It seems frequently assumed that architecting and deploying Highly Available (HA) solutions requires Application Server and/or Operating System clustering. When it comes to SOA and Integration solutions this is not necessarily a correct assumption. Load Balanced (LB) and Highly Available HA) SOA and Integration solutions may not require that degree of complexity and sophistication. Frequently, protocol, binding component, JBI and architectural application design properties can be exploited to design highly available solutions. Testing LB and HA solutions requires infrastructure consisting of multiple hosts and the ability to “crash” hosts at will. With virtualization technologies available now it is far easier to use multiple virtual machines then to use physical machines. It is also easier and potentially less destructive to “crash” virtual machines then it is to do so with physical machines.
In this Note a heterogeneous, non-clustered collection of hosts will be used to implement and exercise three load balanced, highly available GlassFish ESB-based solutions. The environment consists of a number of independent “machines”, which are not a part of an Operating System Cluster. Each “machine” hosts a GlassFish Application Server. Application Servers are independent of one another and are not clustered. This is to demonstrate that load balanced, highly available, horizontally scalable solutions, based on the GlassFish ESB software alone, can be designed and implemented.
The specific class of solutions to which this discussion applies is the class of solutions which:
1. are exposed as request/reply services
a. HL7 messaging with explicit Application Acknowledgment
b. Request/Reply Web Services
c. JMS in Request/Reply mode
2. implement business logic as short lived processes
b. are idempotent
c. tolerant of duplicate messages
Classes of solutions with characteristics different from these named above require different approaches to high availability and horizontal scalability, and are not discussed here.
In this Note only high availability and scalability of receiver solutions is addressed. This aspect is the focus because a failure to process a message by a receiver may result in message loss –generally a bad thing.
Paradoxical as it may sound; senders are special cases of receivers. Just as a receiver is triggered by arrival of a message so too is a sender. Making sure that the sender trigger message does not get lost is much the same as making sure the message a receiver receives does not get lost. This means that the same considerations apply to senders and to receivers.
This note discusses an exercise involving an example load balanced, highly available, horizontally scalable healthcare environment, processing HL7 v2 messages. Discussion includes customization of generic GlassFish ESB v2.2 VMware Virtual Appliances for a specific Load Balancing and High Availability exercise and deploying ready-made GlassFish ESB solutions. The exercise for HL7 BC-based, Web Service-based and JMS-based highly available, load balanced, and horizontally scalable receivers, processing HL7 v2.3.1 messages, will be conducted and discussed.
At the end of the Note we will have three GlassFish ESB VMware Appliances with GlassFish ESB v2.2 Runtime infrastructure, ready to use for further GlassFish ESB Load Balancing and High Availability exercises.
The reader will be convinced, one hopes, that for the applicable class of GlassFish ESB-based solutions, load balancing and dynamic failover without message loss work. For that class of solutions this provides for high availability and horizontal scalability without resorting to Application Server or Operating System clustering.
The complete Note is available as 03_Conducting_HL7_LB_and_HA_Exercise_v22.214.171.124.pdf at http://blogs.czapski.id.au/wp-content/uploads/2010/03/03_Conducting_HL7_LB_and_HA_Exercise_v126.96.36.199.pdf