[jira] Created: (AMQ-2475) If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks

If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
-----------------------------------------------------------------------------------------------------------------------

                 Key: AMQ-2475
                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
             Project: ActiveMQ
          Issue Type: Bug
          Components: Broker, Message Store, Transport
    Affects Versions: 5.3.0
         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
            Reporter: Martin Murphy


I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.

Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
{code}
    public void add(MessageReference node) throws Exception { //... snip ...
            if (maximumPendingMessages != 0) {
                synchronized (matchedListMutex) {   // We have this mutex
                    matched.addMessageLast(node); // ends up waiting for space
                    // NOTE - be careful about the slaveBroker!
                    if (maximumPendingMessages > 0) {
{code}
Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
{code}
    private void dispatchMatched() throws IOException {      
        synchronized (matchedListMutex) {  // never gets passed here.
            if (!matched.isEmpty() && !isFull()) {
{code}
This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.


--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Martin Murphy updated AMQ-2475:
-------------------------------

    Attachment: hangtest.zip

Oops, didn't realize that the test never attached properly last time

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>         Attachments: hangtest.zip
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Rob Davies reassigned AMQ-2475:
-------------------------------

    Assignee: Rob Davies

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: hangtest.zip
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

    [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=55210#action_55210 ]

Dominic Tootell commented on AMQ-2475:
--------------------------------------

I had an investigate into attempted to patch this locally in activemq-core, on a fusesource 5.3.0.4 (MacOSX 10.6.1).

I've run the following tests on the patch, will I'll attach: (the patch diffs, the patched .java and the broker xml I used in testing):

The test cases I've run overnight and this morning/afternoon are:

- Virtual Topic (  VirtualTopic.iplayer  -> Consumer.A.VirtualTopic.iplayer)
- 3 x Producer, 4,000,000 messages each onto Virtual Topic (12million in total)
- 1 x Consumer
- 100mb tmp_store limit

- Virtual Topic (  VirtualTopic.iplayer  -> Consumer.A.VirtualTopic.iplayer)
- 6 x Producer, 2,000,000 messages each onto Virtual Topic (12million in total)
- 1 x Consumer
- 512mb tmp_store limit

The tmp_storage was definitely limiting ok, and niether the broker, producer or consumer blocked:

du -sh of the tmp_storage area:
{code}
dominic-tootells-macbook-pro:data dominict$ du -sh *
 96M journal
 48K kr-store
  0B lock
512M tmp-test-broker

dominic-tootells-macbook-pro:data dominict$ du -sh *
 96M journal
 48K kr-store
  0B lock
483M tmp-test-broker

dominic-tootells-macbook-pro:data dominict$ du -sh *
 96M journal
 48K kr-store
  0B lock
490M tmp-test-broker

dominic-tootells-macbook-pro:data dominict$ du -sh *
 64M journal
 48K kr-store
  0B lock
 38M tmp-test-broker

dominic-tootells-macbook-pro:data dominict$

{code}


I've also run the junit provided by Martin; this ran ok too; with no blockage.

I shall attach the potential patches.  I haven't run any other tests against the patches; to see if they potentially cause any other unforeseen issues (i.e. normal persistent queue - will do this later on)

cheers
/dom

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: hangtest.zip
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment: Queue.patchfile.txt
                Topic.patchfile.txt
                TopicSubscription.patchfile.txt

all in package org.apache.activemq.broker.region


TopicSubscription.patchfile.txt      (Changes to TopicSubscription.java)
Topic.patchfile.txt                          (Changes to Topic.java)
Queue.patchfile.txt                       (Changes to Queue.java)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment: TopicSubscription.java
                Topic.java
                Queue.java

The patched java files:

Queue.java
Topic.java
TopicSubscription.java

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment: activemq.xml

activemq.xml test broker configuration I was using to test the scenario

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

    [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=55218#action_55218 ]

Dominic Tootell commented on AMQ-2475:
--------------------------------------

I've ran some more tests on the patch I uploaded yesterday, and come across a small issue with  sendFailIfNoSpace="true".  The ResourceAllocationException would only be thrown if the producer noticed the out of space condition before the message was added to the cursor.  However, there was a slight chance space would be available when producer 1 checked, but this space was then eaten by producer 2.  Producer 1 would then be within the waiting for space loop; and not send a ResourceAllocationException.  

I have added checks within the waiting for space loop, to check if a ResourceAllocationException should be thrown if sendFailIfNoSpace="true".  

I shall update the patches attached yesterday, to reflect this.

apologies,
/dom

I'm also currently running the test against a normal persistent queue; to make sure all is ok with that; I'll comment back once the run has finished.

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment:     (was: Topic.patchfile.txt)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment:     (was: Queue.patchfile.txt)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment:     (was: Queue.java)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment:     (was: TopicSubscription.patchfile.txt)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment: TopicSubscription.patchfile.txt
                Topic.patchfile.txt
                Queue.patchfile.txt

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment:     (was: TopicSubscription.java)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment:     (was: Topic.java)

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.patchfile.txt, Topic.patchfile.txt, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

     [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Dominic Tootell updated AMQ-2475:
---------------------------------

    Attachment: TopicSubscription.java
                Queue.java
                Topic.java

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

    [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=55222#action_55222 ]

Dominic Tootell commented on AMQ-2475:
--------------------------------------

Test for a normal persistent queue ran ok:

- 1 consumer
- 5 * producer
- 10,000,000 * 1k messages

hope this helps,
/dom

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

    [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=55413#action_55413 ]

Rob Davies commented on AMQ-2475:
---------------------------------

I applied the patch to trunk - but got some JUnit test failures - going to resolve the dead lock differently.
I'm not clear if you are experiencing the same problem with Queues ?

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

    [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=55413#action_55413 ]

Rob Davies edited comment on AMQ-2475 at 11/13/09 11:50 AM:
------------------------------------------------------------

I applied the patch to trunk - but got some JUnit test failures -  org.apache.activemq.broker.BrokerTest is the main one (a few JUnit tests are derived from this one).
Going to try resolve the dead lock differently.
I'm not clear if you are experiencing the same problem with Queues ?

      was (Author: rajdavies):
    I applied the patch to trunk - but got some JUnit test failures - going to resolve the dead lock differently.
I'm not clear if you are experiencing the same problem with Queues ?
 

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

    [ https://issues.apache.org/activemq/browse/AMQ-2475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=55493#action_55493 ]

Dominic Tootell commented on AMQ-2475:
--------------------------------------

I wasn't experiencing the same issue with queue's, just topics (non persistent when they overflowed to tmp_storage)

What test was failing in the org.apache.activemq.broker.BrokerTest? out of interest.
/dom

> If tmp message store fills up, broker can deadlock due to while producers wait on disk space and consumers wait on acks
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-2475
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2475
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker, Message Store, Transport
>    Affects Versions: 5.3.0
>         Environment: Tested on Windows XP with JDK 1.60_13, but fairly sure it will be an issue on all platforms
>            Reporter: Martin Murphy
>            Assignee: Rob Davies
>         Attachments: activemq.xml, hangtest.zip, Queue.java, Queue.patchfile.txt, Topic.java, Topic.patchfile.txt, TopicSubscription.java, TopicSubscription.patchfile.txt
>
>
> I will attach a simple project that shows this. In the test the tmp space is set to 32 MB and two threads are created. One thread will constantly produce 1KB messages and the other consumes these, but sleeps for 100ms, note that producer flow control is turned off as well. The goal here is to ensure that the producers block while the consumers read the rest of the messages from the broker and catch up, this in turn frees up the disk space and allows the producer to send more messages. This config means that you can bound the broker based on disk space rather than memory usage.
> Unfortunately in this test using topics while the broker is reading in the message from the producer it has to lock the matched list it is adding it to. This is an abstract from the Topic's point of view and doesn't realize that the file may block based on the file system.
> {code}
>     public void add(MessageReference node) throws Exception { //... snip ...
>             if (maximumPendingMessages != 0) {
>                 synchronized (matchedListMutex) {   // We have this mutex
>                     matched.addMessageLast(node); // ends up waiting for space
>                     // NOTE - be careful about the slaveBroker!
>                     if (maximumPendingMessages > 0) {
> {code}
> Meanwhile the consumer is sending acknowledgements for the 10 messages it just read in (the configured prefetch) from the same topic, but since they also modify the same list in the topic this waits as well on the mutex held to service the producer:
> {code}
>     private void dispatchMatched() throws IOException {      
>         synchronized (matchedListMutex) {  // never gets passed here.
>             if (!matched.isEmpty() && !isFull()) {
> {code}
> This is a fairly classic deadlock. The trick is now how to resolve this given the fact that the topic isn't aware that it's list may need to wait for the file system to clean up.

--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.