It looks like we can start the system vm, immediately write cmdline to the vm's socket, and then we can dump it out of the character device in the system vm (literally just cat the char dev) in cloud-early-config where it would normally mount the patch disk. After that, I think we may want to use our own daemon listening on the character device, rather than qemu-guest-agent, because it doesn't seem like qemu-guest-agent allows us to do much besides read/write files and a few system level things like freezing filesystems. If anyone has more info that I'm missing on that please speak up. I'm assuming we'd want to have something we can communicate with to launch the various system vm scripts. Plus we could distribute it in with the scripts.
For the time being though, we can leverage the socket to push cmdline/authorized keys, and skip any special changes to how scripts are called or having to install qemu-guest-agent. On Tue, Mar 5, 2013 at 1:15 AM, Rohit Yadav <bhais...@apache.org> wrote: > Marcus, let me know I can install qemu-guest-agent during > postinstallation phase of the systemvmtemplate, and in > patches/systemvm you can configure cloud-early-config or a script > which runs this service? During startup of a systemvm template, > systemvm.iso is used to patch this template to create one of ssvm, > cpvm or rvm. > > Regards. > > On Tue, Mar 5, 2013 at 4:28 AM, Marcus Sorensen <shadow...@gmail.com> wrote: >> I have a request that if we provide a system vm going forward that it >> have virtio_console support from early on, as well as qemu-ga >> (qemu-guest-agent) service started before cloud-early-config, so we >> can use it to control the system vms on KVM. >> >> On Fri, Mar 1, 2013 at 2:43 PM, Sudha Ponnaganti >> <sudha.ponnaga...@citrix.com> wrote: >>> Hi All, >>> >>> Just want to summarize what QA is doing for 4.1 validation and 4.2 so we >>> are all on the same page. >>> >>> 4.1 >>> * we will use 4.0 template for IPV4 testing ( all QA uses this one) - >>> (Even for this VMWare template is broken [1] and that is a different topic) >>> *Sangeetha who is testing IPV6 or any others who are interested can use >>> template provided by Sheng. >>> >>> 4.2 ( Master) >>> * Continue to use older templates (4.0) temporarily till Rohit makes the >>> template process streamlined for all HVs >>> * Reason is being that blockers are being logged against new procedure done >>> by Rohit [2]. It need to be tested and all templates ( xen/vmware/kvm) need >>> to completed. >>> * once he gives a go ahead all QA will consume that. >>> >>> [1] https://issues.apache.org/jira/browse/CLOUDSTACK-1252 >>> [2] https://issues.apache.org/jira/browse/CLOUDSTACK-1462 >>> >>> Hope this approach is fine for everyone. >>> >>> Thanks >>> /Sudha >>> >>> -----Original Message----- >>> From: Chip Childers [mailto:chip.child...@sungard.com] >>> Sent: Thursday, February 28, 2013 1:13 PM >>> To: cloudstack-dev@incubator.apache.org >>> Subject: Re: What are the System VM Templates for ASF 4.1 Release? >>> >>> On Thu, Feb 28, 2013 at 10:47:25AM -0800, Sheng Yang wrote: >>>> On Wed, Feb 27, 2013 at 9:52 PM, Chiradeep Vittal >>>> <chiradeep.vit...@citrix.com> wrote: >>>> > >>>> > >>>> > On 2/27/13 10:12 AM, "Sheng Yang" <sh...@yasker.org> wrote: >>>> > >>>> >>Per this case, if people thinks systemvm template can be hosted >>>> >>alone, I would suggest use the tested ipv6 template for the whole >>>> >>4.1 release, to avoid confusion. >>>> > >>>> > As long as it is documented, it shouldn't cause too much confusion. >>>> > People are not likely to be using ipv6 by accident, especially since >>>> > it is considered experimental. >>>> > I am sure your template is fine, but an abundance of caution at this >>>> > stage of the game would lead me to believe that it is best to go >>>> > with the 2-pronged approach. If we were making this decision 3 weeks >>>> > ago, I'd say, 'yeah, probably OK'. >>>> >>>> I've sent out the notice when I branch out for IPv6, said it would >>>> need a template. I stated so again when check in for 4.1 branch. And I >>>> opened the bug for fixing this issue in 4.1. Thanks to Rohit, we >>>> started discussion [3]. Everything looks fine. >>>> >>>> But this thing still happened. Bug changed to 4.1 fix version, the >>>> issue raised by QA at last minute. >>>> >>>> I don't know how loud should I speak if we need a template for IPv6 in >>>> 4.1. Seems nobody cares. >>>> > >>>> >> >>>> >>Document the step to switch is fine, but two set of systemvm >>>> >>template for one release would be tricky I think. >>>> > >>>> > Yes, but it is experimental. >>>> > >>>> >> >>>> >>And the change to the ipv6 systemvm template, is it just contained >>>> >>upgraded dnsmasq(version 6.22). That's it, nothing changed beside >>>> >>that. I kind of believe it should be mostly the same as before, >>>> >>tested enough for default template. >>>> > >>>> > These are not strong, confident statements. To make it simpler, we >>>> > could use approach 'B' with the caveat that it does not run the >>>> > apt-get unless some explicit action is taken by the cloud admin. For >>>> > example: >>>> > - a global flag (systemvm.ipv6.enable) or >>>> > - whenever an ipv6 subnet is created. >>>> >>>> I don't think the thing would depends on if my statement is strong or >>>> confident. >>>> >>>> I don't think we should let systemvm run apt-get things. >>>> >>>> According to what I observed in the community, I think probably it's >>>> right that people not quite interested in ipv6. >>> >>> To be clear, I personally am *very* interested in getting IPv6 support. >>> I think what we are talking about is the fact that this is experimental for >>> 4.1 (as previously agreed). >>> >>>> >>>> Probably we just revert the UI for 4.1 branch, and make API usable >>>> with updated template. >>> >>> +1 to that approach. >>> >>> And another +1 to the implied implementation of IPv6, plus a new template, >>> plus a new template build process, plus the UI, plus lots of testing... to >>> make IPv6 support a full feature for 4.2. >>> >>>> >>>> --Sheng >>>> >>>> [1] http://article.gmane.org/gmane.comp.apache.cloudstack.devel/10785 >>>> [2] http://article.gmane.org/gmane.comp.apache.cloudstack.devel/11387 >>>> [3] >>>> http://thread.gmane.org/gmane.comp.apache.cloudstack.devel/12183/focus >>>> =15159 >>>> > >>>> >> >>>> >>VMware template may need some work, I remember last time we upgrade >>>> >>the vmware template by installing some vmware tools, which didn't >>>> >>affect other two templates(KVM and Xen). We would need to do it >>>> >>again, Kelven should able to help with it. >>>> >> >>>> >>--Sheng >>>> >> >>>> >>On Wed, Feb 27, 2013 at 8:12 AM, Chip Childers >>>> >><chip.child...@sungard.com> wrote: >>>> >>> On Tue, Feb 26, 2013 at 10:23:04PM -0800, Chiradeep Vittal wrote: >>>> >>>> Another work-around may be to not require new systemvms unless >>>> >>>>the ipv6 feature is required in which case: >>>> >>>> A. We provide the bits of the systemvm of whatever Sheng's been >>>> >>>>testing with (with the caveat that it is under development/beta) >>>> >>>>B. Write a patch for cloud-early-config (or ssh in after VR is >>>> >>>>created) to >>>> >>>> apt-get update + apt-get install <ipv6 packages> >>>> >>> >>>> >>> I like option A. We had actually already agreed that IPv6 would >>>> >>> be considered "experimental" in this release anyway. So if >>>> >>> someone wants to try it out with 4.1, IMO it's OK to have them do >>>> >>> a little more work to get the correct system VM. >>>> >>> >>>> >>> As long as we document it, I think that option A is the right one. >>>> >>> >>>> >>> Other thoughts? >>>> >>> >>>> >>>> >>>> >>>> On 2/26/13 10:15 PM, "Rohit Yadav" <bhais...@apache.org> wrote: >>>> >>>> >>>> >>>> >On Wed, Feb 27, 2013 at 3:45 AM, Sheng Yang <sh...@yasker.org> wrote: >>>> >>>> >> When I first report the bug >>>> >>>> >> https://issues.apache.org/jira/browse/CLOUDSTACK-1066 >>>> >>>> >> >>>> >>>> >> I've set the target for 4.1 because of ipv6 need. >>>> >>>> >> >>>> >>>> >> When Rohit fixed it, it was changed to 4.2, sorry I didn't >>>> >>>> >>aware of that. >>>> >>>> > >>>> >>>> >Yes Sheng is correct, I was responsible for that because the >>>> >>>> >feature/code to create systemvms was not even started and since >>>> >>>> >I started working on it after the code freeze, I moved the >>>> >>>> >version to >>>> >>>> >4.2 >>>> >>>> >It was only recently when I found out that ipv6 is going to make >>>> >>>> >it in 4.1, in that case the feature is code complete [1] and >>>> >>>> >we've an automated jenkins job. The only problems are: >>>> >>>> > >>>> >>>> >- Code syncing: I did not cherry-pick the code to 4.1 >>>> >>>> >- Testing: We need to test against 4.1 branch that the >>>> >>>> >appliance/template really works [2] >>>> >>>> > >>>> >>>> >I'm sorry Sheng if ipv6 won't make in 4.1 because of this. But I >>>> >>>> >would try my best to test/fix the template for Xen at least >>>> >>>> >before 28/2, I really want to see your feature go in 4.1 Since, >>>> >>>> >4.1 is frozen, community would have to make an exception to at >>>> >>>> >least allow the new systemvms templates (if not the code) to be >>>> >>>> >used in case it works fine for all three (kvm, xen and vmware) >>>> >>>> >and we could still fix/test ahead of time, we still have few >>>> >>>> >more weeks before the release; otherwise we can always use the same >>>> >>>> >old template. >>>> >>>> > >>>> >>>> >Comments, suggestions, especially from Chip and ppmc? >>>> >>>> > >>>> >>>> >Regards. >>>> >>>> > >>>> >>>> > >>>> >>>> >[1] https://issues.apache.org/jira/browse/CLOUDSTACK-1066 >>>> >>>> >[2] https://issues.apache.org/jira/browse/CLOUDSTACK-1340 >>>> >>>> > >>>> >>>> >> >>>> >>>> >> --Sheng >>>> >>>> >> >>>> >>>> >> On Tue, Feb 26, 2013 at 2:12 PM, Chip Childers >>>> >>>> >> <chip.child...@sungard.com> wrote: >>>> >>>> >>> On Tue, Feb 26, 2013 at 02:07:37PM -0800, Chandan >>>> >>>> >>> Purushothama >>>> >>>>wrote: >>>> >>>> >>>> Building System VM Template is a 4.2 feature >>>> >>>> >>>>https://issues.apache.org/jira/browse/CLOUDSTACK-1340. The >>>> >>>> >>>>system >>>> >>>>VM >>>> >>>> >>>>Templates posted by Rohit is for the Master branch >>>> >>>> >>>> >>>>>>>>http://jenkins.cloudstack.org/view/master/job/build-systemvm-m >>>> >>>>>>>>aster/ >>>> >>>>>>>>las >>>> >>>> >>>>tSuccessfulBuild/artifact/tools/appliance/dist/ . I am >>>> >>>> >>>>referring to the ASF 4.1 Release System VM Templates in my >>>> >>>> >>>>question. >>>> >>>> >>> >>>> >>>> >>> So in that case, I guess the only system VMs we have to use >>>> >>>> >>> now >>>> >>>>are the >>>> >>>> >>> same ones we used for 4.0 (which were inherited from Citrix >>>> >>>>pre-ASF). >>>> >>>> >>>> >>>> >>>> > >>>>
