[
https://issues.apache.org/jira/browse/ARTEMIS-4003?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Clebert Suconic updated ARTEMIS-4003:
-------------------------------------
Description:
LargeMessages over BridgeImpl (including Clustering) will request more credits
eventually than what they are entitled. This is because we don't flow control
the large message producer on the Bridge (at least not yet), and the sender
could get to a window where it requested more credits than it was entitled. The
flow control on the Bridge is limited to handle returning busy or not (and that
should be enough really).
as a Result, the flow control will never refill and the bridge would freeze and
get stuck.
To fix this situation we should check for the starve at the end of the receive
credits and do the proper refill.
I am writing a test validating this issue, it will be under smokeTests and it
will be named
org.apache.activemq.artemis.tests.smoke.clusteredLargeMessage.ClusteredLargeMessageTest
was:
LargeMessages over BridgeImpl (including Clustering) will request more credits
eventually than what they are entitled. This is because we don't flow control
the large message producer on the Bridge (at least not yet), and the sender
could get to a window where it requested more credits than it was entitled. The
flow control on the Bridge is limited to handle returning busy or not (and that
should be enough really).
as a Result, the flow control will never refill and the bridge would freeze and
get stuck.
To fix this situation we should check for the starve the condition at the end
of the receive credits and do the proper refill.
I am writing a test validating this issue, it will be under smokeTests and it
will be named
org.apache.activemq.artemis.tests.smoke.clusteredLargeMessage.ClusteredLargeMessageTest
> FlowControl on BridgeImpl is broken for Large Messages
> ------------------------------------------------------
>
> Key: ARTEMIS-4003
> URL: https://issues.apache.org/jira/browse/ARTEMIS-4003
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Reporter: Clebert Suconic
> Assignee: Clebert Suconic
> Priority: Major
>
> LargeMessages over BridgeImpl (including Clustering) will request more
> credits eventually than what they are entitled. This is because we don't flow
> control the large message producer on the Bridge (at least not yet), and the
> sender could get to a window where it requested more credits than it was
> entitled. The flow control on the Bridge is limited to handle returning busy
> or not (and that should be enough really).
> as a Result, the flow control will never refill and the bridge would freeze
> and get stuck.
> To fix this situation we should check for the starve at the end of the
> receive credits and do the proper refill.
> I am writing a test validating this issue, it will be under smokeTests and it
> will be named
> org.apache.activemq.artemis.tests.smoke.clusteredLargeMessage.ClusteredLargeMessageTest
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
