Repairs ranges repeated for each node #1089
Description
Looking at the range repairs done during a VNODE repair I have noticed that it does the repair for all ranges the node contains, not just the primary range.
So if a keyspace has a replication factor of 3, then each range will be repaired 3 times during a repair cycle of all the nodes on the cluster. This seems like overkill.
The more normal way to do a repair across a cluster is for primary ranges only, using the -pr option within nodetool, this will ensure each range is repaired only once, per cycle.
One way to do this would just do repairs for primary repairs for each node, however that may cause another inefficiency as described by Tommy:
"For scheduled repair: as long as we make sure a range is only repaired once. If we do that by doing primary ranges only or by making sure each node knows if the range has been repaired by another node doesn't matter that much to me. But I think ecchronos has a feature to combine ranges next to each other which make repair more efficient in large cluster and I think that will be much more useful if don't do only primary."
So further investigation should be made as to which way to fix this is best/practical.