Hi,
I am trying to understand the resource demand for pagination queries over large datasets using search_after and a Point-In-Time.
The application I work on exposes "datasets" to its customers. These datasets may be stored on our ES cluster in the same index (it is not easy to change this setup right now). We use aliases with a simple filter like dataset_id=N for that purpose.
I understand allocating a PIT on an index avoids purging old documents that are deleted after the PIT (and avoid segment merges to guarantee a constant shord_doc ?) but what does it do when created on an alias ?
If a PIT is created on an alias with a filter, will it "freeze" all documents of the index, or only those covered by the associated filter ?
Put differently: what is the cost of maintaining a PIT on a filtered alias that never changes when other documents on the same index (with a different dataset_id in our case) are deleted/updated ?
Thanks for your help