Is it best practice to allow auto scale out/down ES cluster nodes?

I am building my Elasticsearch cluster on Azure VMs. I am thinking to enable auto scale out/in the VMs in the cluster based on the some performance throttle, such as CPU usage.
But my concern is, since the scale out/down VMs will add/remove VMs to/from cluster, it will introduce the shard re-balance which will have impact on performance.

Does anyone allow AUTO Scale on your ES nodes? Is it a good or bad idea?

I've seen this before, it's not super common but it makes sense when dealing with bursty loads.

The big thing you will need to watch, as you highlighted, is the cluster allocating things.

I have seen issues with auto-scaling as it is often difficult to get it to scale in/out at the right time. As scaling out adds load to the cluster due to data being reallocated, the worst time to do it is generally exactly when you need the extra performance.