Details
-
Bug
-
Resolution: Fixed
-
Major
-
7.6.0
-
Triaged
-
0
-
Unknown
Description
The RR% guard rail includes a check on rebalance to avoid crossing the guard rail limit when it is expected to be breached by the rebalance (MB-57365). We should do similar checks for the other guard rails.
- Max number of buckets - We should check that the cpu count of added nodes does not violate the min cores per buckets guard rail
- Max data per node - We should be able to use a similar estimation method to the RR% check, blocking a rebalance when it is expected to breach the max data per node limit
- Disk space - I think that it's reasonable to block topology changes that net-remove nodes (as discussed in
MB-58863) when the disk usage limit is already reached on a node, although when a customer has a node-specific disk issue they will need to failover the node in order to be allowed to eject it. There are other cases we should consider handling (swap rebalance, net-removing nodes before the disk limit is reached, etc), but this requires further thought