David Sitsky created AMQ-6066:
---------------------------------
Summary: Performance issue in OrderedPendingList
Key: AMQ-6066
URL: https://issues.apache.org/jira/browse/AMQ-6066
Project: ActiveMQ
Issue Type: Bug
Components: Broker
Affects Versions: 5.12.1
Reporter: David Sitsky
Fix For: 5.13.0
I have updated my application from ActiveMQ 5.3 to 5.11.1 and have noticed a
serious performance degregation. Running a number of jstacks I can see the
broker is often stuck here:
{noformat}
"Queue:master-items" Id=122 RUNNABLE
at
org.apache.activemq.broker.region.cursors.OrderedPendingList.contains(OrderedPendingList.java:144)
at
org.apache.activemq.broker.region.Queue.doPageInForDispatch(Queue.java:1930)
at
org.apache.activemq.broker.region.Queue.pageInMessages(Queue.java:2119)
at org.apache.activemq.broker.region.Queue.iterate(Queue.java:1596)
- locked java.lang.Object@253c3089
at
org.apache.activemq.thread.DedicatedTaskRunner.runTask(DedicatedTaskRunner.java:112)
at
org.apache.activemq.thread.DedicatedTaskRunner$1.run(DedicatedTaskRunner.java:42)
Number of locked synchronizers = 1
- java.util.concurrent.locks.ReentrantReadWriteLock$NonfairSync@2eb46567
{noformat}
For this specific queue in my application, there were a large number of items
in it.. around 100,000. However I noticed the code for contains has:
{code}
public boolean contains(MessageReference message) {
if (message != null) {
for (PendingNode value : map.values()) {
if (value.getMessage().equals(message)) {
return true;
}
}
}
return false;
}
{code}
This will obviously be very slow. Given the Map is keyed by message ID, can't
we do a .contains(message.getMessageId()) instead? I noticed the remove()
method does this. I changed the code to do this:
{code}
@Override
public boolean contains(MessageReference message) {
if (message != null) {
return map.containsKey(message.getMessageId());
}
return false;
}
{code}
I got a speedup for my test which now takes 29 minutes from 41 minutes. Can we
get this change in to the upcoming 5.13 release (pretty please??) It is a tiny
(but very important) change for us.. many thanks.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
