This chapter describes the high availability features in Oracle GlassFish Server 3.1.2.
The following topics are addressed here:
High availability applications and services provide their functionality continuously, regardless of hardware and software failures. To make such reliability possible, GlassFish Server provides mechanisms for maintaining application state data between clustered GlassFish Server instances. Application state data, such as HTTP session data, stateful EJB sessions, and dynamic cache information, is replicated in real time across server instances. If any one server instance goes down, the session state is available to the next failover server, resulting in minimum application downtime and enhanced transactional security.
GlassFish Server provides the following high availability features:
The HTTP Load Balancer Plug-in for Oracle GlassFish Server 3.1.2 accepts HTTP/HTTPS requests and forwards them to application server instances in a cluster. If an instance fails, becomes unavailable (due to network faults), or becomes unresponsive, the load balancer redirects requests to existing, available hosts. The load balancer can also recognize when a failed instance has recovered and redistribute the load accordingly. The Load Balancer Plug-in is compatible with the Oracle iPlanet Web Server, Oracle HTTP Server, Apache HTTP Server, and Microsoft Internet Information Server.
By distributing workload among multiple physical hosts, the load balancer increases overall system throughput. It also provides higher availability through failover of HTTP requests. For HTTP session information to persist, you must configure HTTP session persistence.
For simple, stateless applications a load-balanced cluster may be sufficient. However, for mission-critical applications with session state, use load balanced clusters with replicated persistence.
Server instances and clusters participating in load balancing have a homogenous environment. Usually that means that the server instances reference the same server configuration, can access the same physical resources, and have the same applications deployed to them. Homogeneity assures that before and after failures, the load balancer always distributes load evenly across the active instances in the cluster.
GlassFish Server provides high availability of HTTP requests and session data (both HTTP session data and stateful session bean data).
Java EE applications typically have significant amounts of session state data. A web shopping cart is the classic example of a session state. Also, an application can cache frequently-needed data in the session object. In fact, almost all applications with significant user interactions need to maintain session state. Both HTTP sessions and stateful session beans (SFSBs) have session state data.
Preserving session state across server failures can be important to end users. If the GlassFish Server instance hosting the user session experiences a failure, the session state can be recovered, and the session can continue without loss of information. High availability is implemented in GlassFish Server by means of in-memory session replication on GlassFish Server instances running in a cluster.
For more information about in-memory session replication in GlassFish Server, see How GlassFish Server Provides High Availability. For detailed instructions on configuring high availability session persistence, see Configuring High Availability Session Persistence and Failover.
GlassFish Server supports the Java Message Service (JMS) API and JMS messaging through its built-in jmsra resource adapter communicating with GlassFish Server Message Queue as the JMS provider. This combination is often called the JMS Service.
The JMS service makes JMS messaging highly available as follows:
By default, when a GlassFish cluster is created, the JMS service automatically configures a Message Queue broker cluster to provide JMS messaging services, with one clustered broker assigned to each cluster instance. This automatically created broker cluster is configurable to take advantage of the two types of broker clusters, conventional and enhanced, supported by Message Queue.
Additionally, Message Queue broker clusters created and managed using Message Queue itself can be used as external, or remote, JMS hosts. Using external broker clusters provides additional deployment options, such as deploying Message Queue brokers on different hosts from the GlassFish instances they service, or deploying different numbers of Message Queue brokers and GlassFish instances.
For more information about Message Queue clustering, see Using Message Queue Broker Clusters With GlassFish Server.
The use of Message Queue broker clusters allows connection failover in the event of a broker failure. If the primary JMS host (Message Queue broker) in use by a GlassFish instance fails, connections to the failed JMS host will automatically fail over to another host in the JMS host list, allowing messaging operations to continue and maintaining JMS messaging semantics.
For more information about JMS connection failover, see Connection Failover.
With RMI-IIOP load balancing, IIOP client requests are distributed to different server instances or name servers, which spreads the load evenly across the cluster, providing scalability. IIOP load balancing combined with EJB clustering and availability also provides EJB failover.
When a client performs a JNDI lookup for an object, the Naming Service essentially binds the request to a particular server instance. From then on, all lookup requests made from that client are sent to the same server instance, and thus all
EJBHome objects will be hosted on the same target server. Any bean references obtained henceforth are also created on the same target host. This effectively provides load balancing, since all clients randomize the list of target servers when performing JNDI lookups. If the target server instance goes down, the lookup or EJB method invocation will failover to another server instance.
IIOP Load balancing and failover happens transparently. No special steps are needed during application deployment. If the GlassFish Server instance on which the application client is deployed participates in a cluster, the GlassFish Server finds all currently active IIOP endpoints in the cluster automatically. However, a client should have at least two endpoints specified for bootstrapping purposes, in case one of the endpoints has failed.
For more information on RMI-IIOP load balancing and failover, see RMI-IIOP Load Balancing and Failover.
GlassFish Server provides high availability through the following subcomponents and features:
Storing session state data enables the session state to be recovered after the failover of a server instance in a cluster. Recovering the session state enables the session to continue without loss of information. GlassFish Server supports in-memory session replication on other servers in the cluster for maintaining HTTP session and stateful session bean data.
In-memory session replication is implemented in GlassFish Server 3.1.2 as an OSGi module. Internally, the replication module uses a consistent hash algorithm to pick a replica server instance within a cluster of instances. This allows the replication module to easily locate the replica or replicated data when a container needs to retrieve the data.
The use of in-memory replication requires the Group Management Service (GMS) to be enabled. For more information about GMS, see Group Management Service.
If server instances in a cluster are located on different hosts, ensure that the following prerequisites are met:
To ensure that GMS and in-memory replication function correctly, the hosts must be on the same subnet.
To ensure that in-memory replication functions correctly, the system clocks on all hosts in the cluster must be synchronized as closely as possible.
A highly available cluster integrates a state replication service with clusters and load balancer.
When implementing a highly available cluster, use a load balancer that includes session-based stickiness as part of its load-balancing algorithm. Otherwise, session data can be misdirected or lost. An example of a load balancer that includes session-based stickiness is the Loadbalancer Plug-In available in Oracle GlassFish Server.
Clusters, server instances, load balancers, and sessions are related as follows:
A server instance is not required to be part of a cluster. However, an instance that is not part of a cluster cannot take advantage of high availability through transfer of session state from one instance to other instances.
The server instances within a cluster can be hosted on one or multiple hosts. You can group server instances across different hosts into a cluster.
A particular load balancer can forward requests to server instances on multiple clusters. You can use this ability of the load balancer to perform an online upgrade without loss of service. For more information, see Upgrading in Multiple Clusters.
A single cluster can receive requests from multiple load balancers. If a cluster is served by more than one load balancer, you must configure the cluster in exactly the same way on each load balancer.
Each session is tied to a particular cluster. Therefore, although you can deploy an application on multiple clusters, session failover will occur only within a single cluster.
The cluster thus acts as a safe boundary for session failover for the server instances within the cluster. You can use the load balancer and upgrade components within the GlassFish Server without loss of service.
GlassFish Server uses the Distributed Component Object Model (DCOM) remote protocol or secure shell (SSH) to ensure that clusters that span multiple hosts can be administered centrally. To perform administrative operations on GlassFish Server instances that are remote from the domain administration server (DAS), the DAS must be able to communicate with those instances. If an instance is running, the DAS connects to the running instance directly. For example, when you deploy an application to an instance, the DAS connects to the instance and deploys the application to the instance.
However, the DAS cannot connect to an instance to perform operations on an instance that is not running, such as creating or starting the instance. For these operations, the DAS uses DCOM or SSH to contact a remote host and administer instances there. DCOM or SSH provides confidentiality and security for data that is exchanged between the DAS and remote hosts.
The use of DCOM or SSH to enable centralized administration of remote instances is optional. If the use of DCOM SSH is not feasible in your environment, you can administer remote instances locally.
For more information, see Enabling Centralized Administration of GlassFish Server Instances.
You can use various techniques to manually recover individual subcomponents after hardware failures such as disk crashes.
The following topics are addressed here:
Loss of the Domain Administration Server (DAS) affects only administration. GlassFish Server clusters and standalone instances, and the applications deployed to them, continue to run as before, even if the DAS is not reachable
Use any of the following methods to recover the DAS:
Back up the domain periodically, so you have periodic snapshots. After a hardware failure, re-create the DAS on a new host, as described in "Re-Creating the Domain Administration Server (DAS)" in Oracle GlassFish Server Administration Guide.
Put the domain installation and configuration on a shared and robust file system (NFS for example). If the primary DAS host fails, a second host is brought up with the same IP address and will take over with manual intervention or user supplied automation.
Zip the GlassFish Server installation and domain root directory. Restore it on the new host, assigning it the same network identity.
GlassFish Server provide tools for backing up and restoring GlassFish Server instances. For more information, see To Resynchronize an Instance and the DAS Offline.
There are no explicit commands to back up only a web server configuration. Simply zip the web server installation directory. After failure, unzip the saved backup on a new host with the same network identity. If the new host has a different IP address, update the DNS server or the routers.
This assumes that the web server is either reinstalled or restored from an image first.
The Load Balancer Plug-In (
plugins directory) and configurations are in the web server installation directory, typically
/opt/SUNWwbsvr. The web-install
/config directory contains the
When a Message Queue broker becomes unavailable, the method you use to restore the broker to operation depends on the nature of the failure that caused the broker to become unavailable:
Power failure or failure other than disk storage
Failure of disk storage
Additionally, the urgency of restoring an unavailable broker to operation depends on the type of the broker:
Standalone Broker. When a standalone broker becomes unavailable, both service availability and data availability are interrupted. Restore the broker to operation as soon as possible to restore availability.
Broker in a Conventional Cluster. When a broker in a conventional cluster becomes unavailable, service availability continues to be provided by the other brokers in the cluster. However, data availability of the persistent data stored by the unavailable broker is interrupted. Restore the broker to operation to restore availability of its persistent data.
Broker in an Enhanced Cluster. When a broker in an enhanced cluster becomes unavailable, service availability and data availability continue to be provided by the other brokers in the cluster. Restore the broker to operation to return the cluster to its previous capacity.
When a host is affected by a power failure or failure of a non-disk component such as memory, processor or network card, restore Message Queue brokers on the affected host by starting the brokers after the failure has been remedied.
To start brokers serving as Embedded or Local JMS hosts, start the GlassFish instances the brokers are servicing. To start brokers serving as Remote JMS hosts, use the
imqbrokerd Message Queue utility.
Message Queue uses disk storage for software, configuration files and persistent data stores. In a default GlassFish installation, all three of these are generally stored on the same disk: the Message Queue software in as-install-parent
/mq, and broker configuration files and persistent data stores (except for the persistent data stores of enhanced clusters, which are housed in highly available databases) in domain-dir
/imq. If this disk fails, restoring brokers to operation is impossible unless you have previously created a backup of these items. To create such a backup, use a utility such as
tar to create archives of these directories and all their content. When creating the backup, you should first quiesce all brokers and physical destinations, as described in "Quiescing a Broker" and "Pausing and Resuming a Physical Destination" in Oracle GlassFish Server Message Queue Administration Guide, respectively. Then, after the failed disk is replaced and put into service, expand the backup archive into the same location.
Restoring the Persistent Data Store From Backup. For many messaging applications, restoring a persistent data store from backup does not produce the desired results because the backed up store does not represent the content of the store when the disk failure occurred. In some applications, the persistent data changes rapidly enough to make backups obsolete as soon as they are created. To avoid issues in restoring a persistent data store, consider using a RAID or SAN data storage solution that is fault tolerant, especially for data stores in production environments.
For information about planning a high-availability deployment, including assessing hardware requirements, planning network configuration, and selecting a topology, see the Oracle GlassFish Server Deployment Planning Guide. This manual also provides a high-level introduction to concepts such as:
GlassFish Server components such as node agents, domains, and clusters
IIOP load balancing in a cluster
Message queue failover
For more information about developing applications that take advantage of high availability features, see the Oracle GlassFish Server Application Development Guide.
For information on how to configure and tune applications and GlassFish Server for best performance with high availability, see the Oracle GlassFish Server Performance Tuning Guide, which discusses topics such as:
Tuning persistence frequency and persistence scope
Checkpointing stateful session beans
Configuring the JDBC connection pool
Configuring load balancers for best performance