A race condition occurs when concurrent operations produce different results depending on their timing or execution order. An operation may succeed when A finishes before B but fail when B finishes first. section 11.1.2
The error appears when a shared state is checked and changed through separate, non-atomic steps. Two cache updates can both observe available capacity, skip eviction, and insert entries, exceeding the configured limit or overwriting a value. section 7.7
Handle race conditions by coordinating access or by representing work already in progress:
- Mark a cache miss as loading before computation begins, then let later requests poll for the result or subscribe for notification. section 7.2.4
- Do not rely on an extra check alone: if the complete set operation remains non-atomic, another operation can still interleave before the update. section 7.7
Two requests add to a nearly full cache
A cache has maxSize−1 entries. Concurrent calls A and B both test the size before either inserts, so both conclude that no eviction is needed. Without synchronization, both add an entry. A synchronized cache serializes the write; a loading marker instead coordinates requests waiting for the same computation. section 7.7
