HP Integrity NonStop H-Series - Manual (Page 19)

of instances of NSASJ are always running, and hence, the business services offered by NSASJ
are always available.
•
Infinispan: Each installation of NSASJ has an Infinispan cache cluster to maintain state
information.
Infinispan cache cluster is configured in such a way that at least one node acts as a backup
for another. Thus, all state information is replicated and is available in another node too.
Failure of a node does not result in loss of state information. Upon failure of a node, the
configuration of the cluster ensures that any future requests for the state is serviced by the
backup node. When the failed node restarts, it acquires all its state information from the
backup node and thus restores parity.
•
Domain/Host Controller: Each installation of NSASJ has only one instance of Domain/Host
Controller. This is configured as a TS/MP serverclass with only one instance and serves as
the single point of management for the container.
In case the Domain/Host Controller fails, the management interface becomes unavailable.
However, upon a restart, the management interface becomes available again.
The following example describes how the NSASJ components together with NonStop infrastructure
provide HA:
•
Remote object lookup: The client first looks up the remote EJB object. The request gets routed
to an NSASJ Server instance through the Post Master.
◦ Since more than one instance of the Post Master is running, the remote lookup service
survives any infrastructure (hardware or software) failure.
◦ TCP subsystem (provided by either TCPIPv6 or IP-CIP) provides the unique feature that
allows multiple instances to all service requests on a server port. Since the Post Master
uses Pathsend API to communicate with NSASJ, TS/MP ensures the availability of
NSASJ.Thus it can be seen how the integration with TCP/IP and TS/MP subsystems make
NSASJ services (in this case Remote Object Lookup) highly available.
•
Remote EJB method invocation: The client invokes a method on the remote EJB object. The
request once again gets routed to an NSASJ Server instance through the Post Master.
◦ Since the Post Master communicates with the NSASJ Server using Pathsend API, the entire
setup inherits the TS/MP features.
◦ With the appropriate configuration, the remote EJB invocations get distributed across the
NSASJ serverclass. Load distribution is a feature of TS/MP.
◦ Failure of any one or more NSASJ serverclass instances (either due to software or hardware
failure) has no impact on the client as TS/MP ensures high availability of its serverclasses.
NOTE: In case of an NSASJ instance failure during processing a client request “Remoting
connection can no longer process messages” is returned to the client. In such cases, the
client needs a restart.
•
Web session replication handling: The web client accesses a Servlet or JSP and starts a session.
The request is routed to one of the NSASJ Server instances through the Post Master.
◦ Since the Post Master communicates with the NSASJ Server using Pathsend API, the entire
setup inherits the features of TS/MP.
◦ The NSASJ Server that processes the request either creates a new session if the session
is not present or loads the session from the Infinispan Cache server.
◦ After the request is processed, the NSASJ Server instances store the updated session data
back to the Infinispan Cache server.
High Availability 19