[ 
https://issues.apache.org/jira/browse/ARTEMIS-2057?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16595072#comment-16595072
 ] 

ASF GitHub Bot commented on ARTEMIS-2057:
-----------------------------------------

Github user gemmellr commented on a diff in the pull request:

    https://github.com/apache/activemq-artemis/pull/2272#discussion_r213341266
  
    --- Diff: 
artemis-protocols/artemis-amqp-protocol/src/main/java/org/apache/activemq/artemis/protocol/amqp/broker/AMQPSessionCallback.java
 ---
    @@ -581,8 +581,8 @@ public void offerProducerCredit(final SimpleString 
address,
              Runnable creditRunnable = () -> {
                 connection.lock();
                 try {
    -               if (receiver.getRemoteCredit() <= threshold) {
    -                  receiver.flow(credits);
    +               if (receiver.getCredit() <= threshold) {
    +                  receiver.flow(credits - receiver.getCredit());
    --- End diff --
    
    I think this should protect against trying to flow <=0 after the 
calculation.


> AMQP: Credit window can grow beyond max credit configuration
> ------------------------------------------------------------
>
>                 Key: ARTEMIS-2057
>                 URL: https://issues.apache.org/jira/browse/ARTEMIS-2057
>             Project: ActiveMQ Artemis
>          Issue Type: Improvement
>          Components: AMQP
>    Affects Versions: 2.6.2
>            Reporter: Timothy Bish
>            Assignee: Timothy Bish
>            Priority: Blocker
>             Fix For: 2.6.3
>
>
> The broker currently grants credit based on the 'amqpCredit' configuration 
> which default to 1000 using a check on the "remote credit" view which 
> includes messages currently queued into proton but not processed by the 
> protocol head yet.  This accounting can lead to a runaway credit situation if 
> the sender is particularly fast or a significant number of small messages are 
> batched into one TCP frame and the broker isn't able to clear the backlog 
> from the proton queue fast enough such that the queue grows to the point that 
> the granting algorithm starts granting "amqpCredit" number of credits on each 
> new messages it processes from proton.
> Another smaller issue is that the granting of credit allows for a client to 
> have an outstanding balance of more than the configured "amqpCredits" which 
> may not be what was originally intended by that option and doesn't match what 
> many of the clients do when topping off credit at a low water mark. 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to