None of the above mention the cluster structure attributes that should be taken into consideration in order to decide how to allocate dedicated master nodes.
From these articles I understand that:
Master nodes should avoid having a tie - an odd number of nodes can solve this.
The amount of RAM can range from 512MB to 4GB, what affects this configuration?
How much block-storage should be allocated? and what type?
You want 3 master nodes regardless of your setup. You can start with something like 1 dedicated master and 2 data+master nodes. A dedicated master node should have around 8GB to 16GB of physical memory with a heap size of 75% of the physical memory. At minimum you would want at least a 4GB java heap. If you grow beyond this configuration, (after adding dedicated data nodes) then you should have 3 dedicated master nodes. The master node doesn't need much physical storage since it only stores the cluster state.
What are Master Nodes doing that they require gigabytes of memory? I thought they merely need to keep track of cluster state, which should be in the kilobyte range rather than the Gigabyte, no? Also, do they tend to be more CPU intensive or more memory intensive? How many CPUs should they have?
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.