There are lots of docs talking about the distribution of shards across nodes in a cluster, but very little about the actual usage of multiple data paths within nodes.
It is clear that a shard cannot be split over different paths, but WHEN and HOW does Elasticsearch (say 7.6) decide to create or to move a shard from one data path to another (across nodes, or within the same node)?
It is written that when a disk reaches a given watermark (e.g. 90%) the system decides to reallocate shards.
BUT does this include the possibility to reallocate them within the same node, i.e. moving it to a new empty disk, for instance?
Is it possible to tell Elastic to use a given data path when creating a new index, for instance?
Is there any clear document explaining these behaviours in detail?