You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@activemq.apache.org by "Chen Maoqian (JIRA)" <ji...@apache.org> on 2016/08/18 02:33:22 UTC
[jira] [Updated] (ARTEMIS-684) Client connection are not balanced
across cluster with any load balancing policy
[ https://issues.apache.org/jira/browse/ARTEMIS-684?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Chen Maoqian updated ARTEMIS-684:
---------------------------------
Description:
When we use
*Scenario*
There two nodes in cluster. Nodes in cluster are connected and both nodes have same jms destinations and connection factory defined like the following one
{noformat}
<connection-factory name="RemoteConnectionFactory" connectors="http-connector" entries="java:jboss/exported/jms/RemoteConnectionFactory" ha="true" connection-ttl="120000" call-timeout="60000" block-on-acknowledge="true" retry-interval="1000" reconnect-attempts="-1" connection-load-balancing-policy-class-name="org.apache.activemq.artemis.api.core.client.loadbalance.RandomConnectionLoadBalancingPolicy"/>
{noformat}
There are clients connecting to cluster. With _RandomConnectionLoadBalancingPolicy_, I would expect that connections are randomly distributed among the nodes in cluster. However, when I list connections on both nodes (/subsystem=messaging-activemq/server=default:list-connection-ids), all connections are on one server.
Method {{private TransportConfiguration selectConnector()}} in class {{ServerLocatorImpl}} checks whether the {{topologyArray}} is null and makes decision between choosing connector to other node in topology or initial connector of connection factory. In method {{private TransportConfiguration selectConnector()}} {{topologyArray}} is still null, so clients make connection using initial connector. It seems like {{topologyArray}} is only updated during {{notifyNodeUp/Down}} methods.
Could you please explain how is this client connection balancing supposed to work?
was:
* Start two EAP instances with configuration attached to this JIRA as a {{standalone-full-ha.xml}} file.
* Unzip simple reproducer attached as a {{connection-balancing-reproducer.zip}}. It is simple maven project, only slightly adjusted helloword jms quickstart. There are 100 producers, each send 1000 messages to queue with 5 second delay between sends.
* Run reproducer and open CLI for both servers
* In CLI switch to {{/subsystem=messaging-activemq/server=default}} and invoke {{:list-connection-ids}} on both servers
*Scenario*
There two nodes in cluster. Nodes in cluster are connected and both nodes have same jms destinations and connection factory defined like the following one
{noformat}
<connection-factory name="RemoteConnectionFactory" connectors="http-connector" entries="java:jboss/exported/jms/RemoteConnectionFactory" ha="true" connection-ttl="120000" call-timeout="60000" block-on-acknowledge="true" retry-interval="1000" reconnect-attempts="-1" connection-load-balancing-policy-class-name="org.apache.activemq.artemis.api.core.client.loadbalance.RandomConnectionLoadBalancingPolicy"/>
{noformat}
There are clients connecting to cluster. With _RandomConnectionLoadBalancingPolicy_, I would expect that connections are randomly distributed among the nodes in cluster. However, when I list connections on both nodes (/subsystem=messaging-activemq/server=default:list-connection-ids), all connections are on one server.
Method {{private TransportConfiguration selectConnector()}} in class {{ServerLocatorImpl}} checks whether the {{topologyArray}} is null and makes decision between choosing connector to other node in topology or initial connector of connection factory. In method {{private TransportConfiguration selectConnector()}} {{topologyArray}} is still null, so clients make connection using initial connector. It seems like {{topologyArray}} is only updated during {{notifyNodeUp/Down}} methods.
Could you please explain how is this client connection balancing supposed to work?
> Client connection are not balanced across cluster with any load balancing policy
> --------------------------------------------------------------------------------
>
> Key: ARTEMIS-684
> URL: https://issues.apache.org/jira/browse/ARTEMIS-684
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Affects Versions: 1.4.0
> Reporter: Chen Maoqian
> Priority: Critical
>
> When we use
> *Scenario*
> There two nodes in cluster. Nodes in cluster are connected and both nodes have same jms destinations and connection factory defined like the following one
> {noformat}
> <connection-factory name="RemoteConnectionFactory" connectors="http-connector" entries="java:jboss/exported/jms/RemoteConnectionFactory" ha="true" connection-ttl="120000" call-timeout="60000" block-on-acknowledge="true" retry-interval="1000" reconnect-attempts="-1" connection-load-balancing-policy-class-name="org.apache.activemq.artemis.api.core.client.loadbalance.RandomConnectionLoadBalancingPolicy"/>
> {noformat}
> There are clients connecting to cluster. With _RandomConnectionLoadBalancingPolicy_, I would expect that connections are randomly distributed among the nodes in cluster. However, when I list connections on both nodes (/subsystem=messaging-activemq/server=default:list-connection-ids), all connections are on one server.
> Method {{private TransportConfiguration selectConnector()}} in class {{ServerLocatorImpl}} checks whether the {{topologyArray}} is null and makes decision between choosing connector to other node in topology or initial connector of connection factory. In method {{private TransportConfiguration selectConnector()}} {{topologyArray}} is still null, so clients make connection using initial connector. It seems like {{topologyArray}} is only updated during {{notifyNodeUp/Down}} methods.
> Could you please explain how is this client connection balancing supposed to work?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)