Re: [PATCH 00/10] cgroups: Task counter subsystem v8

[Rik van Riel <riel-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

On 02/01/2012 02:51 PM, Andrew Morton wrote:
> On Wed, 1 Feb 2012 19:50:01 +0100
> Frederic Weisbecker<fweisbec-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>  wrote:
>
>> On Wed, Feb 01, 2012 at 08:31:26AM -0800, Tejun Heo wrote:
>>> On Wed, Feb 01, 2012 at 04:37:40AM +0100, Frederic Weisbecker wrote:
>>>> Changes In this version:
>>>>
>>>> - Split 32/64 bits version of res_counter_write_u64() [1/10]
>>>>    Courtesy of Kirill A. Shutemov
>>>>
>>>> - Added Kirill's ack [8/10]
>>>>
>>>> - Added selftests [9/10], [10/10]
>>>>
>>>> Please consider for merging. At least two users want this feature:
>>>
>>> Has there been further discussion about this approach?  IIRC, we
>>> weren't sure whether this should be merged.
>>
>> The doubts I have noticed were:
>>
>> Q: Can't we rather focus on a global solution to fight forkbombs?
>>
>> If we can find a reliable solution that works in any case and that
>> prevent from any forkbomb to impact the rest of the system then it
>> may be an acceptable solution. But I'm not aware of such feature.
>>
>> Besides, another point in having this task counter is that we
>> have a per container limit. Assuming all containers are running under
>> the same user, we can protect against a container starving all others
>> with a massive amount of processes close to the NR_PROC rlimit.

> What I struggle with is "is this feature useful enough to warrant
> merging it"?

I have seen thunderbird create as many child processes
as it could (until I hit my rlimit NR_PROC), and have
seen web servers go wrong under a combination of load
and buggy scripts, forking as many processes as they
could.

Since we know rlimit NR_PROC is useful, having the
equivalent per cgroup will be useful, too.

What we need to lose is the focus on malicious
forkbombs - buggy programs are a real issue, and
protecting against them is useful.