[Nut-upsdev] Asking hard questions about the NUT architecture

[Eric S. Raymond]

Peter Selinger <selinger at mathstat.dal.ca>:
> Interesting. I just followed your advice and pulled the plug from my
> Linux machine, and none of the stuff you described happened. Instead
> of receiving and sending signals and making my applications shut down
> and save their data, the processor went ... well ... off. As in, no
> power. 

Is your hardware defective or ancient?  I haven't seen a Linux system
behave that way since the last century :-)
 
> The file system journaled and was left in a consistent state, as a
> modern filesystem should.  However, the last few hundred kilobyte of
> buffered data had not yet been synched back to disk when the drive
> lost power.

How do you actually know this?  I'm wondering because I had outages
happen to me without an UPS in the several-month interval between when
my APC died and when I purchased the Belkin, and I never lost partial
writes from my Emacs or gpsd logs or anything.  Nor did my Mozilla
session state or cache ever get hosed.

My systems behave as though the account I posted is true.  Since,
according to various standards and specifications and design
documents, I've also described the way it's *supposed* to work, my
confidence that it actually works that way is fairly high.

Hmmmm...are your disks RAIDed?  That's an edge case.

> On the other hand, last week I experienced a 15-second power failure,
> which is typical where we live. My UPS handled the situation as it
> should: it supplied power to the system. Perhaps this functionality is
> "dumb" as it does not require a USB interface. But it is quite useful!

No argument there.  It's the utility of UPS-controlled shutdowns I'm
questioning, not the utility of UPSes themselves.  I did, after all,
just buy two of them.
 
> OK, back to serious.  I agree that installing NUT from sources in the
> typical single-user single-machine single-power-supply case is too
> complicated. Even installing from a packaged distribution can be a
> pain.  It would be nice to have an "installer" script a la Acrobat
> Reader that prompts a few simple questions, lets the user select their
> hardware from a pull-down menu (if it can't be auto-detected), and
> takes care of the rest automatically, including some simple testing
> and trouble shooting

My only argument with this is that I don't want to see the writing of
a configure script turn into an excuse for not being zero-configuration
in the cases where that's possible.
-- 
		<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>