Thursday, June 13, 2013

Leader election of Camel router through Zookeeper

Most modern distributed systems somewhere comes across the problem where only one instance of the process or job should runs and others instance of the same job or process should be standby. Apache zookeeper project provides this above functionality out of box. I am not going to describe all the usecases or solutions that could be solved by the Zookeeper. You can read the following great article for quick start.

Let me describe our use case with more details, we have apache camel component which polling some resources and only one instance of the component should run in any time. It means there should be one leader and a few followers. When master or leader goes down, any follower should replace the master. You also can use doozerd instead of zookeeper for leader election, but for me zookeeper is easy to maintain.

It's very easy to install a multi server zookeeper cluster. I have recommend you to follow the step from this blog to setup a cluster. Now when we have done our cluster setup we are ready to go for camel.
For leader election in camel we can apply ZooKeeperRoutePolicy in camel route component as follows:
ZooKeeperRoutePolicy zooKeeperRoutePolicy = new ZooKeeperRoutePolicy("zookeeper://172.18.5.36:2182/ELECTION", 1);
for details please see the link.
Once the policy is defined, it could be plugged into route as follows:
from("quartz://zookeeper/heartbeat?cron=0/50+*+*+*+*+?")
    .routePolicy(zooKeeperRoutePolicy)
    .to("jms:queue:xyz");
Above route will be execute every 50 seconds and will send message to queue xyz if it could be bind the relevant znode. Whats going on under the hood. Assume we have the znode named ELECTION in our cluster. When we have execute our first instance of this camel route, it will creates a Ephemeral znode as follows:
[zk: 172.18.5.36:2182(CONNECTED) 0] ls /ELECTION
[602-bsha-154fcb3e-2768-4d24-9803-92b9d2d9c77c0000000005]
when camel route will disconnect the znodes will be deleted automatically. If we we have started another instance of this camel route we will go following znodes:
[zk: 172.18.5.36:2182(CONNECTED) 0] ls /ELECTION
[602-bsha-61918db4-6ebe-4283-b381-c3c4688920f10000000006, 602-bsha-154fcb3e-2768-4d24-9803-92b9d2d9c77c0000000005]
By the zookeeper leader election algorithm the ephemeral znodes 602-bsha-154fcb3e-2768-4d24-9803-92b9d2d9c77c0000000005 will be the leader because it's has the smallest sequence number.
If we will stop the first camel route, 602-bsha-61918db4-6ebe-4283-b381-c3c4688920f10000000006 znode will be the leader and second camel route will start polling.
Now imagine what happens, if the first zookeeper server failed? our first camel router would be also failed because its only connected with one zookeeper server. For high availability if we want to connect to more zookeeper server from camel ZookeeperRoutePolicy we have to add more server ip address and port to the configuration. But camel ZookeeperRoutePolicy doesn't provide this functionality out of box. One of my college found a very quick fix of this problem by using ZooKeeperConfiguration class.
We have to first add the bean in spring context, it could be autowired.
<bean id="zookeeperComponent" class="org.apache.camel.component.zookeeper.ZooKeeperComponent"/>
After that we can add ZooKeeperConfiguration properties to the above component as follows:
ZooKeeperConfiguration zooConfig = new ZooKeeperConfiguration();
        zooConfig.addZookeeperServer("172.18.5.36:2181");
        zookeeperComponent.setConfiguration(zooConfig);
It's provide camel route component to bind with another zookeeper server for getting high availability. If one of the zookeeper server will fail, camel route component will still running.

2 comments:

Nagarajan said...

Can you please let me know how zookeeperComponent is used tied to the zookeeperRoutePolicy?

Nagarajan said...

Can you please let me know how zookeeperComponent is used tied to the zookeeperRoutePolicy?