changefeed.balance_range_distribution.sensitivity isn't very useful. It's currently used to make rebalancing less uniform (by preserving the original plan provided by distsql).
This can be made more useful. For example, if the number of ranges is below X, rebalancing should put all the spans on one partition. This is a rough idea which needs design. There are more complications - for example, what happens when the table grows to a larger number of ranges? What if there are many changefeeds which all get planned on one node?
changefeed.balance_range_distribution.sensitivity
isn't very useful. It's currently used to make rebalancing less uniform (by preserving the original plan provided by distsql).This can be made more useful. For example, if the number of ranges is below X, rebalancing should put all the spans on one partition. This is a rough idea which needs design. There are more complications - for example, what happens when the table grows to a larger number of ranges? What if there are many changefeeds which all get planned on one node?
Jira issue: CRDB-36687