Re: [PATCH RFC 0/2] add nproc cgroup subsystem

[Austin S Hemmelgarn <ahferroin7@gmail.com>]

On 2015-02-28 11:43, Tejun Heo wrote:
> Hello, Tim.
>
> On Sat, Feb 28, 2015 at 08:38:07AM -0800, Tim Hockin wrote:
>> I know there is not much concern for legacy-system problems, but it is
>> worth adding this case - there are systems that limit PIDs for other
>> reasons, eg broken infrastructure that assumes PIDs fit in a short int,
>> hypothetically.  Given such a system, PIDs become precious and limiting
>> them per job is important.
>>
>> My main point being that there are less obvious considerations in play than
>> just memory usage.
>
> Sure, there are those cases but it'd be unwise to hinge long term
> decisions on them.  It's hard to even argue 16bit pid in legacy code
> as a significant contributing factor at this point.  At any rate, it
> seems that pid is a global resource which needs to be provisioned for
> reasonable isolation which is a good reason to consider controlling it
> via cgroups.
If 16-bit PID's aren't a concern anymore, then why do we still default 
to treating it like a 16-bit signed int (the default for 
/proc/sys/kernel/pid_max is 32768)?