Since SERVER-124364/SERVER-124363, on FCV upgrade we interrupt non-authoritative DB & collection shard filtering metadata refreshes. This is done on the FcvOpObserver, immediately after the Authoritative Shards CRUD flag is enabled following the global server FCV change.

However this happens before the draining so there may still be in-flight operations still using OFCV=kUpgrading which may initiate new non-authoritative refreshes. Thus the interruption did not accomplish its purpose of ensuring all refreshes are authoritative from that point.

We should investigate if this is an issue and if so, find a solution (e.g. rely on some other op/OpObserver that runs after the draining).

(The same case applies to FCV downgrade.)