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

Ronald Reed rreed at ops.sgp.arm.gov
Wed May 30 13:40:05 UTC 2007


On Tue, 2007-05-29 at 23:23, Eric S. Raymond wrote:
> If they're on the other side of the planet, what good does a
> *notification* do? :-) It's not like you're going to be able to
> parachute someone in until well after the battery drains.  If what you
> want is just remote is-it-alive monitoring, ping is your friend.

I really don't need notification for me. I need notification for the NFS
server, so that it can tell the other computers that mount that file
system to shutdown before the NFS server does. That saves lots of work
when power is restored. Since everything shutdown nicely, it all comes
back nicely.

> I'm not (just) being snarky.  You seem to be kind of wobbling back
> and forth between use cases for UPS-controlled shutdown and
> remote alerts.  I agree that both actually have use cases, but
> the cases for one don't support the other very well.

I have never mentioned anything about remote alerts, I am only concerned
with a graceful shutdown that will allow my servers to restart when the
power is restored.

> Yes, the systems I administer have been behaving that way since the
> mid-1980s (except that it never worked on the ISA boxes).  It's a
> *really* short time, like on the order of 250ms, but enough for buffer
> flushes.  If that weren't so, ext3-like file systems couldn't achieve
> their main purpose.  Heck, if that weren't so disk drives wouldn't
> be able to autopark their heads.

250ms to close all the files and flush all the buffers and mark the
filesystems clean. I will trust NUT to do a graceful shutdown and not
roll the dice and hope that the hardware can do all the work that needs
to be done in less than 250ms.

> Basically because I don't want to fork the driver codebase.  That would
> be silly -- the driver design is fit for purpose.  It's the stuff above
> that layer I'm questioning.

Questioning things are always good, and this is the most discussion that
I have seen on this list since I joined it. But I think that NUT works
very well. I admit that it is rather complicated to configure, but once
you have it going, it does its job and does it very well. It is one of
the reasons that I supply a buildbot for the project.

-- 
Ronald Reed - RHCE, GCUX, GCIH
Manager of Computer Operations
ARM SGP ACRF Site
(580)388-4053 Ext. 114
rreed at ops.sgp.arm.gov




More information about the Nut-upsdev mailing list