I have a couple of questions about coordinating nodes:
- What are the minimum resources (RAM, CPU, disk) for a coordinating node?
- What do I need to do in order to estimate the resources for this kind of node correctly?
Thank you in advance,
You might not be able to find the official minimum resources for a coordinating node, because often the answer is it depends. The documentation does say you'd need to worry more about RAM and CPU than disk for coordinating nodes. Disk should be negligable since by definition it won't be storing anything elastic related that is. I'd try some settings out on a coordinating node and see how it performs. For example: 4GB ram (for elastic, 8GB for host overall), 2.XGhz 4 core enterprise grade cpu.
An elastic blog post about that subject can be found here.
I have seen a lot of clusters where there have been no need for dedicated coordinating only nodes at all, so the first thing I would like to understand is the use case and what problem you are trying to solve by adding them. Can you please elaborate on this?
We have deployed a cluster with 5 nodes. We have seen that those nodes have a high usage of CPU and RAM. We would like to use Coordinating nodes in order to reduce the usage of CPU and RAM.
I have a coordinating node in a production cluster that is working well. The nice thing is it doesn't hurt to try and see what works for your unique situation. Good luck!
Have you tracked the root cause of the high CPU and RAM usage using hot threads?
What is the configuration of your nodes? Do you have 3 nodes dedicates and 2 data nodes or all 5 nodes can be master/data?
A 5 node cluster seems too small to need a coordinating only node, but if you want to add one you could create a node with the same CPU/Ram configuration of your data nodes.
All 5 nodes can be master and they are data too. We have found out that nodes are continuously doing merging index, and disk access time increase significantly. As a result, there is an increase of CPU and RAM usage.
If merging is causing issues it sounds like you may be suffering from slow storage. What type of storage are you using?
If the tasks that would be taken over by added coordinating-only nodes are not visible in the hot threads I am not sure adding dedicated coordinating-only nodes will make much difference. Fell free to try though as it is hard to know for sure.
We are using Azure Premium SSD LRS (P40 - 7500IOPS - 250MB/s). However, after increasing shards number, the issue is mitigated.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.