This message was deleted.

So I think that doesn’t quite work because pulumi will hang then “forever” if the spot price is too low—which may be the status quo. Also, just because the instances are fulfilled, doesn’t mean they’re up yet.

Yeah - the use case totally makes sense.

pulumi will hang then “forever”

Technically, we only wait 10m and then fail the update. Which I think is what you want here - since otherwise you would have downtime, right?

Also, just because the instances are fulfilled, doesn’t mean they’re up yet.

Is this the missing piece then - that you don't want to allow the replacement to progress until the newly created fleet is "healthy", where that is defined by some custom logic you provide?

Approximately, yes. And on the fulfillment issue: Also, since it’s a spot fleet, what I really want is status quo. So if the current fleet is preempted I don’t care if the new request isn’t immediately fulfilled. If there are hosts running under the old request, I want the new request to bring up hosts first.