Re: [PATCH] new cgroup controller "fork"

[Alan Cox <alan@lxorguk.ukuu.org.uk>]

> Putting a reasonable limit on jobs that are expected to run only for a
> limited amount of time, with a limited amount of total resources.  For

fork is an almost irrelevant resource. Address space (ie memory), file
handles and the like are actual constrained resources.

I can't help thinking this is focussing on a completely irrelevant
cornercase issue. It belongs as part of a general resource limiting
cgroup that also deals with memory, I/O and the like. In fact most of
these resources are a balancing act.

Who cares if you have 10,000 threads. We can handle that without trying.
10,000 different mappings is a whole different ball game, and 100,000 file
handles in some configurations might also matter.

In short in your specific environment a fork runaway is a symptom you can
use to detect and sometimes halt the failure case. It's not the actual
resource problem and it doesn't solve the general case.

> Similar existing feature: RLIMIT_CPU.  Millions of users have it in
> their kernels, but nobody uses it nowadays.  And it's not even
> optional.

It's required by the standards, and basically unmeasurable overhead.

> Btw. I have no problem with maintaining this patch (and a whole bunch
> of others) in my proprietary git repository at work forever.  They're
> very useful for my employer.  I'm just trying to be a good citizen by
> sharing them.

Sure - I'm just not seeing that a whole separate cgroup for it is
appropriate or a good plan. Anyone doing real resource management needs
the rest of the stuff anyway.