core: forget a decommissioned node

[mberhault]

I can't find an existing issue for this.

Per user request: it would be nice to be able to really forget a node after decommission in cases where the node is truly gone (intentional removal or lost data).

This should either be a separate command or a special flag on decommission for safety.

[tbg]

I think this is covered by various items from https://github.com/cockroachdb/cockroach/issues?utf8=%E2%9C%93&q=is%3Aissue+is%3Aopen+decommissioned

Let's say #19863 #19541 #19863

In short, you don't need a separate command. Just decommission the node in absentia and it should be removed from the UI and various other pages.

[mberhault]

I'm not sure that's quite the same. The other issues are about still including decommissioned nodes as if they were dead. Even when that is fixed, they will still show up as decommissioned nodes. The ask here is to forget about them completely: they wouldn't even show up under "Decommissioned"

[dianasaur323]

yep, this has been requested from quite a few users. I thought it was stylistic that we chose to do it this way, since it also prevents a decommissioned node from ever coming back, but perhaps that is a bit too heavy handed.

[tbg]

@mberhault #15609 (comment)

I'll retitle that issue. My general intuition is that we should do "dead + decommissioned = don't show the node anywhere".

[fat0]

What happens if a new node comes up with the same IP of a decommissioned node IP? Ideally, we should be able to re-use old IPs that are not used by the cluster any more.

[a-robinson]

Reusing IPs should already work. As long as it isn't started up with the old node's data directory, it will be treated as a new node that hasn't been decommissioned. Have you run into any problems doing so?