Yarn : Do we need nodemanager on namenode?

Question

Do we need to start nodemanager on a node which is namenode i.e i wont make it datanode? I will be starting namenode and resourcemanager on a machine. Since I wont start datanode on this machine, I think I dont need to start nodemanager as well ?

Thank

Answer 1

It depends on you. You can opt for keeping a Nodemanager on a node where Datanode service is not installed and in this case data will travel over the network for task execution if executed by this Nodemanager.

But if you want to reduce load from the Master Node then you should avoid installing such heavy process on it (Datanode and NodeManager). In Many cases it improves the performance of the Cluster.

Answer 2

Theoretically , you can have Nodemanager on a machine where you DO NOT have datanode but this is not useful practically .

Hadoop has one fundamental rule : Data Locality So if you do not have datanode on the machine where you run nodemanager then system will move data to that node required for computation and it will slow down the process.

Ideally, both Nodemanager and Datanode should run on a machine.

Hope this helps