Details
-
Improvement
-
Resolution: Unresolved
-
Minor
-
4.1.0, 4.5.0
-
Security Level: Public
Description
Currently when a disk backfill is scheduled, we backfill all (most) items on disk irrespective of end seqno requested.
Though functionally not incorrect, it is an overkill (and hence not best performance wise) to backfill all items.
It would be more efficient only to backfill till the snapshot_end close to endSeqno
Example:
Client requests from seqno 1 to seqno 500 from disk.
Say the disk has 2 snapshots one from 1 to 600 and another from 601 to 700. Currently, as I understand we send backfill snapshot from 1 to 700.
It would be more efficient to send the backfill snapshot from 1 to 600.