Atomic replication: Allocator changes

[jseldess]

Background: https://airtable.com/tblD3oZPLJgGhCmch/viwchkdQkRUw6hZ3z/reckezB9xNbQbj8hi?blocks=hide
PM: @awoods187
Eng: @bdarnell, @andreimatei

[rmloveland]

@tbg I have questions (asking because you are the author of cockroachdb/cockroach#39936 and cockroachdb/cockroach#40234).

Based on my reading of our replication layer docs (especially the Membership changes: rebalance / repair section) we simply refer to this activity as "rebalancing" without defining whether it is atomic or not. A reasonable person could assume that our rebalancing is already atomic.

Therefore I'm inclined to think that there is no need to update the replication layer docs based on this code change since it is addressing a limitation. Do you agree with that assessment?

Related: @jseldess is this known limitation about rebalances not being atomic documented anywhere? I was thinking I'd need to remove a known limitation entry about this but I'm not seeing anything obvious on our known limitations page. At least my searches for "rebalancing" and related terms on that page are not returning results that I think are related to this. But maybe I'm missing it?

[tbg]

Yep Rich that sounds right.

[rmloveland]

Awesome. In that case I will close this issue.