But BOINC clients cache the IP forever (at least older ones do) and will continue giving you the "connect" error message even if the server is already up and running, just with a fresh IP.

Don't remember the exact version, but this was fixed around v6.10.30, so any resent BOINC-clients shouldn't normally have any problems getting the new IP.

What I was suggesting (a bit subtly perhaps) is that perhaps the error is in fact HTTP 500, "Internal Error", and not HTTP 502, "Gateway Timeout" (= slow response from the database server), nor HTTP 503, "Service [temporarily] Unavailable", i.e. overload. Given that the machine has been rebuilt, and has changed to a new IP address, and all.

Edit: The point being that waiting would fix the latter two problems, but not the first.