One of the storage pods in metadata service got overloaded, and both reads and writes started queuing causing transactions to fail with timeouts