[clue-tech] upstart

Nate Duehr nate at natetech.com
Mon Sep 8 21:55:03 MDT 2008


Kevin Fenzi wrote:

> That said, I agree in some cases new stuff that provides no advantages
> or is not mature enough gets shipped. It happens. I really can't see 
> why anyone would argue thats the case with SysVinit/upstart however. 

The problem lies in knowing that it's there and something changed. 
Machines are rarely admined by only one person, and it only takes one 
person taking it out of SysV mode and then you're stuck knowing how to 
deal with it in a production environment.

Changes like this should be more blatent... no "SysV" mode.  If there's 
a backlash, then the devs would have pretty good reason to simply fix 
whatever was wrong in SysVInit instead of making something new.

Instead, they "sneak" something like this out -- it's on my production 
boxes whether I want it or not, and sooner or later it WILL bite me in 
the ass, if I'm not paying attention to mailing lists, etc... and know 
it's "there" on that particular flavor of Linux.  (Whichever ones are 
using the thing.)

"Compatibility modes" are supremely dangerous in production 
environments... things get deployed and work right up until someone who 
thinks they like the non-compatible stuff better, messes with it.

Heavy control of integration/release to production is the only way to 
"fight" such silliness by the upstream distros, which imposes resource 
limitations it would have been easier just not to have to deal with.

In the Solaris world (as an example of how to BETTER do this), this type 
of thing is held off until a MAJOR release -- when us "production" folks 
have already scheduled time to learn all the differences, and scheduled 
testers and resources to do regression testing, etc.  Sun pre-warned 
everyone for at least a YEAR that a new startup method was coming, and 
published how it would work BEFORE it was released.

Linux distros don't make the effort to do that very often.

Nate


More information about the clue-tech mailing list