[procps] Re: cgroup support for new top

  • From: "Dr. Werner Fink" <werner@xxxxxxx>
  • To: procps@xxxxxxxxxxxxx
  • Date: Thu, 14 Apr 2011 13:01:14 +0200

On Wed, Apr 13, 2011 at 01:37:06PM -0500, Jim Warner wrote:
> 
> On Apr 13, 2011, at 8:35 AM, Dr. Werner Fink wrote:
> ...
> 
> > The top of the git repository of newtop does not respect
> > the personal ~/.toprc and it is not possible to run it
> > a batch job.  It seems also miss support for oom_score
> > and oom_adjustment.  But it does not crash anymore within
> > small windows least I was not able to do so. Also the
> > overflow in win_names() is fixed.
> > 
> 
> Werner,
> 
> I'm not sure we're starting off on the right foot.

Hi Jim,

it doesn't look like the right foot ;)

> 
> The newtop branch that Craig created was against procps, not wfinks-procps.
> Craig is about to commit some changes that will also embrace your library
> hacks via a #define.
> 
> What you claim is fixed was never broken in this newtop.

Hmmm ... I'm missing oom support as well as support for
hotplugged cpus (very common on s390/ppc64 zSeries) and
better support for batch job (that is no SIGHUP handler
if the batch option is used).

> This newtop supports hotplugged memory.

Good to hear.

> This newtop supports hotplugged cpus, but your library's
> sysinfo.c does not.  Why you transformed smp_num_cpus into
> a function only to return an invariant value is a mystery.

This patch based on a change from SGI of a bug report, the
request was to minimize read access to /proc/stat which will
be caused e.g. by a call to sysconf(_SC_NPROCESSORS_ONLN)
the former solution with a constructor had caused that /proc/stat
was always scanned for every program using libproc.

> This newtop supports hundreds of cpus and terabytes of memory.
> And it does so without a useless left shift as in your top.
> 
> And yes, this vastly more capable newtop will not respect a
> old .toprc file.  It's a small price to pay for the improvements.
> Still, by enabling a #define, one can alert the user of that
> rejection rather than silently defaulting.  Your current top
> doesn't even offer that option.

The problem I've with this is that in my experience this will
lead to a lot of regressions, that is that the number of my
bugzilla entries will increase a lot.

   Werner

-- 
  "Having a smoking section in a restaurant is like having
          a peeing section in a swimming pool." -- Edward Burr

Other related posts: