The leap second of doom
Posted Aug 2, 2012 22:32 UTC (Thu) by nix
Parent article: The leap second of doom
Martin Burnicki described a problem his customers saw after June's leap second in which certain older NTP servers did not reset the leap flag after the event.
Later, he described a more interesting case, where multiple NTP servers had each other set as upstreams. If something (perhaps network latency) caused a query to be sent to one of these servers from one of the others just before the end of June, such that it arrived just after the start of July, that server would instantly flip the leap-second flag (that it had just flipped off) back on. After that, because it's the upstream for all the other NTP servers in this little 'liar group', the lie would spread to all the others, even if they had a higher-stratum machine or an accurate time source telling them otherwise.
to post comments)