Re: [PATCH] xen: Implement domain runstates

[George Dunlap <George.Dunlap@eu.citrix.com>]

On 23/11/10 15:00, Keir Fraser wrote:
> Well, actually what we care about is #users-who-care. Plumbing into xentop
> to satisfy an acceptance criteria is probably misguided. Reducing
> patch-queue length just to make a patch-queue short is also probably
> misguided.

I wouldn't offer to put it in xentop just because I wanted to get the 
patch out of the patchqueue.  I think it's a useful feature, that many 
people would use if they knew about it.  If I didn't think it was 
useful, by far the simplest way to get it out of the patchqueue would be 
to simply remove the functionality from the next release of XenServer.

I think the "wait for people to ask for it" won't work very well for 
this particular feature, because it's not the kind of thing I think most 
people would think to ask for.  A bit like tmem -- I think it's a 
solution to a problem that people have that's very non-obvious, and must 
therefore be suggested to people rather than waiting for them to ask for 
it.  I could write a blog post about it or otherwise try to promote the 
idea, but how many people are going to download a patch in order to try 
it out?

(I'm aware the difference is that tmem's effectiveness can be measured 
with benchmarks while this feature cannot; I'm just saying that in both 
cases, you'd be waiting for a long time for people to demand it.)

xenalyze will calculate and report domain runstate numbers based on vcpu 
runstate_change records, and I've found it useful to get a rough idea 
what's going on when I get traces in from the field.

Of course there's no rush, but it's likely to be forgotten about for 
another two years. :-)  (It's already been in the XS patchqueue since 
September 2008.)

  -George