I'm not sure why operators care so much about the HSDir flag. It naturally comes and goes. Try not worry about it. :)
I've noticed that it can take 30+ minutes after a version upgrade before the directory service on my nodes is fully responsive again [1]. I'm not entirely sure what's happening in this timeframe, but it might be a good reason not to leave the HSDir flag in place.
----
1) To monitor the health of my relays I periodically query the directory service via a GET request to a URI. For example:
http://197.231.221.211:1080/tor/server/authority
(That's not my relay. I picked a top relay from Globe).
The monitoring software looks for a 200 response and validates the fingerprint.
After the upgrade to 0.2.7.6 all my relays returned a 404 on that URI for at least 30 minutes.