From mboxrd@z Thu Jan 1 00:00:00 1970 From: Lucas Meneghel Rodrigues Subject: Re: [KVM-AUTOTEST PATCH 0/5] KVM test: Support for Parallel install of guest OS Date: Wed, 10 Feb 2010 17:41:19 -0200 Message-ID: <1265830879.2273.28.camel@localhost.localdomain> References: <1265818995.6747.25.camel@yogi-laptop> <1265821989.2273.15.camel@localhost.localdomain> <20100210175450.GU6083@us.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Cc: autotest@test.kernel.org, kvm@vger.kernel.org To: Ryan Harper Return-path: In-Reply-To: <20100210175450.GU6083@us.ibm.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: autotest-bounces@test.kernel.org Errors-To: autotest-bounces@test.kernel.org List-Id: kvm.vger.kernel.org On Wed, 2010-02-10 at 11:54 -0600, Ryan Harper wrote: > * Lucas Meneghel Rodrigues [2010-02-10 11:14]: > > On Wed, 2010-02-10 at 21:53 +0530, yogi wrote: > > > Hello everyone, > > > > > > I like to submit a set of patches to enable installing multiple guests > > > in parallel. > > > > Hi Yogi, great to receive your patches! > > > > > To install multiple guests in parallel the role of the host and the > > > guest need to be changed. In this patch host acts as a client and the > > > guest acts as server, enabling us to install multiple guest at same > > > time. > > > > I see your point, but I am concerned that we already have a parallel > > test execution implementation (see control.parallel), so either your > > patchset needs to fit in with the control.parallel approach or we might > > think of another approach. I have thought of some ideas, will check your > > approach and comment about it soon. > > If I've read the patches right, I believe Yogi is relying upon > control.parallel; these changes are needed to ensure that the > unattended_install test invoked in parallel doesn't clobber itself. > Yogi hit all of the major issues: Yes, after reading the patches more carefully I realized that I was too hasty in making that comment :) > 1) tftp code uses the same path and filename for each guest; Yogi > switched that to include a per-guest path so we can have multipl > kernel/initrd pairs in flight at the same time > > 2) only one service side listening server and no way agreeing upon a > port number; Yogi moved the listen to the guest, and now the host pokes > at the pre-defined port in each guest to see if it has completed. Yes, the last months I've been running the tests on a internal test grid, as there were a lot of details to take care, I decided to keep serial execution until I was completely comfortable with all the tests, so I haven't run into this issue yet. Good point, let's fix this as soon as possible.