Rora reads could keep replication in mind to avoid unnecessary fallback to proxy
Opened this issue · 3 comments
E.g. in case of replication on an all flash backend multiple fragments of the same chunk are equivalent.
The ocaml code knows about this, and can leverage that. The c++ code however will do a fallback to the proxy in case it can't read from the first fragment of a certain chunk (either because it isn't there, or because the asd is down).
The same applies for an accelerated backend with replication on the cache backend.
What about HPRM?
The pracc server uses the rora code inside, so same issue applies there.
Do note that given the repair-on-partial-read behaviour this isn't too bad...
(Also: how often will the fastest tier/cache use replication?)
(Also: how often will the fastest tier/cache use replication?)
Hence set to Roadmap ;-)