Description
Let's assume the following situation. On a cluster of 3 hosts, there are 3 mount-points. For example, Host 1, 2 and 3 (H1, H2, and H3) each have mounts /mldata1, /mldata2, and /mldata3 (M1, M2, and M3). For example, each mount is a separate GP2 filesystem on AWS. The forest assignment policy should multiplex the forests on hosts with the following constraints:
- for each host, primary forests should be balanced across the mounts (as much as possible).
- replica forests should be balanced across the remaining hosts, such a failover will spread the load across the cluster.
- on each host, the replicas should be balanced across the mount points.
For example, H1 has mounts M1, M2, and M3. With two forests per mount point, that should imply M1 has forests F1 and F2, M2 has forests F3 and F4 and M4 should have forests M5 and M6. H2 (which has Forests F7-F12), would also be spread over its three mounts, etc.
The replica for H1, M1, F1 might be on H2, M2. the replicat for H1, M1, F2 might be on H3, M3. The replicat for H1, M2, F3 might be on H1, M1. The replica for H1, M2, F4 might be on H3, M1, etc.