HelmaSwarm is a Helma extension that allows multiple instances of Helma to be combined into a cluster. It consists of three tools that plug into various parts of Helma:
- SwarmCache, which acts as a replacement object cache that propagates notifications of changed objects to other members in the swarm.
- SwarmSessionManager, which is a replacement for the Helma session manager that replicates sessions among swarm members and propagates session updates and expiry.
- SwarmIDGenerator, which is necessary to coordinate primary key generation for new persistent objects when the underlying database uses SELECT max(id) for creating primary keys (as is usually the case with MySQL). It is not needed if Sequences are used for key generation (as usually the case with Oracle, for instance).
HelmaSwarm uses JGroups <http://jgroups.org> for communication between Helma instances. This version of Helmaswarm comes bundled with JGroups 2.4.1-sp3.
This version of HelmaSwarm requires Helma 1.6.0-rc2 (Release Candidate 2) or later.
You can download the latest HelmaSwarm release from http://adele.helma.org/download/helma/contrib/helmaswarm/.
HelmaSwarm is built with Apache Ant.
- Edit build.properties to match your Helma installation directory.
- Run the following in the command line:
This should compile and build helmaswarm-version.jar and copy it to the lib/ext directory of your Helma installation along with the JGroups jar file.
To enable SwarmCache, SwarmSessionManager, and SwarmIDGenerator for a Helma application, add the respective lines to its app.properties file:
cacheImpl = helma.swarm.SwarmCache sessionManagerImpl = helma.swarm.SwarmSessionManager idGeneratorImpl = helma.swarm.SwarmIDGenerator
HelmaSwarm uses a group name to identify and connect to a particular swarm. By default, the application name is used as the group name. If you want use a different group name, for instance because your swarm is made up of applications with different names, you can set the HelmaSwarm group name with the swarm.name entry in app.properties:
swarm.name = mySwarmName
HelmaSwarm uses an XML configuration file from which it reads its properties. This file is called swarm.conf and is either set by just copying it to the application directory, or by setting the helma.conf app property:
swarm.conf = /path/to/helmaswarm/swarm.conf
The most important setting in helma.conf is the JGroups network stack. By default, HelmaSwarm uses a UDP multicast stack called "udp". helma.swarm also contains a TCP stack. The JGroups stack is configured with the following app property:
swarm.jgroups.stack = [udp|tcp|custom]
The default UDP multicast stack uses port 22024 on multicast address 18.104.22.168. It is advisable to use a different setting if multiple swarm instances are operated on the same local network to avoid unnecessary network traffic.
HelmaSwarm uses asynchronous communication for keeping consistent state among Helma instances. This means that messages are sent without the sender waiting for confirmation of receipt by other swarm members. While this greatly reduces group communication overhead and complexity, it makes it possible that a client that has altered some state on one Helma instance may still see the old state in a subsequent request to another swarm member, if that request gets ahead of the swarm notification of the state change.
For this reason it is advisable (although not strictly necessary) to use sticky sessions with HelmaSwarm clusters. See doc/README-Apache.txt for information on how to implement sticky sessions with Apache and mod_jk. Round Robin DNS should minimize the risk of this scenario, since DNS names aren't refreshed that often, and if they are chances should be small that DNS lookup and HTTP request gets ahead of HelmaSwarm communication on the server LAN.
HelmaSwarm is written by Hannes Wallnoefer (hannes at helma dot at). This sofware was initially inspired by SwarmCache <http://swarmcache.sf.net/>, from which it borrows part of its name.