Partial read optimisation in case of replication
Closed this issue · 1 comments
domsj commented
We needlessly differentiate between the first fragment and any additional fragments (which are in fact exactly the same).
When the first fragment is unavailable we'll fall back to rebuilding the chunk (aka fetching another copy of the fragment entirely) while we could just do a partial read from another copy of the fragment.
With the repair on partial read path this situation will quickly be resolved, but it still costs more than it should.