[PATCH] Use the same process and krb5 ccache when invoking samba-tool in drs tests

[Andrew Bartlett]

On Wed, 2017-07-26 at 09:11 +0200, Andreas Schneider wrote:
> On Wednesday, 26 July 2017 08:36:33 CEST Andrew Bartlett wrote:
> > On Wed, 2017-07-26 at 07:58 +0200, Andreas Schneider wrote:
> > > On Tuesday, 25 July 2017 22:07:24 CEST Andrew Bartlett via samba-technical
> > > 
> > > wrote:
> > > > On Sat, 2017-07-22 at 09:00 +1200, Andrew Bartlett via samba-technical
> > > > 
> > > > wrote:
> > > > > On Sat, 2017-07-22 at 08:19 +1200, Andrew Bartlett via samba-technical
> > > > > 
> > > > > wrote:
> > > > > > This patch removes a number of calls to fork() inside python, and
> > > > > > instead of passing -Uusername%password on the internal command line,
> > > > > > we
> > > > > > pass around the name of the krb5 credentials cache we already have.
> > > > > > 
> > > > > > I've not been able to quantify the speed benefits - there seems to
> > > > > > be
> > > > > > more noise than signal in the timing here, hence my other patch
> > > > > > reducing idle time, but when we had these tests under
> > > > > > perf/FlameGraph
> > > > > > the string2key routines really dominated the time what was spent on
> > > > > > the
> > > > > > CPU.
> > > > > 
> > > > > Actually, if I use my tevent_immediate patch as the baseline, it saves
> > > > > about 80 seconds per repl_move test on my laptop.
> > > > > 
> > > > > Aside from making the timing more reliable, it is however independent
> > > > > of that patch.
> > > > > 
> > > > > > Earlier in the thread, Jeremy looked over the C parts, but if
> > > > > > someone
> > > > > > could look over the rest that would be great!
> > > > > > 
> > > > > > Please review!
> > > > 
> > > > Can I please get a review on this patch series?  As above, it makes a
> > > > measurable reduction to the time taken for make test, which we would
> > > > all appreciate.
> > > 
> > > I was busy with metze his async spnego patchset, that one is the next on
> > > my
> > > list.
> > 
> > Thanks for all your work on that.  I got some very helpful feedback
> > (this patch set has a botched patch split up) and then Douglas' review
> > on this today, and so it is in autobuild.
> > 
> > Thanks!
> > 
> > Andrew Bartlett
> 
> https://git.samba.org/?p=asn/samba.git;a=shortlog;h=refs/heads/master-review
> 
> pycredentials: Add get_name() for a credentials cache
> 
> 
> This one had an incorrect commit message (set_ccache_name). Also the pointers 
> where not initialized with NULL. We agreed to always init them with NULL and 
> especially out pointers should be.
> 
> 
> pycredentials: Add set_named_ccache()
> 
> Pointers not initialized
> memory leak on error path of cli_credentials_set_ccache
> exceeded 80 cols ...
> 
> 
> I'm would push that branch to autobuid if you're fine with it ...

Sadly that branch doesn't have the other fixes I tried to push today.

Those are at 
http://git.catalyst.net.nz/gw?p=samba.git;a=shortlog;h=refs/heads/faster-drs-krb5-ccache-2

Something has gone wrong, as my autobuild is neither running, nor do I
have a failed autobuild message.  I've reset it to make sure.

I'll figure out the difference between your fix-up and my new branch
and get a combined improved set into master tomorrow.

Thanks!

Andrew Bartlett

-- 
Andrew Bartlett                       http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba