Re: [RFC][PATCH 0/1] MAZE: Mazed processes monitor

[Hirofumi Nakagawa]

Hi Andi,

Thank you for your comment and proposal.

> How about you implement a way to change the RLIMIT_CPU limit
> of a running task (and possibly fix it to use a finer grained unit
> if you need <1s resolution).
>
> Then you could run a user space daemon running with lower priority that just regularly resets the RLIMIT_CPUs of all running processes.
>
> If some RT process uses so much time that the user daemon cannot
> keep up its cpu time limit will be eventually exceeded and it will
> be killed.
>
> I think that would be a far cleaner and generic way to implement
> this.
>
> -Andi

I also watch normal processes, which run into infinite loop but
cleanly scheduled out by sched when a time slice expires.
In this case, your deamon still can reset the counter.

> Isn't that very similar to RLIMIT_CPU? The main difference seems to be
> that they're regularly reset and that they can be more fine grained
> than seconds.

The differences with MAZE and rlimit are as follows.
 - MAZE detects excessive CPU cycle usage, but rlimits limits total amount
   of CPU usage.
   MAZE can safely handle CPU intensive but correctly running processes.

 - User processes can add watched processes in MAZE.

 - MAZE allows users to choose a way how to act on the process,
   selecting which signal to send.

Following example shows the differences of features MZE and rlimit

----
void foo(void)
{
    for(;;);
}
----
This code receives a signal shortly,
if registered under either rlimit or MAZE.

----
void foo(void)
{
    for(;;) {
        sleep(1);
    }
}
----
This code receives a signal in case of rlimit.
But, it doesn't receive a signal under MAZE.
There are cases when you need to distinguish these two types.

Thanks.
Hirofumi Nakagawa


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/