LATEST VERSION: 9.2.0 - CHANGELOG
Pivotal GemFire® v9.2

Overview of the Cluster Configuration Service

The Pivotal GemFire cluster configuration service persists cluster configurations created by gfsh commands to the locators in a cluster and distributes the configurations to members of the cluster.

Why Use the Cluster Configuration Service

We highly recommend that you use the gfsh command line and the cluster configuration service as the primary mechanism to manage your distributed system configuration. Using a common cluster configuration reduces the amount of time you spend configuring individual members and enforces consistent configurations when bringing up new members in your cluster. You no longer need to reconfigure each new member that you add to the cluster. You no longer need to worry about validating your cache.xml file. It also becomes easier to propagate configuration changes across your cluster and deploy your configuration changes to different environments.

You can use the cluster configuration service to:

  • Save the configuration for an entire Pivotal GemFire cluster.
  • Restart members using a previously-saved configuration.
  • Export a configuration from a development environment and migrate that configuration to create a testing or production system.
  • Start additional servers without having to configure each server separately.
  • Configure some servers to host certain regions and other servers to host different regions, and configure all servers to host a set of common regions.

Using the Cluster Configuration Service

To use the cluster configuration service in GemFire, you must use dedicated, standalone locators in your deployment. You cannot use the cluster configuration service with co-located locators (locators running in another process such as a server) or in multicast environments.

The standalone locators distribute configuration to all locators in a cluster. Every locator in the cluster with --enable-cluster-configuration set to true keeps a record of all cluster-level and group-level configuration settings.

Note: The default behavior for gfsh is to create and save cluster configurations. You can disable the cluster configuration service by using the --enable-cluster-configuration=false option when starting locators.

Subsequently, any servers that you start with gfsh that have --use-cluster-configuration set to true will pick up the cluster configuration from the locator as well as any appropriate group-level configurations (for member groups they belong to). To disable the cluster configuration service on a server, you must start the server with the --use-cluster-configuration parameter set to false. By default, the parameter is set to true.

You can also load existing configuration files into the cluster configuration service by starting up a standalone locator with the parameter --load-cluster-configuration-from-dir set to true. See Loading Existing Configuration Files into Cluster Configuration.

How the Cluster Configuration Service Works

When you use gfsh commands to create Pivotal GemFire regions, disk-stores, and other objects, the cluster configuration service saves the configurations on each locator in the cluster (also called a GemFire distributed system). If you specify a group when issuing these commands, a separate configuration is saved containing only configurations that apply to the group.

When you use gfsh to start new Pivotal GemFire servers, the locator distributes the persisted configurations to the new server. If you specify a group when starting the server, the server receives the group-level configuration in addition to the cluster-level configuration. Group-level configurations are applied after cluster-wide configurations; therefore you can use group-level to override cluster-level settings.

gfsh Commands that Create Cluster Configurations

The following gfsh commands cause the configuration to be written to all locators in the cluster (the locators write the configuration to disk):

  • configure pdx*
  • create region
  • alter region
  • alter runtime
  • destroy region
  • create index
  • destroy index
  • create disk-store
  • destroy disk-store
  • create async-event-queue
  • deploy jar
  • undeploy jar

* Note that the configure pdx command must be executed before starting your data members. This command does not affect any currently running members in the system. Data members (with cluster configuration enabled) that are started after running this command will pick up the new PDX configuration.

The following gateway-related commands use the cluster configuration service, and their configuration is saved by locators:

  • create gateway-sender
  • create gateway-receiver

gfsh Limitations

There are some configurations that you cannot create using gfsh, and that you must configure using cache.xml or the API:

  • Client cache configuration
  • You cannot directly modify the attributes of the following objects:

    • function
    • custom-load-probe
    • cache-listener
    • cache-loader
    • cache-writer
    • compressor
    • serializer
    • instantiator
    • pdx-serializer

      Note: The configure pdx command always specifies the org.apache.geode.pdx.ReflectionBasedAutoSerializer class. You cannot specify a custom PDX serializer in gfsh.

    • custom-expiry

    • initializer

    • declarable

    • lru-heap-percentage

    • lru-memory-size

    • partition-resolver

    • partition-listener

    • transaction-listener

    • transaction-writer

  • Adding or removing a TransactionListener

  • Adding JNDI bindings

  • Deleting an AsyncEventQueue

In addition, there are some limitations on configuring gateways using gfsh. You must use cache.xml or the Java APIs to configure the following:

  • Configuring a GatewayConflictResolver
  • You cannot specify parameters and values for Java classes for the following:
    • gateway-listener
    • gateway-conflict-resolver
    • gateway-event-filter
    • gateway-transport-filter
    • gateway-event-substitution-filter

Disabling the Cluster Configuration Service

If you do not want to use the cluster configuration service, start up your locator with the --enable-cluster-configuration parameter set to false or do not use standalone locators. You will then need to configure the cache (via cache.xml or API) separately on all your distributed system members.