2013/06/11

Clustering in jBPM v6


Clustering in jBPM v5 was not an easy task, there where several known issues that had to be resolved on client (project that was implementing solution with jBPM) side, to name few:

  • session management - when to load/dispose knowledge session
  • timer management - required to keep knowledge session active to fire timers
This is not the case any more in version 6 where several improvements made their place in code base, for example new module that is responsible for complete session management was introduced - jbpm runtime manager. More on runtime manager in next post, this one focuses on how clustered solution might look like. First of all lets start with all the important pieces jbpm environment consists of:


  1. asset repository - VFS based repository backed with GIT - this is where all the assets are stored during authoring phase
  2. jbpm server that includes JBoss AS7 with deployed jbpm console (bpm focused web application) of kie-wb (fully features web application that combines BPM and BRM worlds)
  3. data base - backend where all the state date is kept (process instances, ksessions, history log, etc)

Repository clustering

Asset repository is GIT backed virtual file system (VFS) that keeps all the assets (process definitions, rules, data model, forms, etc) is reliable and efficient way. Anyone who used to work with GIT understands perfectly how good it is for source management and what else assets are if not source code?
So if that is file system it resides on the same machine as the server that uses it, that enforces it to be kept in sync between all servers of a cluster. For that jbpm makes use of two well know open source projects:

Zookeeper is responsible for gluing all parts together where Helix is cluster management component that registers all cluster details (cluster itself, nodes, resources).

So this two components are utilized by the runtime environment on which jbpm v6 is based on:
  • kie-commons - provides VFS implementation and clustering 
  • uber fire framework - provides backbone of the web applications

So let's take a look at what we need to do to setup cluster of our VFS:

Get the software

  • download Apache Zookeeper (note that 3.3.4 and 3.3.5 are the only versions that were currently tested so make sure you get the correct version)
  • download Apache Helix  (note that version that was tested was 0.6.1)

Install and configure

  • unzip Apache Zookeeper into desired location - ( from now one we refer to it as zookeeper_home)
  • go to zookeeper_home/conf and make a copy of zoo_sample.conf to zoo.conf
  • edit zoo.conf and adjust settings if needed, these two are important in most of the cases:
# the directory where the snapshot is stored.
dataDir=/tmp/zookeeper
# the port at which the clients will connect
clientPort=2181

  •  unzip Apache helix to into desired location (from now one we refer to it as helix_home)

Setup cluster

Now we have all the software available locally so next step is to configure the cluster itself. We start with start of the Zookeeper server that will be master of the configuration of the cluster:
  • go to zookeeper_home/bin
  • execute following command to start zookeeper server:
sudo ./zkServer.sh start
  • zookeeper server should be started, if the server fails to start make sure that the data directory defined in zoo.conf file exists and is accessible
  • all zookeeper activities can be viewed zookeeper_home/bin/zookeeper.out
To do so, Apache Helix provides utility scripts that can be found in helix_home/bin.

  • go to helix_home/bin
  • create cluster
./helix-admin.sh --zkSvr localhost:2181 --addCluster jbpm-cluster
  • add nodes to the cluster 
node 1
./helix-admin.sh --zkSvr localhost:2181 --addNode jbpm-cluster nodeOne:12345
node2
    ./helix-admin.sh --zkSvr localhost:2181 --addNode jbpm-cluster nodeTwo:12346
add as many nodes as you will have cluster members of jBPM server (in most cases number of application servers in the cluster)
NOTE: nodeOne:12345 is the unique identifier of the node, that will be referenced later on when configuring application severs, although it looks like host and port number it is use to identify uniquely logical node.
  • add resources to the cluster
./helix-admin.sh --zkSvr localhost:2181 
           --addResource jbpm-cluster vfs-repo 1 LeaderStandby AUTO_REBALANCE
  • rebalance cluster to initialize it
./helix-admin.sh --zkSvr localhost:2181 --rebalance jbpm-cluster vfs-repo 2

  • start the Helix controller to manage the cluster
./run-helix-controller.sh --zkSvr localhost:2181 
                        --cluster jbpm-cluster 2>&1 > /tmp/controller.log &
Values given above are just examples and can be changed according to the needs:
cluster name: jbpm-cluster
node name: nodeOne:12345, nodeTwo:12346
resource name: vfs-repo
zkSvr value must match Zookeeper server that is used.

Prepare data base 


Before we start with application server configuration data base needs to be prepared, for this example we use PostgreSQL data base. jBPM server will create all required tables itself by default so there is no big work required for this but some simple tasks must be done before starting the server configuration.

Create data base user and data base

First of all PostgreSQL data base needs to be installed, next user needs to be created on the data base that will own the jbpm schema, in this example we use:
user name: jbpm
password: jbpm

Once the user is ready, data base can be created, and again for the example jbpm is chosen for the data base name.

NOTE: this information (username, password, data base name) will be used later on in application server configuration.

Create Quartz tables

Lastly Quartz related tables must be created, to do so best is to utilize the data base scripts provided with Quartz distribution, jbpm uses Quartz 1.8.5. DB scripts are usually located under QUARTZ_HOME/docs/dbTables.

Create quartz definition file 

Quartz configuration that will be used by the jbpm server needs to accomodate the needs of the environment, as this guide is about to show the basic setup obviously it will not cover all the needs but will allow for further improvements.

Here is a sample configuration used in this setup:
#============================================================================
# Configure Main Scheduler Properties  
#============================================================================

org.quartz.scheduler.instanceName = jBPMClusteredScheduler
org.quartz.scheduler.instanceId = AUTO

#============================================================================
# Configure ThreadPool  
#============================================================================

org.quartz.threadPool.class = org.quartz.simpl.SimpleThreadPool
org.quartz.threadPool.threadCount = 5
org.quartz.threadPool.threadPriority = 5

#============================================================================
# Configure JobStore  
#============================================================================

org.quartz.jobStore.misfireThreshold = 60000

org.quartz.jobStore.class=org.quartz.impl.jdbcjobstore.JobStoreCMT
org.quartz.jobStore.driverDelegateClass=org.quartz.impl.jdbcjobstore.PostgreSQLDelegate
org.quartz.jobStore.useProperties=false
org.quartz.jobStore.dataSource=managedDS
org.quartz.jobStore.nonManagedTXDataSource=notManagedDS
org.quartz.jobStore.tablePrefix=QRTZ_
org.quartz.jobStore.isClustered=true
org.quartz.jobStore.clusterCheckinInterval = 20000

#============================================================================
# Configure Datasources  
#============================================================================
org.quartz.dataSource.managedDS.jndiURL=jboss/datasources/psjbpmDS
org.quartz.dataSource.notManagedDS.jndiURL=jboss/datasources/quartzNotManagedDS


Configure JBoss AS 7 domain


1. Create JDBC driver module - for this example PostgreSQL
a) go to JBOSS_HOME/modules directory (on EAP JBOSS_HOME/modules/system/layers/base)
b) create module folder org/postgresql/main
c) copy postgresql driver jar into the module folder (org/postgresql/main) as postgresql-jdbc.jar          name
d) create module.xml file inside module folder (org/postgresql/main) with following content:
         <module xmlns="urn:jboss:module:1.0" name="org.postgresql">
         <resources>
           <resource-root path="postgresql-jdbc.jar"/>
         </resources>

             <dependencies>
      <module name="javax.api"/>
      <module name="javax.transaction.api"/>
         </dependencies>
</module>

2. Configure data sources for jbpm server
a) go to JBOSS_HOME/domain/configuration
b) edit domain.xml file
for simplicity sake we use default domain configuration which uses profile "full" that defines two 
        server nodes as part of main-server-group
c) locate the profile "full" inside the domain.xml file and add new data sources
main data source used by jbpm
   <datasource jndi-name="java:jboss/datasources/psjbpmDS" 
                pool-name="postgresDS" enabled="true" use-java-context="true">
        <connection-url>jdbc:postgresql://localhost:5432/jbpm</connection-url>
        <driver>postgres</driver>
        <security>
            <user-name>jbpm</user-name>
                <password>jbpm</password>
        </security>
   </datasource>
        
        additional data source for quartz (non managed pool)
        <datasource jta="false" jndi-name="java:jboss/datasources/quartzNotManagedDS"   
           pool-name="quartzNotManagedDS" enabled="true" use-java-context="true">
      <connection-url>jdbc:postgresql://localhost:5432/jbpm</connection-url>
        <driver>postgres</driver>
        <security>
         <user-name>jbpm</user-name>
                <password>jbpm</password>
        </security>
    </datasource>
defined the driver used for the data sources
<driver name="postgres" module="org.postgresql">
      <xa-datasource-class>org.postgresql.xa.PGXADataSource</xa-datasource-class>
    </driver>
        
3. Configure security domain 
     a) go to JBOSS_HOME/domain/configuration
     b) edit domain.xml file
for simplicity sake we use default domain configuration which uses profile "full" that defines two
        server nodes as part of main-server-group
      
     c) locate the profile "full" inside the domain.xml file and add new security domain to define security 
         domain for jbpm-console (or kie-wb) - this is just a copy of the "other" security domain defined 
         there by default
    
<security-domain name="jbpm-console-ng" cache-type="default"> <authentication>
            <login-module code="Remoting" flag="optional">
               <module-option name="password-stacking" value="useFirstPass"/>
            </login-module>
            <login-module code="RealmDirect" flag="required">
                  <module-option name="password-stacking" value="useFirstPass"/>
            </login-module>
        </authentication>
    </security-domain>  
        
for kie-wb application, simply replace jbpm-console-ng with kie-ide as name of the security domain.  
        
4. Configure server nodes

    a) go to JBOSS_HOME/domain/configuration
    b) edit host.xml file
    c) locate servers that belongs to "main-server-group" in host.xml file and add following system  
        properties:
    


property nameproperty valuecomments
org.uberfire.nio.git.dir/home/jbpm/node[N]/repolocation where the VFS asset repository will be stored for the node[N]
org.quartz.properties/jbpm/quartz-definition.propertiesabsolute file path to the quartz definition properties
jboss.node.namenodeOneunique node name within cluster (nodeOne, nodeTwo, etc)
org.uberfire.cluster.idjbpm-clustername of the helix cluster
org.uberfire.cluster.zklocalhost:2181location of the zookeeper server
org.uberfire.cluster.local.idnodeOne_12345unique id of the helix cluster node, note that ':' is replaced with '_'
org.uberfire.cluster.vfs.lockvfs-reponame of the resource defined on helix cluster
org.uberfire.nio.git.daemon.port9418port used by the GIT repo to accept client connections, must be unique for each cluster member
org.uberfire.nio.git.ssh.port8001port used by the GIT repo to accept client connections (over ssh), must be unique for each cluster member
org.uberfire.nio.git.daemon.hostlocalhosthost used by the GIT repo to accept client connections, in case cluster members run on different machines this property must be set to actual host name instead of localhost otherwise synchronization won't work
org.uberfire.nio.git.ssh.hostlocalhosthost used by the GIT repo to accept client connections (over ssh), in case cluster members run on different machines this property must be set to actual host name instead of localhost otherwise synchronization won't work
org.uberfire.metadata.index.dir/home/jbpm/node[N]/indexlocation where index for search will be created (maintained by Apache Lucene)
org.uberfire.cluster.autostartfalsedelays VFS clustering until the application is fully initialized to avoid conficts when all cluster members create local clones
    
examples for the two nodes:
    
  •     nodeOne
<system-properties>
  <property name="org.uberfire.nio.git.dir" value="/tmp/jbpm/nodeone" 
                                       boot-time="false"/>
  <property name="org.quartz.properties" 
      value="/tmp/jbpm/quartz/quartz-db-postgres.properties" boot-time="false"/>
  <property name="jboss.node.name" value="nodeOne" boot-time="false"/>
  <property name="org.uberfire.cluster.id" value="jbpm-cluster" 
                                           boot-time="false"/>
    <property name="org.uberfire.cluster.zk" value="localhost:2181" 
                                           boot-time="false"/>
  <property name="org.uberfire.cluster.local.id" value="nodeOne_12345" 
                                                 boot-time="false"/>
  <property name="org.uberfire.cluster.vfs.lock" value="vfs-repo" 
                                                 boot-time="false"/>
  <property name="org.uberfire.nio.git.daemon.port" value="9418" boot-time="false"/>
  <property name="org.uberfire.metadata.index.dir" value="/tmp/jbpm/nodeone" boot-time="false"/>
  <property name="org.uberfire.cluster.autostart" value="false" boot-time="false"/>
</system-properties>
    
  •     nodeTwo
<system-properties>
    <property name="org.uberfire.nio.git.dir" value="/tmp/jbpm/nodetwo" 
                                         boot-time="false"/>
    <property name="org.quartz.properties" 
       value="/tmp/jbpm/quartz/quartz-db-postgres.properties" boot-time="false"/>
    <property name="jboss.node.name" value="nodeTwo" boot-time="false"/>
    <property name="org.uberfire.cluster.id" value="jbpm-cluster" 
                                             boot-time="false"/>
    <property name="org.uberfire.cluster.zk" value="localhost:2181" 
                                             boot-time="false"/>
    <property name="org.uberfire.cluster.local.id" value="nodeTwo_12346" 
                                                   boot-time="false"/>
    <property name="org.uberfire.cluster.vfs.lock" value="vfs-repo" 
                                                   boot-time="false"/>
    <property name="org.uberfire.nio.git.daemon.port" value="9419" boot-time="false"/>
    <property name="org.uberfire.metadata.index.dir" value="/tmp/jbpm/nodetwo" boot-
     time="false"/>
    <property name="org.uberfire.cluster.autostart" value="false" boot-time="false"/>
</system-properties>

NOTE: since this example runs on single node host properties for ssh and git daemons are omitted.

Since repository synchronization is done between git servers make sure that GIT daemons are active (and properly configured - host name and port) on every cluster member.
5. Create user(s) and assign it to proper roles on application server

Add application users
In previous step security domain has been created so jbpm console (or kie-wb) users could be authenticated while logging on. Now it's time to add some users to be able to logon to the application once it's deployed. To do so:
 a) go to JBOSS_HOME/bin
 b) execute ./add-user.sh script and follow the instructions on the screen
  - use Application realm not management
  - when asked for roles, make sure you assign at least:
  for jbpm-console: jbpm-console-user
  for kie-wb: kie-user
 
add as many users you need, same goes for roles, listed above are required to be authorized to use the web application. 

Add management (of application server) user
To be able to manage the application server as domain, we need to add administrator user, it's similar to what was defined for adding application users but the realm needs to be management
 a) go to JBOSS_HOME/bin
 b) execute ./add-user.sh script and follow the instructions on the screen
  - use Management realm not application

Application server should be now ready to be used, so let's start the domain:

JBOSS_HOME/bin/domain.sh

after few seconds (it's still empty servers) you should be able to access both server nodes on following locations:
administration console: http://localhost:9990/console

the port offset is configurable in host.xml for given server.


Deploy application - jBPM console (or kie-wb)

Now it's time to prepare and deploy application, either jbpm-console or kie-wb. As by default both application comes with predefined persistence that uses ExampleDS from AS7 and H2 data base there is a need to alter this configuration to use PostgreSQL data base instead.

Required changes in persistence.xml

  • change jta-data-source name to match one defined on application server
             java:jboss/datasources/psjbpmDS
  • change hibernate dialect to be postgresql 
             org.hibernate.dialect.PostgreSQLDialect

Application build from source

If the application is built from source then you need to edit persistence.xml file that is located under:
jbpm-console-ng/jbpm-console-ng-distribution-wars/src/main/jbossas7/WEB-INF/classes/META-INF/
next rebuild the jbpm-distribution-wars module to prepare deployable package - once that is named: 
jbpm-console-ng-jboss-as7.0.war

Deployable package downloaded

In case you have deployable package downloaded (which is already a war file) you need to extract it change the persistence.xml located under:
WEB-INF/classes/META-INF
once the file is edited and contains correct values to work properly with PostgreSQL data base application needs to be repackaged:
NOTE: before repackaging make use that previous war is not in the same directory otherwise it will be packaged into new war too.

jar -cfm jbpm-console-ng.war META-INF/MANIFEST.MF *

IMPORTANT: make sure that you include the same manifest file that was in original war file as it contains valuable entires.


To deploy application logon as management user into administration console of the domain and add new deployments using Runtime view of console. Once the deployment is added to the domain, assign it to the right server group - in this example we used main-server-group it will be default enable this deployment on all servers within that group - meaning deploy it on the servers. This will take a while and after successful deployment you should be able to access jbpm-console (or kie-wb) on following locations:


the context root (jbpm-console-ng) depends on the name of the war file that was deployed so if the filename will be jbpm-console-ng-jboss7.war then the context root will be jbpm-console-ng-jboss7. Same rule apply to kie-wb deployment.

And that's it - you should have fully operational jbpm cluster environment!!!

Obviously in normal scenarios you would like to hide the complexity of different urls to the application from end users (like putting in front of them a load balancer) but I explicitly left that out of this example to show proper behavior of independent cluster nodes.

Next post will go into details on how different components play smoothly in cluster, to name few:
  • failover - in case cluster node goes down
  • timer management - how does timer fire in cluster environment
  • session management - auto reactivation of session on demand
  • etc
As we are still in development mode, please share your thoughts on what would you like to see in cluster support for jBPM, your input is most appreciated!

Update:
There was a change in naming of system properties since the article was written so for those that configured it already for 6.0.0.Final there will be a need to adjust name of following system properties:

  • org.kie.nio.git.dir -> org.uberfire.nio.git.dir
  • org.kie.nio.git.daemon.port -> org.uberfire.nio.git.daemon.port
  • org.kie.kieora.index.dir -> org.uberfire.metadata.index.dir
  • org.uberfire.cluster.autostart - new parameter
Table above already contains proper values for 6.0.0.Final