Slashdot Mirror
Preparing Your Datacenters for DST Changes?
Cheeze asks: "As I am sure some of you know, Daylight Saving Time is slated to change this year thanks to The Energy Policy Act of 2005. This means nothing to the large majority of the population except they will either sleep late one day or have to commute in the dark. To a select few, this is a crunch time akin to the Y2K fiasco, only there has been almost zero publicity recently. These select few are the ones responsible for updating the millions of computers, both servers and workstations, with the new time zone information. For newer servers, this usually means just install a patch and reboot (which is slightly more than mildly inconvenient). For older servers, this is basically an 'End of Life' declaration. Servers running software for which no patch is available will be unable to update their own clocks. This doesn't seem like such a big deal until you realize Microsoft is only offering patches for Windows XP and beyond, and Sun will not be supporting Solaris 7 and older. That should knock a large percentage of the computers 1 hour off for a few weeks this spring. What are you doing in your datacenters to prepare?"
Using UTC.
In the US, move your servers to Arizona or eastern Indiana or Hawaii or Alaska, so you don't have to deal with time zone changes anymore.
-ez
I'm more concerning of a switch back to the old DST (the one we had in 2006) if they decide it's the energy saving isn't good enough. How difficult would it be to unpatch out systems?
There is no reason for having a single server clock on non-UTC. DST, even if it was to stay the same forever, makes things a pain in the arse (once a year, an hour happens twice).
Your client should convert the timezone. For a client machine, it's not much hassle just to go into clock settings and change it.
Worst case is web-apps - the server clock should be UTC, but you still need to convert the time zone in the client-facing PHP/Perl/Java/etc code. There you're just looking at a patch to the language runtime or whatever.
I just got my servers switched over from the Julien to the Gregorian calender. Now this. When will it ever end.
Do not try to read the dupe, thats impossible. Instead, only try to realize the truth
What truth?
There is no dupe
This should fix the problems for your Unix based machines. You can download a set of time zone source files from ftp://elsie.nci.nih.gov/pub/, these files are in the public domain. From these files you can build a full set of zoneinfo files (read the man page for zic(1)), DON'T FORGET TO BACKUP YOUR CURRENT ONES IN CASE OF PROBLEMS!! For most Linux based distros this is in a package called tzdata (which will compile the source files into zoneinfo files for you). zic(1) should put the compiled files into the right directory for you. These source files should work with most zic(1)'s, if not you may need to massage them with sed or perl. Don't forget to update /etc/localtime (or whatever other mechanism your OS uses, /etc/timezone et al) to set your localtime to the right zoneinfo file.
Mod parent up. ISTM that you're better off if all computers have their hardware clocks set to UTC and the conversions to local time take place at the last possible time in the clients (if at all). If a user wants to use some client software or OS that doesn't understand the silly daylight savings rules (which are extremely crazy in some places); at least they don't damage anything in the datacenter and will only confuse themselves.
OTOH, what kind of evil company won't even give as trivial a patch as updating a time-zone file to their customers!?! If I had any such software and the company still existed I'd be wanting a full refund.
"...Microsoft is only offering patches for Windows XP and beyond"
They are offering support for Windows 2000, free with the right volume license, $$$ without, or they have made a manual registry hack available (which I don't have handy because I'm posting from home).
And before you ask, there are companies that still specify Win2K server and workstation to run their software, Thompson/Prometric being one of them, which is how I came to find this out.
Some days it's just not worth
chewing through my restraints.
Thats a good idea, but won't always solve all of your problems. Here's a problem I experienced recently with one of our web apps.
We have a web event calendar where people can schedule events (for far into the future). When people input an event, they specify the date/time it will begin/end. The application then converts the date/time into a unix timestamp and stores it in the database. Later, when an event is being viewed, the timestamp is converted back to a textual date/time.
The conversion from local time to timestamp is done via PHP functions, which uses the systems timezone file. The OS patch to fix this problem simply updates the timezone file and everything should automatically work.
This is fine and dandy for most things, but I ran into one small glitch. For any events that are scheduled between the new start of DST and the old start of DST (roughly a 3 week period), if they were created BEFORE the patch was applied, they are now off by 1 hour AFTER the patch is applied.
The 1 hour difference is simple enough, and I might not have even noticed...or if I had, I probably would have just blamed it on user error when entering the data. However, this caused a strange side effect. The application built a secondary index to simplify searching for events. The index consisted of the event_id and the starting timestamp truncated to midnight. Of course, for those events in that 3 week period that were created before the patch was applied, they now had an index timestamp that was actually 1 AM. Of course, this caused events to mysteriously disappear from the calendar. You could search for them, but not see them when browsing.
I wasted a good couple of hours on this problem (thinking at first that it was isolated to a single event) before I finally did an analysis of the entire index and discovered that almost all events in that 3 week period were wrong. Then it instantly dawned on me what had happened.
You should have been around for the BC to AD conversion.
We thought we were going to be fine with that here, as we had some wise men who worked out when they were going to start. We were planning on just using an integer for number of years:positive for AD, negative for BC. However, some idiot politician somewhere decided to start labelling the years at 1 instead of 0, so now we've got an out-by-one bug every time you try to do a calculation that crosses the AD/BC boundary.
When our name is on the back of your car, we're behind you all the way!