Monday, November 16, 2009

yes, we've been having windstorms

I tend to think that when they warn about 40-50mph gusts it's not such a big deal, but we did get a 2-second power glitch this evening during which time the hardware clock mysteriously gained about 2 hours -- given the interval, I am less certain that this had anything to do with DST or UTC-vs-local nonsense this time. Buh.

Fortunately I was sitting right there and caught it in time to not have to do (yet another) timewarp. Yay.

Well okay, having looked back over the last five years of blog and seeing that we were doing a timewarp pretty much every fall for at least 3 years in a row -- missed last year for some reason -- and having also seen that my noting of the hardware clock being on localtime and not UTC was not actually accompanied by some kind of assertion that I was going to fix this as I thought I had. And lo and behold, after grepping the init scripts for every instance of hwclock I found a setting in /etc/default/rcS that I hadn't seen before that actually addresses the problem and was set the wrong way. Whee.

Also found out that the hardware clock, which is only used to keep time while the power is out, apparently only ever gets synched when the system does a clean shutdown. So if every shutdown for months on end is a power failure, all bets are off. While things up and running, hwclock can be wildly wrong and nothing will notice or care. So now it's making a lot more sense why a power failure, say a month after the DST switchover, results in an hour jump.

Anyway, we now definitively have the hwclock on UTC (rebooted while the network was down), so there is now a high level of confidence that we won't be having to do timewarps ever again (knock wood... I suppose it probably would also be nice to know why NTP wasn't taking care of this ... project for another day...)