[systemd-devel] systemd and kernel process

[Steve Dickson]

On 10/03/2015 08:18 AM, Lennart Poettering wrote:
> On Wed, 30.09.15 16:49, Steve Dickson (SteveD at redhat.com) wrote:
> 
>> Hello,
>>
>> Is there a way for systemd to monitor kernel process?
> 
> To add to what Kay said:
> 
> No. Kernel threads cannot really be tracked by userspace in any
> sensible manner. We won't get SIGCHLD for them, and they cannot be
> moved outside of the root cgroup. This means we will not get events
> about their life-time, and cannot track their existance at all. 
> 
> With the current kernel this is hence technically not possible. On top
> of that though, I am not sure this would even be desirable
> semantically: babysitting kernel facilities from userspace sounds
> wrong to me, I think we want a clear hierarchy, that the kernel is
> responsible for PID 1, and then PID 1 for the rest of the
> services. But making PID1 also responsible for the kernel appears like
> the wrong way around.
Its for an HA environment so they need to know the
"health" of the NFS server, which is a combination of
userspace daemon and kernel processes.

The userspace daemon all have systemd units so monitoring
their health not a problem. I just hoping I could used 
systemd to monitor the kernel processes as well. But
I guess that's not possible. 

Thanks for the input!

steved.