Due to anormal objects placement we are going to use the rsync replication method to recover the situation. We are currently applying some configurations on the cluster to lower the impact on performances
Update(s):
Date: 2018-03-31 06:48:00 UTC with the last configuration changes, read requests were spread across 2 out of the 3 replicas. We updated this setting to spread these requests over the 3 replicas. This will give more resources for GET requests
Date: 2018-03-30 12:11:33 UTC We are switching back the replication method from rsync to ssync
Date: 2018-03-26 14:50:05 UTC misplaced objects are less than 2% now in the first replica, we should be able to swift rsync on another one tomorrow
Date: 2018-03-16 14:34:17 UTC we are modifying the affinity rules to lower the requests number on the replica running rsync
Date: 2018-03-16 14:11:54 UTC We switched the replication method to rsync on the first replica servers