[ovs-dev] [v1 1/1] Fix "raceful" E2E ovn tests

Lance Richardson lrichard at redhat.com
Mon Jul 18 18:51:48 UTC 2016


----- Original Message -----
> From: "Ben Pfaff" <blp at ovn.org>
> To: "Lance Richardson" <lrichard at redhat.com>
> Cc: "Amitabha Biswas" <abiswas at us.ibm.com>, dev at openvswitch.org
> Sent: Monday, July 18, 2016 2:34:02 PM
> Subject: Re: [ovs-dev] [v1 1/1] Fix "raceful" E2E ovn tests
> 
> On Mon, Jul 18, 2016 at 01:50:40PM -0400, Lance Richardson wrote:
> > ----- Original Message -----
> > > From: "Ben Pfaff" <blp at ovn.org>
> > > To: "Amitabha Biswas" <abiswas at us.ibm.com>
> > > Cc: dev at openvswitch.org
> > > Sent: Monday, July 18, 2016 1:33:56 PM
> > > Subject: Re: [ovs-dev] [v1 1/1] Fix "raceful" E2E ovn tests
> > > 
> > > On Fri, May 20, 2016 at 11:17:57AM -0700, Amitabha Biswas wrote:
> > > > This patch fixes the "raceful" situation that occurs when ovn-nbctl
> > > > configures the OVN Northbound database till the point when the
> > > > ovn-controller picks up the corresponding logical flows and applies
> > > > them as OpenFlow rules on the hypervisor.
> > > 
> > > I posted a patch that does something similar:
> > >         http://openvswitch.org/pipermail/dev/2016-July/075516.html
> > > 
> > > I forgot, when I started writing it, that this previous patch had
> > > existed.
> > 
> > I have spent some time recently looking into the root causes of OVN
> > E2E test failures.  Initially I was thinking that this would lead to
> > a better understanding of these races, but as it turned out it seems
> > that most of these failures are not really due to races at all, but
> > are instead mostly due to other types of bugs. Here's the list:
> > 
> > 
> >    https://patchwork.ozlabs.org/patch/631074/ (merged)
> >    https://patchwork.ozlabs.org/patch/643536/ (merged)
> >    https://patchwork.ozlabs.org/patch/645650/ (queued for review)
> >    https://patchwork.ozlabs.org/patch/646179/ (queued for review)
> >    https://patchwork.ozlabs.org/patch/646257/ (queued for review)
> >    https://patchwork.ozlabs.org/patch/648815/ (merged)
> 
> Hmm, that's interesting.  I guess that we are all too eager to blame
> bugs on "races", when in fact there can be (and it appears that there
> are in practice!) many root causes for bugs.  I guess that fixing the
> races, then, will make it more likely that we won't ignore other bugs.
> 

++, agreed!... I think that eliminating races is hugely important for
real deployments, while improving the reliability of the test suite will
pretty much only be interesting for developers.



More information about the dev mailing list