@Thejas-bhat For couchbase it seems if the node were to go down during this situation, KV would ask us to rollback yes?
And I suppose this is related to the concurrent persister model that we just introduced?

nope i don't think so - this is because bolt separately stores the high seqno. at vBucket level inside the index snapshot bucket of bolt which is tracked correctly. However, the segment snapshot bucket won't be filled with all the file segment's entries. So when the node goes down and comes back up, the seq no.s are correct (so no rollback) however the loadSegment() itself will be applied only to a subset of file segments that should've been there

I see, and does this come into play because of the concurrent persister model or has it been always around? Do we need to backport this asap?

i dont think we have to backport it because this is a bug that got introduced with the concurrent persister PR (finally a better name for it) that got merged recently in 2.5.0

Thanks, I've updated the commit description^.