Hello,
On a rather large architecture (about 32Tb of data x2 for the replica), we regularly update documents (need in terms of search engine so no regular ingestion, but rather updates of documents once a week).
We make one snapshot per index (we have three in all, so an index is about 10 TB).
The problem we encounter is that during a minimal update (1000 documents), we find ourselves having to re-snapshot a lot of data, which causes an extremely long backup time (6-7h). Compared to the number of modified documents, this seems like a lot.
I think that the problem comes from the segments that are certainly "merged" between the old snapshot, after the document update, and before the new snapshot.
Do you have any tips or solution to try to reduce this snapshot time as much as possible? Maybe we can play on the number of segments? Avoid a merge ? etc.
Thanks a lot.