Steak and sizzle
Eric S. Raymond
esr at thyrsus.com
Fri Oct 28 09:36:17 UTC 2016
Hal Murray <hmurray at megapathdsl.net>:
> > So...what else would it be interesting to watch and update in real time?
> In general, stuff doesn't change fast enough for a real-time display to be
> interesting. Most of the time, all that's going to change is the when column.
I think I've seen offset and jitter change on relatively short timescales.
Was I hallucinating?
> It might be interesting to be able to display internal data from refclocks.
> That's a whole nother can of worms. Do we want to open that before the 1.0
Probably not. For pre-1.0 I'd prefer to stick to things we can mine out of
existing Mode 6 reponses.
> There is a bunch of stuff I'd like to be able to see easily but isn't worth
> permanent screen space. The obvious one is the version of ntpd. Others
> would include
> ntpd distro version and kernel version.
> uptime and load averages.
> Temperature from wherever you can get it (CPU, disk, refclock...).
> That all probably fits on a reasonable chunk of screen space. But it's easy
> to break up if it gets too big.
There's a design pattern I tend to apply to this UI that will give us a natural
place to put things like ntpd distro version and kernel version. That is, the
display is two panels separated by a one-line status bar. Peer data goes in the
top panel, ntpd distro version and kernel version on the status bar, botttom
panel is a scrolling display of some kind of event data. Think gpsmon(1),
top(1), or mutt(1).
(Until we can think of a good thing to put in the bottom panel the status bar
will hang out at the lower edge of the window.)
> Other options would include the last N lines of ntpd.log or loopstats, or
> peerstats, ...
Last N lines of ntpd.log was my first thought. However, when you dig
into this it turns out to be remarkably difficult. syslog(3) wants to hide
the disposition of log data, and many systems (including mine) don't
configure a separate ntpd log at all.
NTP's own logs seem like a better bet, if we can figure out what is interesting.
> Or the last few lines of peerstats/rawstats/clockstats for a selected
Possibly, though ny initial esign sketch is KISS - no controls.
> It might be interesting to watch the top of the mrulist. Abusive clients
> stand out if you do something like:
> ntpq -nc "mru mincount=1000 sort=avgint"
> For a live display, mrulist probably needs something like "just give me the
> first N slots". But I think the sort is done by ntpq so that would have to
> be pushed into ntpd.
And that's something we have ready access to through Mode 6, so it's
a good candidate.
<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>
More information about the devel