Cluster member shutdown

In general, a PolicyCenter cluster contains the following types of servers:

One or more user interface servers that process web requests, perform business transactions, and render web pages.
One or more non-user interface servers that manage batch processing, work queues, scheduling, message destinations, and startable services (plugins).

The shutdown procedure for a Guidewire PolicyCenter cluster member is dependent on the role of server involved.

ui-role cluster member shutdown

Guidewire PolicyCenter provides the means to schedule a planned PolicyCenter server shutdown. You schedule a server shutdown through the Server Tools Cluster Components screen.

After you schedule a planned shutdown on a specific server:

PolicyCenter shows an on-screen banner for that server instance that warns of the impending shutdown.
The banner shows a count-down timer to the planned shutdown and recommends that any logged-in user log out of PolicyCenter.

See Schedule a planned cluster member shutdown for more information.

Non-ui role cluster member shutdown

As with PolicyCenter user interface servers, you schedule a shutdown of a server cluster member with a non-user interface role through the Server Tools Cluster Components screen. However, there are additional requirements for shutting down a non-user interface server.

Stop any running background tasks in the following order:

Background task		Shutdown considerations
1.	Scheduler	The scheduler typically runs on several servers, as a distributed component. Thus, stopping the scheduler on one server does not affect the cluster.
2.	Batch processing	It is necessary for all running batch processes to complete before starting the server shutdown. In many cases, it is better to let a batch process complete without attempting to terminate the process. However, as some batch processes can take significant amount of time to complete, potentially hours, you often need to make an individual determination for each batch process. See About planned server shutdowns for more information.
3.	Work queue	It is necessary to stop all work queue workers on the server that you intend to shut down. As PolicyCenter distributes work queues across cluster members, stopping workers on one server does not typically affect the cluster. However, it is possible for the overall performance of work items processing to drop. Stop work queues after stopping batch processing to stop generating business transactions that can potentially generated messages.
4.	Messaging destinations	It is necessary to stop messages destinations on the server that you intend to shut down. Other servers in the cluster must take over the stopped message destination.
5.	Startable plugins	It is necessary to stop startable plugins (services) on the server that you intend to shut down. Other servers in the cluster must take over the stopped services.

Most background tasks, except batch processes, stop quickly as their units of work are small. The actual task managers, for example the Batch Process Manager of the Message Destination Manager, do not instantly stop in a server shutdown. Instead, each lease manager moves to a passive mode in which it does not start new background tasks and moves to stop or complete any currently running tasks.

After all components stop their background tasks, you can shut down the batch server safely.

Server shutdown monitoring

As it may take some time to stop all background tasks, you can use the following API method to track the background task shutdown progress:

ISystemTools.getClusterState

You can also use the following system_tools command options to gather information about a server and the state of the server components:

system_tools -components
system_tools -nodes

See system_tools command options for a discussion of these command options.