[
https://issues.apache.org/jira/browse/OAK-4263?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15251618#comment-15251618
]
Thomas Mueller commented on OAK-4263:
-------------------------------------
Possible patch (uses an AtomicBoolean):
{noformat}
===================================================================
--- src/main/java/org/apache/jackrabbit/oak/cache/CacheLIRS.java
(revision 1738734)
+++ src/main/java/org/apache/jackrabbit/oak/cache/CacheLIRS.java
(working copy)
@@ -27,6 +27,7 @@
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.ExecutionException;
+import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import javax.annotation.Nonnull;
@@ -74,9 +75,13 @@
*/
public class CacheLIRS<K, V> implements LoadingCache<K, V> {
- private static final Logger LOG = LoggerFactory.getLogger(CacheLIRS.class);
+ static final Logger LOG = LoggerFactory.getLogger(CacheLIRS.class);
+ static final ThreadLocal<Integer> CURRENTLY_LOADING = new
ThreadLocal<Integer>();
private static final AtomicInteger NEXT_CACHE_ID = new AtomicInteger();
- private static final ThreadLocal<Integer> CURRENTLY_LOADING = new
ThreadLocal<Integer>();
+
+ static {
+ LOG.info("CacheLIRS with less notifyAll");
+ }
/**
* Listener for items that are evicted from the cache. The listener
@@ -104,7 +109,7 @@
void evicted(@Nonnull K key, @Nullable V value, @Nonnull RemovalCause
cause);
}
- private final int cacheId = NEXT_CACHE_ID.getAndIncrement();
+ final int cacheId = NEXT_CACHE_ID.getAndIncrement();
/**
* The maximum memory this cache should use.
@@ -138,8 +143,8 @@
* value to be loaded need to wait on the synchronization object. The
* loading thread will notify all waiting threads once loading is done.
*/
- final ConcurrentHashMap<K, Object> loadingInProgress =
- new ConcurrentHashMap<K, Object>();
+ final ConcurrentHashMap<K, AtomicBoolean> loadingInProgress =
+ new ConcurrentHashMap<K, AtomicBoolean>();
/**
* Create a new cache with the given number of entries, and the default
@@ -952,14 +957,14 @@
// to prevent a deadlock, we also load the value ourselves
return load(key, hash, valueLoader);
}
- ConcurrentHashMap<K, Object> loading = cache.loadingInProgress;
+ ConcurrentHashMap<K, AtomicBoolean> loading =
cache.loadingInProgress;
// the object we have to wait for in case another thread loads
// this value
- Object alreadyLoading;
+ AtomicBoolean alreadyLoading;
// synchronized on this object, even before we put it in the
// cache, so that all other threads that get this object can
// synchronized and wait for it
- Object loadNow = new Object();
+ AtomicBoolean loadNow = new AtomicBoolean();
// we synchronize a bit early here, but that's fine (we don't
// optimize for the case where loading is extremely quick)
synchronized (loadNow) {
@@ -971,16 +976,20 @@
return load(key, hash, valueLoader);
} finally {
loading.remove(key);
- // notify other threads
- loadNow.notifyAll();
+ if (loadNow.get()) {
+ // notify other threads, but only if
+ // they wait for this to be loaded
+ loadNow.notifyAll();
+ }
CURRENTLY_LOADING.remove();
}
}
}
// another thread is (or was) already loading
synchronized (alreadyLoading) {
+ alreadyLoading.set(true);
// loading might have been finished, so check again
- Object alreadyLoading2 = loading.get(key);
+ AtomicBoolean alreadyLoading2 = loading.get(key);
if (alreadyLoading2 != alreadyLoading) {
// loading has completed before we could synchronize,
// so we repeat
{noformat}
> LIRS cache: excessive use of notifyAll
> --------------------------------------
>
> Key: OAK-4263
> URL: https://issues.apache.org/jira/browse/OAK-4263
> Project: Jackrabbit Oak
> Issue Type: Bug
> Reporter: Thomas Mueller
> Assignee: Thomas Mueller
> Fix For: 1.6
>
>
> The LIRS cache tries to avoid loading the same cache entry concurrently. To
> notify a possible waiting thread, it uses notifyAll. This is a relatively
> expensive operation, and should be avoided if possible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
