring logic: Revising business logic to minimize resource contention.

Example: In PostgreSQL, pg_locks and pg_stat_activity can be used to diagnose deadlocks, and then change the order of operations in transactions to avoid mutual locks.