Hide Forgot
Affects: Documentation (Ref Guide, User Guide, etc.), Release Notes Date of First Response: 2008-02-27 16:27:31 project_key: SOA Document thatAggregator and stateful non-replicating actions in general cannot be used in ESB cluster config
Affects: Added: [Documentation (Ref Guide, User Guide, etc.), Release Notes]
Jiri, could you please update this JIRA with more information, or a pointer to source JIRAs for more information. thanks --josh
From an IRC conversation with Jiri: Aggregator is a stateful action. It assembles a predefined number of messages into a single message, which is then passed on for further processing. While the Aggregator is waiting for all of the messages that it needs to arrive its state is stored in memory. In a clustered ESB environment the Aggregator is running on multiple machines, but it's memory is not replicated across those machines. This means that the Aggregator instance on one machine does not have access to the messages that have been captured by Aggregator instances on the other machines in the cluster. This can lead to the situation where messages are captured by different Aggregator instances in the cluster. Since they are unaware of each other the messages will never be assembled. Because of this you should avoid using the Aggregator action in a clustered environment. Aggregator is the only stateful action that we deliver out of the box. If you create other stateful actions you will need to implement a replication mechanism to use them in a clustered environment.
Slightly edited: Aggregator is a stateful action. It assembles a predefined number of messages into a single message, which is then passed on for further processing. The Aggregator's state is stored in memory while it is waiting for the arrival of all the messages required. In a clustered ESB environment multiple Aggregator instances are running, on the different nodes of the cluster, each with its own memory space. The Aggregator instance on one node does not have access to the messages that have been captured by Aggregator instances on other nodes in the cluster. This can lead to the situation where messages are captured by different Aggregator instances in the cluster, and because they are unaware of each other the messages will never be assembled. Because of this you should avoid using the Aggregator action in a clustered environment. Aggregator is the only stateful action that we deliver out of the box. If you create other stateful actions you will need to implement a replication mechanism to use them in a clustered environment.
What's the roadmap for stateful action replication in a clustered environment? Is there one? Is there a pointer that I can give to more information on developing a replicating stateful action?
<mlittle> jwulf, replication of stateful services is complex. It's not supported in the current ESB. <mlittle> jwulf, it is a future feature though.
"5.5. Aggregator and other stateful actions are not supported in a clustered environment As mentioned throughout the Programmers Guide, high-availability services within JBossESB are currently only supported if they are stateless. The Aggregator action is a stateful instance, which necessarily means that if it is used within the Action Pipeline/Chain of a service, that service is no longer stateless and therefore cannot be used in a high-availability manner. Further information: http://jira.jboss.com/jira/browse/SOA-421 " Release note pushed to online - will be synced in next redhat.com/docs sync.