[16324] in Kerberos_V5_Development
Re: KDC worker processes project
daemon@ATHENA.MIT.EDU (Nicolas Williams)
Thu Sep 16 17:52:47 2010
Date: Thu, 16 Sep 2010 16:51:07 -0500
From: Nicolas Williams <Nicolas.Williams@oracle.com>
To: Greg Hudson <ghudson@mit.edu>
Message-ID: <20100916215107.GN3982@oracle.com>
MIME-Version: 1.0
Content-Disposition: inline
In-Reply-To: <1284671161.5992.1754.camel@ray>
Cc: "krbdev@mit.edu" <krbdev@mit.edu>
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Errors-To: krbdev-bounces@mit.edu
On Thu, Sep 16, 2010 at 05:06:01PM -0400, Greg Hudson wrote:
> On Thu, 2010-09-16 at 15:56 -0400, Nicolas Williams wrote:
> > Why can't you test by starting the KDC and then confirming that all the
> > expected processes exist, with the right parent/child hierarchy?
>
> I hadn't thought about that sort of test; I was thinking more along the
> lines of ensuring that both worker processes could accept UDP packets
> and TCP connections in parallel.
You could use debug messages to test the TCP case fairly easily: have N
processes/threads connect()-but-write-nothing to the KDC's TCP port,
then check that more than one KDC process wrote a debug message about
accepting connections.
But yes, I agree, this appears to be best unit tested.
> Examining the process table is kind of a portability minefield, isn't
> it?
You could just have the KDC processes write debug messages about it.
> > How will the number of processes be configured? What's the default?
> > (Twice the number of CPUs seems like a reasonable default to me.)
>
> With the -w command line option to krb5kdc. Getting it from the profile
> wouldn't be too difficult if that's important.
Seems desirable.
> The default is not to use worker processes. If you don't have KDC load
> issues and anything goes wrong, you're going to have an easier time
> diagnosing it if the KDC didn't aggressively parallelize in order to
> take advantage of system resources.
KDCs are typically run on systems dedicated for that purpose.
Aggressively using available CPU resources seems like a very desirable
default. Though you might want to be more conservative than I was (say,
the number of CPUs minus 1).
Nico
--
_______________________________________________
krbdev mailing list krbdev@mit.edu
https://mailman.mit.edu/mailman/listinfo/krbdev
