You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@activemq.apache.org by "Yuriy Sidelnikov (JIRA)" <ji...@apache.org> on 2012/10/15 10:14:02 UTC
[jira] [Updated] (AMQ-4107) Message order can be broken for Topic
under a high load when topicPrefetch=1 and comsumer is slow
[ https://issues.apache.org/jira/browse/AMQ-4107?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Yuriy Sidelnikov updated AMQ-4107:
----------------------------------
Description:
For <amq:policyEntry topic=">" producerFlowControl="true" memoryLimit="30mb" {color:red}topicPrefetch="1"{color} blockedProducerWarningInterval="30">
Short excerpt from TopicSubscription class:
public void add(MessageReference node) throws Exception {
…..
if ({color:red}!isFull(){color} && matched.isEmpty() && !isSlave()) {
// if maximumPendingMessages is set we will only discard messages which
// have not been dispatched (i.e. we allow the prefetch buffer to be filled)
dispatch(node); <- Second message will go this way and might be dispatched sooner than first one.
setSlowConsumer(false);
} else {
…….
if (matched.tryAddMessageLast(node, 10)) { <- first message will be put in the VMCursor queue and might be dispatched later
break;
}
.....
dispatchMatched(); <- First message won't be dispatched immediately because !isFull() is still false
}
Possible scenario as I can see it from logs:
1. First message has arrived and !isFull() is false because consumer didn't take some previous message yet.
2. First message will be processed by tryAddMessageLast in VMPendingMessageCursor class and will be dispatched very lately because !isFull() is still false.
3. Meanwhile consumer reads some previous message and !isFull() will return true.
4. Second message will be dispatched immediately and might be first to be delivered.
5. Then first message is dispatched.
6. Message order is broken.
was:
For <amq:policyEntry topic=">" producerFlowControl="true" memoryLimit="30mb" topicPrefetch="1" blockedProducerWarningInterval="30">
Short excerpt from TopicSubscription class:
public void add(MessageReference node) throws Exception {
…..
if ({color:red}!isFull(){color} && matched.isEmpty() && !isSlave()) {
// if maximumPendingMessages is set we will only discard messages which
// have not been dispatched (i.e. we allow the prefetch buffer to be filled)
dispatch(node); <- Second message will go this way and might be dispatched sooner than first one.
setSlowConsumer(false);
} else {
…….
if (matched.tryAddMessageLast(node, 10)) { <- first message will be put in the VMCursor queue and might be dispatched later
break;
}
.....
dispatchMatched(); <- First message won't be dispatched immediately because !isFull() is still false
}
Possible scenario as I can see it from logs:
1. First message has arrived and !isFull() is false because consumer didn't take some previous message yet.
2. First message will be processed by tryAddMessageLast in VMPendingMessageCursor class and will be dispatched very lately because !isFull() is still false.
3. Meanwhile consumer reads some previous message and !isFull() will return true.
4. Second message will be dispatched immediately and might be first to be delivered.
5. Then first message is dispatched.
6. Message order is broken.
> Message order can be broken for Topic under a high load when topicPrefetch=1 and comsumer is slow
> -------------------------------------------------------------------------------------------------
>
> Key: AMQ-4107
> URL: https://issues.apache.org/jira/browse/AMQ-4107
> Project: ActiveMQ
> Issue Type: Bug
> Components: Transport
> Affects Versions: 5.6.0
> Reporter: Yuriy Sidelnikov
>
> For <amq:policyEntry topic=">" producerFlowControl="true" memoryLimit="30mb" {color:red}topicPrefetch="1"{color} blockedProducerWarningInterval="30">
> Short excerpt from TopicSubscription class:
> public void add(MessageReference node) throws Exception {
> …..
> if ({color:red}!isFull(){color} && matched.isEmpty() && !isSlave()) {
> // if maximumPendingMessages is set we will only discard messages which
> // have not been dispatched (i.e. we allow the prefetch buffer to be filled)
> dispatch(node); <- Second message will go this way and might be dispatched sooner than first one.
> setSlowConsumer(false);
> } else {
> …….
> if (matched.tryAddMessageLast(node, 10)) { <- first message will be put in the VMCursor queue and might be dispatched later
> break;
> }
> .....
> dispatchMatched(); <- First message won't be dispatched immediately because !isFull() is still false
> }
> Possible scenario as I can see it from logs:
> 1. First message has arrived and !isFull() is false because consumer didn't take some previous message yet.
> 2. First message will be processed by tryAddMessageLast in VMPendingMessageCursor class and will be dispatched very lately because !isFull() is still false.
> 3. Meanwhile consumer reads some previous message and !isFull() will return true.
> 4. Second message will be dispatched immediately and might be first to be delivered.
> 5. Then first message is dispatched.
> 6. Message order is broken.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira