Undocumented(?) change in default for wait_for_active_shards on close index requests

[DaveCTurner]

In 7.x the default for ?wait_for_active_shards on a POST /index/_close request is NONE (i.e. do not wait) whereas in 8.0 it defaults to DEFAULT (i.e. whatever the index metadata says). I took a quick look at the breaking changes docs for 8.0 and didn't see this mentioned, and nor do I see anywhere that we're emitting deprecation logs indicating the change in behaviour.

This caught me out when writing some tests (see #66415). Should we consider this a breaking change and warn on cases where folk are relying on it? Perhaps not, it might not break anything. I'm opening this issue to remind us to discuss it at some point before releasing 8.0.

@tlrx WDYT?

[elasticmachine]

Pinging @elastic/es-distributed (Team:Distributed)

[tlrx]

Looking at the list of changes when the replication of closed indices was merged in master (#39499) and more precisely at #38854 I see nothing that documents the NONE to DEFAULT change so I think I missed to do it. When the change was merged in 7.2 we decided to use NONE to keep the existing behavior of immediate closing (before 7.2 closing indices was also a way to get rid of unassignable indices...) but it should have been documented as a change for 8.0.

So I agree with you and I think the NONE to DEFAULT move should be documented as breaking in 8.0.

[DaveCTurner]

Ok 👍 I opened #66543 for the breaking changes doc and #66542 to add deprecation warnings.