[ltt-dev] [RFC UST] Processes model

[Nils Carlson]

Stefan Hajnoczi wrote:
> On Tue, Jan 18, 2011 at 11:15 AM, Nils Carlson
> <nils.carlson at ericsson.com> wrote:
>   
>> Hi,
>>
>> Stefan Hajnoczi wrote:
>>     
>>> On Mon, Jan 17, 2011 at 9:13 PM, David Goulet <david.goulet at polymtl.ca>
>>> wrote:
>>>       
>>>> I'm submitting a RFC about a "new daemon model" for UST.
>>>>
>>>> This is the *first* draft... so a lot of feedback will be appreciated for
>>>> improvements and ideas!
>>>>         
>>> At first glance I'm concerned about a root ustd daemon.  One of the
>>> things I like about LTTng UST is that it's low on administration
>>> overhead and configuration.  Putting a central service in place that
>>> mediates access introduces a level of complexity that makes it harder
>>> for unprivileged users who want to quickly trace something in an app
>>> they are running or developing.
>>>       
>> The central daemon is already in place. I assume your talking about
>> using the usttrace command?
>>     
>
> I don't remember ever running as root or having the daemon get in the
> way.  It makes sense to have an out-of-process daemon but keeping
> lightweight ad-hoc usage supported is important.  Especially for
> userspace tracing, LTTng has a big advantage over SystemTap here due
> to less headaches installing and configuring the tracing system.
>
>   
The daemon, ustd or ustconsumerd (renamed in the latest version) can run
in multiple instances. If you run it as a normal user using the usttrace
command it will create its socket in a different place than the central
daemon and the application started by usttrace will connect to your its own
daemon using that socket. I think this design should be possible to keep
without too much ado.

Do you ever start applications without using the usttrace command and 
connect
to them with ustctl?

/Nils

> Stefan
>