[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH 0/3] Experimental support for DTrace probes



On Thu, Sep 16, 2010 at 05:04:42PM +0200, Daniel Veillard wrote:
> On Tue, Sep 14, 2010 at 07:30:20PM +0100, Daniel P. Berrange wrote:
> > This patchset provides the infrastructure for supporting dynamic
> > probing of libvirtd, using static DTrace markers. This can be
> > used by SystemTAP on Linux, or DTrace on Solaris/OS-X/BSD for
> > low overhead tracing.
> > 
> > The proof of concept provides a handful of markers wrt to network
> > client connections, security & auth. Obviously it can be expanded
> > to cover a huge area of our codebase for different tasks. The
> > hard bit is deciding what should be exposed as a probe point. 
> > Ideally probes should not be changed/removed once added, since
> > this would break any user tracing scripts. So a little care needs
> > to be taken in placing probes to be robust against future code
> > re-factoring.
> 
>   Very interesting, I'm just a bit surprized by the patch set.
> patch 1 and 2 are really unrelated, I think they should go in
> independantly. Patch 1 is pure refactoring, I would rather apply it
> early in the cycle for 0.8.5 (or whatever our next release might
> be named), and glancing at it it looks finr to me ACK (will jsut need
> a bit of rebase due to Justin patch)
> Patch 2 looks simple enough, ACK too
> 
> For patch 3 I'm a bit surprized, I think I would have started by adding
> probes for all the public API places first on entry and second on exit with
> the return value provided. It can be helpful for debugging connection
> problems but well I would use SystemTap more for debugging and tuning
> of a running system (i.e. it runs and you don't want to disturb it or
> minimally)

Adding probes for public API places would require me to add 266 probes!
This patch let me demo it adding < 10 probes.  I'm not 100% sure whether
we need to add explicit probes for public API places, because I think
dtrace/systemtap may provide generic support for probes at function
enty and return points. If adding explicit probes lets it work without 
needing -debuginfo installed then it could still be worthwhile to mark
public API places.

> We should understand the problem of the restart being needed to before
> pushing patch3 I think, maybe we need to talk to one of the SystemTap
> developpers so he can explain what is going on and how to fix this :-)

Yep, there's already someone looking at it. One of the problems was
caused by LXC containers - uprobes kernel module has some bad assumptions
that LXC invalidates. The other problem looks like a data caching / race
issue in systemtap/uprobes to me. So I don't think this needs to hold
us up.

Daniel
-- 
|: Red Hat, Engineering, London    -o-   http://people.redhat.com/berrange/ :|
|: http://libvirt.org -o- http://virt-manager.org -o- http://deltacloud.org :|
|: http://autobuild.org        -o-         http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505  -o-   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]