During load balancing, we use replica counts as an evaluation criterion for slice migration. However, the size of each slice is uncertain. A node may have fewer slices but larger disk load. If we perform load balancing at this time, it will cause more load pressure on this type of nodes.
In addition, the machines in a cluster may be of different models and have different disk sizes.
So this feature can help us to do more flexible load balancing.
Feature Request
Describe the feature you'd like:
When load balancing, we can set up a blacklist of nodes such that there will be no sharding in and out on those nodes.
Teachability, Documentation, Adoption, Migration Strategy:
During load balancing, we use replica counts as an evaluation criterion for slice migration. However, the size of each slice is uncertain. A node may have fewer slices but larger disk load. If we perform load balancing at this time, it will cause more load pressure on this type of nodes. In addition, the machines in a cluster may be of different models and have different disk sizes. So this feature can help us to do more flexible load balancing.