I'm developing now, but i still waiting for hardware to test a proper ES cluster (so I can not test this) but I would prefere to develope in the right way.
I will have few REST servlets serving results queried to a elasticsearch backend. Now, as REST based, user consequent requests may reach different servlets, that may be connected to different ES nodes.
If i start a scroll (giving the scroll_id to the user) will he be able to get results even if he access other nodes of the cluster or is a scrolling action linked to a specific node and therefore only same-node requests will work?
My scenario is where the user start scrolling the page and keeps downloading results from the initial query. From what I can think a scroll action would improve the ES performances as query was performed once, but in this scenario is really better over from and size?