Can we narrow this down to a reproducer? It sounds like something like this would work:
---- 0) Configure two different pools where you can tell the servers apart. 1) Stop ntpd. 2) Break your DNS. 3) Start ntpd. Wait a few seconds for it to try to resolve and fail. 4) Fix your DNS. Expected results: ntpd recovers (successfully resolves the pools' DNS records) relatively quickly. Most importantly, when ntpd recovers, it recovers for both pools equally. Actual results: Only the last configured pool recovers right away. The other one takes 5 or 10 minutes. ---- Can you try a few more things: 1) Does deleting the nameservers in /etc/resolv.conf work for step 2, or more to the point, does putting them back in /etc/resolv.conf work for step 4? 2) I doubt it matters, but can you set the poll intervals the same on the two pools, to eliminate that variable? 3) If you reduce minsane to 4, what happens? Does it do the same thing, never spin up the second pool, or (far less likely) spin them both up normally? I wonder if ntpd is only "recovering" the second pool because it is below minsane. -- Richard
