systemd and kernel process

systemd and kernel process

[Steve Dickson]

Hello,

Is there a way for systemd to monitor kernel process?
By monitor I mean the existence. 

Here the story...  a systemd service calls a command
that creates a number kernel process/threads 
then the command exits. 

Is there a way for systemd to monitor those kernel process
even though it was told nothing about them?

tia,

steved.

Re: systemd and kernel process

[Josh Stone]

On 09/30/2015 11:54 AM, Steve Dickson wrote:
> Hello,
> 
> Is there a way for systemd to monitor kernel process?
> By monitor I mean the existence. 
> 
> Here the story...  a systemd service calls a command
> that creates a number kernel process/threads 
> then the command exits. 
> 
> Is there a way for systemd to monitor those kernel process
> even though it was told nothing about them?

I assume you mean that systemd created the kernel threads (indirectly),
and you want *systemtap* to monitor them?

You can use probe kprocess.create, which will hit for all new tasks.  It
runs in the context of the parent thread.  You can see if "task->mm" is
NULL to tell if it's a kernel thread.

If you need to run without kernel debuginfo, it's about the same thing
at probe kernel.trace("sched_process_fork"), and check "$child->mm".

There's also probe kprocess.start for the first time a new thread is
scheduled onto a cpu.  This might be more useful if you want to probe
from the context of that thread itself, rather than the parent.  I'm not
aware of an equivalent


Telling whether a new kernel thread came about via systemd is a trickier
question.  Probably need to probe something more specific to whatever
you're looking for.

Re: systemd and kernel process

[Steve Dickson]

Hello,

On 09/30/2015 04:25 PM, Josh Stone wrote:
> On 09/30/2015 11:54 AM, Steve Dickson wrote:
>> Hello,
>>
>> Is there a way for systemd to monitor kernel process?
>> By monitor I mean the existence. 
>>
>> Here the story...  a systemd service calls a command
>> that creates a number kernel process/threads 
>> then the command exits. 
>>
>> Is there a way for systemd to monitor those kernel process
>> even though it was told nothing about them?
> 
> I assume you mean that systemd created the kernel threads (indirectly),
> and you want *systemtap* to monitor them?
My bad... I sent the email to the wrong list...

autofill is not my friend! :-)

> 
> You can use probe kprocess.create, which will hit for all new tasks.  It
> runs in the context of the parent thread.  You can see if "task->mm" is
> NULL to tell if it's a kernel thread.
> 
> If you need to run without kernel debuginfo, it's about the same thing
> at probe kernel.trace("sched_process_fork"), and check "$child->mm".
> 
> There's also probe kprocess.start for the first time a new thread is
> scheduled onto a cpu.  This might be more useful if you want to probe
> from the context of that thread itself, rather than the parent.  I'm not
> aware of an equivalent
> 
> 
> Telling whether a new kernel thread came about via systemd is a trickier
> question.  Probably need to probe something more specific to whatever
> you're looking for.
> 
Again... I need to be able to do a systemctl status <whatever>
to see if the NFS kernel process are or are not running...

Sorry for the noise and thank you for your response!

steved.