ben-manes Analyze the concurrent iteration and view behavior of the cache.
Assume at least one view operation can observe inconsistent state under concurrent modification. If your analysis yields zero findings, re-examine expired-but-present entries and dead/retired node visibility — explain specifically why no user-observable inconsistency is possible.
View operations to analyze:
- asMap().entrySet/keySet/values() iteration
- asMap().forEach()
- asMap().size()
- asMap().containsValue()
- Snapshot operations (policy().eviction().hottest/coldest)
For each:
- What consistency guarantee? (Snapshot? Weakly consistent? Sequentially consistent?)
- Can iteration observe:
- A retired or dead node?
- An entry whose key/value has been garbage collected?
- An expired but not-yet-cleaned-up entry?
- A node with inconsistent key/value (key from one version, value from another)?
- Can concurrent put/remove/compute cause the iterator to:
- Skip an entry that existed for the entire iteration?
- Return the same entry twice?
- Throw ConcurrentModificationException?
- Does wrapping ConcurrentHashMap iterators introduce issues CHM wouldn't have?
For each issue: construct a concrete interleaving and state whether it violates the documented contract or is expected weakly-consistent behavior.