activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Joe Smith <joesmithc...@yahoo.com>
Subject Re: Load testing 5.5 - broker pauses between sends
Date Tue, 02 Aug 2011 15:34:48 GMT
Hi,

Just another note on our test run w/5.4.2 version which was done earlier.  The broker does
not exhibit the "pausing" behavior.  In fact, it actually achieve speed of 40k msg/sec on
some occasions.

However, on a close examination of log and system stats.  We notice the following:

We were using the fileQueueCursor

1. Test producer sending at approx. 40k msg/sec non-persisted.

2. We saw many paging error messages in the broker log.


2011-07-28 20:44:40,270 | ERROR | Failed to page in more queue messages  | org.apache.activemq.broker.region.Queue
| Queue:test.queue
java.lang.RuntimeException: java.lang.NullPointerException
        at org.apache.activemq.broker.region.cursors.FilePendingMessageCursor.getDiskList(FilePendingMessageCursor.java:444)
        at org.apache.activemq.broker.region.cursors.FilePendingMessageCursor.flushToDisk(FilePendingMessageCursor.java:423)
        at org.apache.activemq.broker.region.cursors.FilePendingMessageCursor.release(FilePendingMessageCursor.java:140)
        at org.apache.activemq.broker.region.Queue.doPageIn(Queue.java:1696)
        at org.apache.activemq.broker.region.Queue.pageInMessages(Queue.java:1898)
        at org.apache.activemq.broker.region.Queue.iterate(Queue.java:1425)
        at org.apache.activemq.thread.DedicatedTaskRunner.runTask(DedicatedTaskRunner.java:98)
        at org.apache.activemq.thread.DedicatedTaskRunner$1.run(DedicatedTaskRunner.java:36)
Caused by: java.lang.NullPointerException
        at org.apache.activemq.broker.region.cursors.FilePendingMessageCursor.getDiskList(FilePendingMessageCursor.java:441)
        ... 7 more


3. After variable number of above error messages - the memory usage shot up and the throughput
increased to 40k msg/sec.  The page in exception stopped showing up in the log.

4. It maintain the throughput  until the next broker reboot.

5. The behavior is not consistent.  Most of the times the broker just repeats the page in
exception and does not achieve the throughput.  The memory did not jump as before.

I believe there is a fix in 5.5 that maybe related to this issues.   Does the fix causing
the "pausing" behavior?

Anyone can provide some insight?

Regards.

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message