iprithv commented on code in PR #16212:
URL: https://github.com/apache/lucene/pull/16212#discussion_r3383124919
##########
lucene/core/src/java/org/apache/lucene/search/LRUQueryCache.java:
##########
@@ -1059,6 +1059,35 @@ void remove(QueryCacheKey queryCacheKey) {
remove(queryCacheKey, -1);
}
+ /**
+ * Remove all cache entries whose segment cache key is in {@code
keysToRemove} or whose query is
+ * in {@code queriesToRemove}. Runs under the partition write lock so that
iteration of the
+ * underlying {@code HashMap} does not race with concurrent mutation.
+ *
+ * <p>Note: {@link #remove(QueryCacheKey, long)} re-acquires the write
lock, which is safe
+ * because {@link
java.util.concurrent.locks.ReentrantReadWriteLock.WriteLock} is reentrant.
+ * This matches the pattern used by {@link #evictIfNecessary()}.
+ */
+ void removeMatching(Set<IndexReader.CacheKey> keysToRemove, Set<Query>
queriesToRemove) {
+ writeLock.lock();
Review Comment:
evictIfNecessary() exits early, but the key pattern is the same, hold the
write lock for the duration of the iteration. removeMatching() scans at most
maxSize/16 entries per partition, once per minute. that's not expensive.
snapshot alternative would need a read lock (which also blocks writers),
allocate a full copy, then acquire/release the write lock per removal. more
work, not less.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
