[
https://issues.apache.org/jira/browse/OAK-1666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13961684#comment-13961684
]
Thomas Mueller commented on OAK-1666:
-------------------------------------
I'm convinced that the inUse map is not a problem. According to my test, one
can do around 18 million (!) inUse map operations per second. Test case:
{noformat}
package org.apache.jackrabbit.core.data;
import java.lang.ref.WeakReference;
import java.util.Collections;
import java.util.Map;
import java.util.WeakHashMap;
import java.util.concurrent.CountDownLatch;
public class InUseMapTest {
public static void main(String... args) throws Exception {
new InUseMapTest().test();
new InUseMapTest().test();
new InUseMapTest().test();
}
void test() throws Exception {
int count = 150;
final long repeat = 100000;
final DataIdentifier id = new DataIdentifier("abc");
final Map<DataIdentifier, WeakReference<DataIdentifier>> inUse =
Collections.synchronizedMap(
new WeakHashMap<DataIdentifier,
WeakReference<DataIdentifier>>());
Thread[] threads = new Thread[count];
final CountDownLatch latch = new CountDownLatch(1);
final Exception[] ex = new Exception[1];
for(int i=0; i<count; i++) {
Thread t = new Thread() {
@Override
public void run() {
try {
latch.await();
for(int i=0; i<repeat; i++) {
inUse.put(id, new
WeakReference<DataIdentifier>(id));
}
} catch (Exception e) {
ex[0] = e;
}
}
};
t.start();
threads[i] = t;
}
long time = System.currentTimeMillis();
latch.countDown();
for(int i=0; i<count; i++) {
threads[i].join();
}
if (ex[0] != null) {
throw ex[0];
}
time = System.currentTimeMillis() - time;
System.out.println("millis: " + time);
System.out.println("threads: " + count);
System.out.println("repeat: " + repeat);
System.out.println("# ISO8601.parse calls: " + count * repeat);
System.out.println("# ISO8601.parse calls per second: " +
(count * repeat) * 1000 / time);
}
}
{noformat}
Output for my system:
{noformat}
millis: 801
threads: 150
repeat: 100000
# ISO8601.parse calls: 15000000
# ISO8601.parse calls per second: 18726591
millis: 819
threads: 150
repeat: 100000
# ISO8601.parse calls: 15000000
# ISO8601.parse calls per second: 18315018
millis: 841
threads: 150
repeat: 100000
# ISO8601.parse calls: 15000000
# ISO8601.parse calls per second: 17835909
{noformat}
> FileDataStore inUse map causes contention in concurrent env
> -----------------------------------------------------------
>
> Key: OAK-1666
> URL: https://issues.apache.org/jira/browse/OAK-1666
> Project: Jackrabbit Oak
> Issue Type: Improvement
> Components: core
> Reporter: Chetan Mehrotra
> Assignee: Chetan Mehrotra
> Priority: Minor
> Fix For: 1.1
>
>
> JR2 FileDataStore#inUseMap [1] is currently a synchronized map and that at
> times causes contention concurrent env. This map is used for supporting the
> Blob GC logic for JR2.
> With Oak this map content is not used. As a fix we can either
> # Set inUseMap to a Guava Cache Map which has weak keys and value
> # Set inUseMap to a no op map where all put calls are ignored
> # Modify FDS to disable use of inUseMap or make {{usesIdentifier}} protected
> #3 would be a proper fix and #2 can be used as temp workaround untill FDS
> gets fixed
> [1]
> https://github.com/apache/jackrabbit/blob/trunk/jackrabbit-data/src/main/java/org/apache/jackrabbit/core/data/FileDataStore.java#L118
--
This message was sent by Atlassian JIRA
(v6.2#6252)
