Why does Curator does Leader Election among 'Process' when ZooKeeper has a Leader Election among ZooKeeper Servers?

2
votes

I have recently learned about ZooKeeper and its design. I understand that a ZooKeeper service is supported by multiple ZooKeeper Servers, however, it is necessary to choose one of these server as the leader of the group.

Next I started going through Apache Curator recipes of LeaderLatch and LeaderElection, instead of talking about selecting a leader, they talk about selecting a process as the leader (organizer).

I am confounded by this. Can somebody help and clarify to me why the Curator recipe and ZooKeeper are talking about two different kinds of Leader?

And if indeed they are different, how do these leaders relate to each other?

1
javaapache-zookeeperapache-curator

1 Answers

3
votes

A user of ZooKeeper doesn’t care about ZooKeeper clusters having a leader. Its an implementation detail. ZooKeeper servers elect a leader among themselves as a way to enforce some of the consistency guarantees ZooKeeper as a service offers.

You as a user of ZooKeeper (or user of Curator as a nicer API) don’t really care about that. In fact, ZK doesn’t expose this implementation detail - you don’t know if you are talking to a leader or a follower.

This ZK internal leader election has nothing to do with a use case ZooKeeper as a service offers - leader election for the users of ZooKeeper.

If you are using ZK, you are probably building a distributed service yourself, your service being distributed across several machines. As any other distributed system, you might face a problem at some point (e.g. task coordination) which is commonly solved by the leader election pattern:

In this case, you could go ahead and implement one of the listed algorithms yourself. Even better - you might use a coordination service like ZooKeeper. In ZK api you can implement leader election with sequential and ephemeral nodes. Take a look at this recipe:

As you see, the recipe is not trivial to implement using the bare ZK api, so Curator offers a nicer wrapper to make it easier for you.

In summary, when talking about Leader Election in ZooKeeper, there are two different things:

  • Leader election done internally by ZK servers among themselves, as a mechanism to implement strong consistency
  • Leader election as a feature provided to users of ZooKeeper, which needs to be either implemented manually using the ZK primitives, or used as an available recipe in the Curator library