Video: Configuring an auto-scalable cluster for Liberty elasticity

But you also have autonomic provisioning of new cluster members to other hosts.

Even if a host doesn't have Liberty installed, Liberty elasticity can autonomically install on that host what's called a stack of Liberty software--the Liberty runtime, an IBM JRE, and a Liberty server with your applications--and then autonomically start and stop those installed Liberty servers and their applications. This gives you greater capacity to handle spikes in resource demand.

What you see here is a scaling controller provisioning a stack group of Liberty software to hosts registered with the controller when resource use goes up, and then stopping or starting the provisioned cluster member servers and their applications in response to resource demands.

Show information Autonomic starting and stopping of cluster members based on scaling policies and resource use (JVM elasticity). Show animation of a scaling controller starting and stopping Liberty server cluster members on a registered host computer. When an arrow representing resource use points upward, show scaling member servers on the registered host going from stopped to running. Then, when the arrow representing resource use points downward, show scaling member servers on the registered host going from running to stopped.

Show information + Autonomic provisioning of new cluster members to other hosts. Show animation of a scaling controller provisioning new scaling members to two registered hosts and then starting and stopping those member servers based on the position of an arrow representing resource use.

Show a command window open at wlp/bin, and run a command that creates a server:

server create myController

collective create myController --hostName=controller_host_name --keystorePassword=myController

I add host="*" to the httpEndpoint element. This is optional.

To make myController a scaling controller, I add the scalingController-1.0 feature to a featureManager element.

I also add the adminCenter-1.0 feature so I can view my collective in Admin Center.

1. Add an administrative user name and password.

   <quickStartSecurity userName="adminUser" userPassword="adminPassword"/>

   Show information Add an administrative user name and password.

2. Add host="*" to the httpEndpoint element.
3. Add features to make myController a scaling controller and to enable administration of myController in Admin Center to a featureManager element.

   <feature>scalingController-1.0</feature>
   <feature>adminCenter-1.0</feature>

   Show information The scalingController-1.0 feature is required for a scaling controller and then The adminCenter-1.0 feature is optional for a scaling controller but needed for this example to use Admin Center.

4. Add a scaling policy that is disabled and has a minimum and maximum of one server.

   <scalingDefinitions>
      <defaultScalingPolicy enabled="false" min="1" max="1"/>
   </scalingDefinitions>

   Show information Add a scaling policy. Start with the scaling policy disabled.

5. Save the file.

cd /cygdrive/c/wlp/bin

./server start myController

Show information On Windows computers, using an SSH service to start servers enables a controller to connect to member servers with SSH.

If your browser prompts you to confirm that the connection is trusted, specify an exception or otherwise enable the connection to Admin Center.

In a web browser:

1. Open the browser on the https://controller_host_name:9443/adminCenter/ URL.
2. Select to trust the connection.
3. On the Admin Center login page, enter adminUser for the user name and adminPassword for the password. In the editor, highlight the user name and password in the scaling controller server.xml file at the same time to show needed login values.
4. On the Toolbox page, select .
5. In the Explore tool, select the Servers panel. The Explore tool shows one running server, myController.

Show command window open at wlp/bin, and run:

server create myMember

collective join myMember --host=controller_host_name --hostName=host_name --port=9443 --user=adminUser --password=adminPassword --keystorePassword=myMember

In this example, the controller and member are on the same host, so I need to change the port numbers. If your member server is on a different host from the controller, ensure the ports are unique on that host.

Add the scalingMember-1.0 feature to make the member server a scaling member.

Add a hostSingleton element and set its port to a number that is unique on the host. The host singleton port is used by all Liberty servers on the same host to elect a leader. The job of the leader is to report aggregated scaling metrics for all servers on the host. If the leader stops running, another leader is elected.

1. Set the RPC user and password, which is the user name and password that you enter to log on to the computer. The user must have administrative privileges on the computer.

   <hostAuthInfo rpcUser="admin_user_id"
                 rpcUserPassword="admin_user_password" />

2. Change the port numbers from 9080 to 9081 and 9443 to 9444 so that the member ports are different from the controller ports and from other ports used on the computer.

   <httpEndpoint id="defaultHttpEndpoint"
                 httpPort="9081"
                 httpsPort="9444" />

3. Make myMember a scaling member by adding scalingMember-1.0 to a featureManager element.

   <feature>scalingMember-1.0</feature>

   Show information The scalingMember-1.0 feature is required for a scaling member. You can add both clusterMember-1.0 and scalingMember-1.0 or add only scalingMember-1.0. You don't need clusterMember-1.0 because the scalingMember-1.0 feature enables the clusterMember-1.0 feature.

4. Add a hostSingleton element and set its port to 5164 or any other port that is unique on the computer.

   <hostSingleton name="ScalingMemberSingletonService" port="5164" />

   Show information A scaling member server.xml file must define a hostSingleton element with a port. All scaling members on the same host must use the same port. You can specify any port number, but the port number must be unique on the host computer. This example uses port number 5164.

5. Save the file.

./server start myMember

In the editor open on the controller server.xml file, enable auto scaling. Change:

<defaultScalingPolicy enabled="false" min="1" max="1"/>

<defaultScalingPolicy enabled="true" min="1" max="1"/>

On the Admin Center Explore tool Servers page, the auto scaling icon displays on the myMember card, indicating that the scaling policy is enabled.

<defaultScalingPolicy enabled="true" min="0" max="1"/>

Show information Setting the minimum to 0 (zero) stops the scaling member.

<defaultScalingPolicy enabled="true" min="1" max="1"/>

Show information Setting the minimum to 1 (one) restarts the scaling member.

<defaultScalingPolicy enabled="false" min="1" max="1"/>

Show information Testing confirms that auto scaling is working. For this example, temporarily disable scaling policies until the stack group and its installables and packages are defined..

If your collective has other dynamic cluster members, add the same statement to the other member server.xml files so they belong to the same cluster.

In the editor open on the member server.xml file, add a statement that puts the member in a cluster named myStackGroup.cluster1:

<clusterMember name="myStackGroup.cluster1"/>

Show information myStackGroup is the name of the stack group directory that will hold installables and packages for Liberty elasticity auto scaling and cluster1 is the name of the server package that the scaling controller will provision to registered hosts.

I run the collective registerHost command and specify a Linux® machine for the target host.

Show a command window open at wlp/bin, and then run a command to register the target host:

collective registerHost target_host_name --host=controller_host_name --port=9443 --user=adminUser --password=adminPassword --rpcUser=target_host_OS_user --rpcUserPassword=target_host_OS_user_password

Show information For this example, the registered host does not have Liberty or an IBM JRE that supports Liberty installed. Liberty elasticity auto scaling will need to provision to the registered host:

• Liberty runtime installable (wlp)
• JRE installable (wlp.jre)
• Server package (wlp.usr)

I copy the defaultStackGroup directory, and rename it.

Show a file system browser and move to the wlp/usr/shared/stackGroups directory. Copy the defaultStackGroup directory, paste it into the stackGroups directory, and rename the pasted directory myStackGroup. Then show the installables and packages subdirectories.

Show a file system browser and move to the wlp_installable directory. Show that the directory is empty.

server package --include=wlp --archive=c:\wlp_installable\wlp.855.zip

Move the archive to the stack group installables directory.

Show a file system browser and move to the java directory of a Liberty JRE installation. Show compressing the contents of the java directory to create jre.17.zip. Then copy the file and paste it into the wlp/usr/shared/stackGroups/myStackGroup/installables directory.

Show information: The JRE installable must be appropriate for the operating system of the registered hosts

Show a command window open at wlp/bin, and run:

server create cluster1

server package cluster1 --include=usr

Show information: For Windows registered hosts, create a server.env file that sets JAVA_HOME to the JRE location on the registered hosts. Place the server.env in the same directory as the server.xml in the server package ZIP. Example server.env file contents: JAVA_HOME=C:\wlp.jre\jre.17.zip\jre

In the editor open on the scaling controller server.xml file, add a statement that sets the administrative user name and password for the stack manager:

<stackManager controllerUser="adminUser" controllerUserPassword="adminPassword" />

After I save the changes, I wait for Admin Center to display the provisioned member server.

In the editor open on the scaling controller server.xml file, increase the min and max values from 1 to 2 and enable auto scaling by changing enabled from false to true:

<scalingDefinitions>
   <defaultScalingPolicy enabled="true" min="2" max="2"/>
</scalingDefinitions>

Then, save the changes.

Show information Provisioning installables and a server package to a registered host can take several minutes.

<scalingDefinitions>
   <defaultScalingPolicy enabled="true" min="1" max="2"/>
</scalingDefinitions>

Then, save the changes.

Show information This example sets min=1 and max=2 for the minimum and maximum number of member servers available for auto scaling. You can set a higher maximum, such as max=4, to enable additional cluster1 member servers to be provisioned on the registered host. The scaling controller can start provisioned member servers when resource use increases and stop the member servers when resource use decreases..

For more information about Liberty auto scaling, see:

WebSphere Application Server Liberty documentation on IBM Knowledge Center

http://www.ibm.com/support/knowledgecenter/