can single point of failure in Hadoop can be addressed by integrating Zookeeper?

Question

I am looking for addressing the issue of single point failure by integrating Zookeeper and allow Hadoop to use the namespace from Zookeeper? Is it possible? how can we address this ?

Answer 1

Sure, it might be technically possible, but this is not something that has been implemented in any shape or form just yet. You can't just plug it into ZooKeeper... it'll take some engineering effort to get it done (by you perhaps?!).

Here are a couple pointers to solutions of this:

• AvatarNode
• MapR (commercial software)
• A high-availability NameNode has been in the works and might end up soon in a future release