[LEAPSECS] Longer horizon

Warner Losh imp at bsdimp.com
Tue Jul 10 10:38:19 EDT 2012



On Jul 10, 2012, at 8:26 AM, Rob Seaman wrote:


> On Jul 10, 2012, at 7:09 AM, Warner Losh wrote:

>

>> On Jul 10, 2012, at 7:12 AM, Daniel R. Tobias wrote:

>>

>>> On 9 Jul 2012 at 14:31, Warner Losh wrote:

>>>

>>>> First, the current "right" database can't be updated in place:

>>>> you have to restart.

>>>

>>> M$ Windows people are used to constantly having to restart their

>>> systems at the most trivial updates... *Nix folks are spoiled!

>>

>> I think that you've got it backwards. M$ developers have been spoiled that the don't have to tackle the hard problem of not forcing a restart. Forced restarts kill your '9's of uptime.

>

> Really? Changing all the clocks on the planet is deemed easier than updating some software to be responsive to a HUP?


You really don't understand the depth of the leap second issue in software. If it were that easy, it would have actually been solved. People just don't care, and that's the problem. You care a whole lot, but aren't out there fixing bugs in code, or driving adaptation of new standards that cope with leap seconds rather than forcing people into picking which way to violate them is the least bad for them. Until people care enough to fix the standards and the attitude, leap seconds will continue to be broken in software.


> We have a git seminar here tomorrow. Software has power build tools that are the envy of other engineering disciplines. The woman on the Clapham omnibus is not supposed to care whether it's dark outside during the daytime, but a bunch of programmers tossing around cute jargon like *Nix and M$ can't arrange to update a record in a database at runtime?


The problems run deeper than "just" that. There's the problem of distribution. There's the cold shelf problems. There's the problems of actually getting people to give a shit and fix their broken code when you report bugs on it. There's the problem of getting people to actually test that their locking doesn't cause deadlock on lead-second day.

Really, I've been there done that and tried to effect change in this area. You haven't. So please, keep the snarky "oh, it's just a SIGHUP" comments to yourself, ok?

Warner


More information about the LEAPSECS mailing list