Re: [patch 0/4] taskstats: Improve cumulative time accounting

[Oleg Nesterov <oleg@redhat.com>]

On 11/19, Peter Zijlstra wrote:
>
> On Fri, 2010-11-19 at 21:11 +0100, Michael Holzheu wrote:
> > Due to POSIX POSIX.1-2001, the CPU time of processes is not accounted
> > to the cumulative time of the parents, if the parents ignore SIGCHLD
> > or have set SA_NOCLDWAIT. This behaviour has the major drawback that
> > it is not possible to calculate all consumed CPU time of a system by
> > looking at the current tasks. CPU time can be lost.
> >
> > To solve this problem, this patch set duplicates the cumulative accounting
> > data in the signal_struct. In the second set (cdata_acct) the complete
> > cumulative resource counters are stored. The new cumulative CPU time (utime
> > and stime) is then exported via the taskstats interface.
>
> Maybe this has been treated earlier in the threads and I missed it, but
> the obvious solution doesn't get mentioned:

IIRC, the first version did this.

And it was me who spoiled this approach. But! only because I wasn't sure
this user-visible change is acceptable, and because there was some
misunderstanding. See http://marc.info/?l=linux-kernel&m=128552495203050&w=2

But,

> What would break if we

say, any test-case which does getrusage() after fork() with ignored
SIGCHLD/SA_NOCLDWAIT?.

> violate this silly POSIX rule and account time of
> childs regardless of SIGCHLD/SA_NOCLDWAIT?

+1.

Personally, I'd certainly prefer this way, because I don't care about
POSIX at all ;)


Still. Once again, this breaks the current rules, and we never do
this without strong reason.

I think we should ask Roland. If he thinks this is OK, I'd certainly
agree.

Oleg.