Failover strangeness and a few Thoughts

David W. Hankins David_Hankins at isc.org
Mon Nov 20 17:15:24 UTC 2006


On Sun, Nov 19, 2006 at 08:26:28AM -0600, Martin McCormick wrote:
> 	My point is that a good control mechanism must pass the
> telephone test.  If it also makes somebody's web GUI possible,

We have a slightly different way of expressing this here at ISC.
It's not written down anywhere, it's more of an oral history:

"You have to be able to do this via the only VT100 terminal in
the DataCenter whose keyboard works, over a 9600 baud serial line
half sliced by the rack case door, so only TX and RX are solid
(and even then only because the operator is holding the bare
copper closed with his teeth).

Because some days are like that."


But on the telephony parallel, you're not wrong, and as it turns
out we're more careful than I thought in selecting command line
tool names that are neither difficult to relate over a phone, nor
sound identical to other tools.

> 	The things that David W. Hankins mentioned in the quote
> above are scalable and can be designed to interface with lots of
> different mechanisms as long as one builds them to be so.  This

I'm not sure what message you're talking about anymore.

> is not a rant against anything, but an expression of hope that we
> don't make things difficult for those of us who build automation
> and who have nothing plugged in to our mouse ports.  The

My plans go as far as wanting a mechanism for DHCP similar to
bind9's rndc.  My hope is that by relying on some kind of generic
XML backchannel we'll enable others (eg the fine folks at Gnome
and their NetworkManager) to do the GUI thing with a minimum of
fuss on our side (hopefully none), while we provide an access point
for our own command-line needs (possibly also serving as debug
tools, or how-to examples).

Right now for the client I think it's pretty clear that dBus
would be the right thing for this.  This would integrate with
no middlemen with our largest consumer.

For the server it's a bit more of a fuzzy question.  dBus, for
all its merits, is still system-local, and the 'r' in rndc
after all stands for 'remote'...and there was a reason for that.

> interfaces that let folks who are blind use gnome, MACOS and
> MS-Windows simply make things behave more command-line like with
> keyboard commands that simulate drag-and-drop and mouse clicks.

I think on the one hand you would be right to criticize the Gnome
NetworkManager applet for being less accessible to blind users.

I might even criticize it for being seemingly needlessly opaque
to "advanced users," or as I prefer to call them, "generation
Hex."

All at the expense of being more accessible to "novice users,"
or as I prefer to call them, "food animals."

But since the basis of all these things is a well understood XML
backchannel (dBus and HAL), it is a small matter of programming
to implement command-line control channels for it, and solve
some of both of those critiques.

I just wouldn't expect that in the 'version 0' release, and it
doesn't surprise me that it isn't present (or if it is, that
it's poorly documented) in the current NetworkManager offering,
or in some other track.

So I think that will surface over time.  These things are still
in their infancy: you can't say the gui version of NM is totally
bugless yet.  I often fall into the annoyance that without manually
iterating 'ifup eth0' as root, NM on my laptop's X session refuses
to give me the "use wired interface" option.  Quite frequently the
DHCP protocol exchange will hang somewhere (as I am often in
strange lands), and it is most annoying that SuSe 10.1's solution
to this is to set link-local addresses and 'claim success'.  GAIM,
which senses these dBus events as a trigger to (re)connect to the
jabber server, behaves like a spider on roller skates, happily
throwing up popup windows (stealing focus, sigh) to inform me just
how bad its network connectivity is in this link-local hell.

But I still use it, because the alternative is more work.

And so it goes.


But I would doubt that ISC's initial version wouldn't include
command line tools of at least some utility.  We have different
customers in that sense.

Anyway, I'm not even going to be able to start on this until
next year sometime.

-- 
David W. Hankins	"If you don't do it right the first time,
Software Engineer		you'll just have to do it again."
Internet Systems Consortium, Inc.	-- Jack T. Hankins


More information about the dhcp-users mailing list