Revise replication logic. Currently, any error that is not NotFound leads to replication, even if the HEAD response:
- says
AccessDenied;
- could not answer at all until timeout expiration;
- *any other reason that says nothing about object presence*.
That leads to network operations that could be useless. Also if a cluster is under critical load, continuous replication made that even worse.
Maybe that is an arch issue and we should allow SN's HEADs always.
Revise replication logic. Currently, any error that is not
NotFoundleads to replication, even if theHEADresponse:AccessDenied;That leads to network operations that could be useless. Also if a cluster is under critical load, continuous replication made that even worse.
Maybe that is an arch issue and we should allow SN's
HEADs always.