I wanted to know if the snapshots always run from the nodes holding the primary shards of the indices which are being snapshotted?
Why I want to know this?
There is a setting of "max_snapshot_bytes_per_sec" which restricts the bandwidth utilisation per node. I have a restriction on the overall bandwidth being used by the entire cluster to be limited at 200MBPS. So I find out all the unique nodes which are holding the primary shards of all the indices being backed up and then set max_snapshot_bytes_per_sec as (200/NoOfNodes).
Wanted to know if this assumption is correct.