On Fri, May 19, 2017 at 06:28:36AM +0300, Leni Kadali Mutungi wrote:

I'd like the input of the Debian Virtualization Team on the messages below. When replying, cc the oVirt team <users@ovirt.org>

---------- Forwarded message ---------- From: Yedidyah Bar David <didi@redhat.com> Date: Tue, 16 May 2017 09:51:24 +0300 Subject: Re: [ovirt-users] Building ovirt-host-deploy gives `configure: error: otopi-devtools required but missing` To: Leni Kadali Mutungi <lenikmutungi@gmail.com> Cc: Jason Brooks <jbrooks@redhat.com>, users <users@ovirt.org>

On Mon, May 15, 2017 at 10:42 PM, Leni Kadali Mutungi <lenikmutungi@gmail.com> wrote:

...

On 5/15/17, Yedidyah Bar David <didi@redhat.com> wrote:

...

Now is the time to explain something.

ovirt-host-deploy is not designed to be ran manually like you try.

You will get the exact same error if you try to do this on CentOS or Fedora.

The normal way it works is that the engine "bundles" it in a tarball, copies it to the target host using ssh, untars it there and runs it.

It then talks with it - the engine sends some stuff, host-deploy replies, etc.

I'm guessing this means I can now move on to building ovirt-engine.

When looking into packaging the Java dependencies talk to the Debian Java packaging team: debian-java@lists.debian.org

That would be great, but:

1. It will require a lot of work, mostly probably of packaging in Debian dependencies of the engine that are not yet there.

2. You'll be mostly walking the unwalked path. Most of the work on Debian support so far was on the hosts' side.

Re (1.), you should also note that currently, the engine is not even packaged properly for Fedora/CentOS. I know this sounds weird, as everyone uses the packages we provide, but it's true - these packages are not compliant with Fedora's packaging guidelines. Why? Because we use maven for building the engine, and package also many dependencies simply as a result of maven getting them from maven central. See also:

https://bugzilla.redhat.com/showdependencytree.cgi?id=1168605&hide_resolved=0

Best,

...

...

The protocol they use is described in otopi, in the file README.dialog.

otopi has (currently) two "dialects" - "human" (default) and "machine". The engine and ovirt-host-deploy talk using the machine dialog.

To make ovirt-host-deploy talk with you using the machine dialog, you should run it with:

ovirt-host-deploy DIALOG/dialect=str:machine

To make it let you configure it, run it with:

ovirt-host-deploy DIALOG/dialect=str:machine DIALOG/customization=bool:True

To know what it expects at each stage, I suggest to have a look at an ovirt-host-deploy log generated on el7 or fedora.

...

Anyway, congrats for a nice progress! Thanks. I wouldn't have come this far without the community's help and

This is very useful; will definitely try this out. the documentation.

...

...

I tried starting the libvirtd service to see if that would make the VIRT/enable error go away or at least satisfy the requirements of ovirt-host-deploy, but it didn't seem to work.

If you check such a log file, you'll see there (among other things):

DIALOG:SEND **%QStart: CUSTOMIZATION_COMMAND DIALOG:SEND ### DIALOG:SEND ### Customization phase, use 'install' to proceed DIALOG:SEND ### COMMAND> DIALOG:SEND **%QHidden: FALSE DIALOG:SEND ***Q:STRING CUSTOMIZATION_COMMAND DIALOG:SEND **%QEnd: CUSTOMIZATION_COMMAND DIALOG:RECEIVE env-query -k VIRT/enable DIALOG:SEND **%QStart: VIRT/enable DIALOG:SEND ### DIALOG:SEND ### Please specify value for 'VIRT/enable': DIALOG:SEND ### Response is VALUE VIRT/enable=type:value or ABORT VIRT/enable DIALOG:SEND ***Q:VALUE VIRT/enable DIALOG:SEND **%QEnd: VIRT/enable DIALOG:RECEIVE VALUE VIRT/enable=bool:true

"SEND" is what the host-deploy sends, "RECEIVE" is what the engine replies.

So host-deploy sent a prompt asking for a customization command, the engine sent the command 'env-query -k VIRT/enable', host-deploy then asked the engine to provide a value for 'VIRT/enable', and the engine replied 'VIRT/enable=bool:true'.

Okay. Will try to look into this as well.

...

...

The other errors seem to be related to not having an IP address that ovirt-host-deploy can recognize. To package this for Debian, I would need to find the equivalent of yumpackager.py for aptitude/apt-get/apt, since it seems to be a dependency required by ovirt-host-deploy.

As I said, you can ignore it for now. But IMO this isn't specific to Debian - search a bit and you'll find other similar cases. Alright. Good to know. :)

...

...

TL;DR: How to enable the virt service and assign an IP address that ovirt-host-deploy can use. Write/Find a python script that is equivalent to yumpackager.py and miniyum.py so that that dependency for ovirt-host-deploy is satisfied as well.

Last one will indeed be very interesting, but isn't mandatory for you to continue, if your stated goal is to have a Debian host managed by an oVirt engine. You can manually install all your stuff on the host, and use offlinepackager so that host-deploy will not try to install stuff for you. You'll then have a harder first-time-install, and will miss checking for updates etc. For these you'll indeed need to write something like debpackager, and probably minideb as well - engine-setup uses it directly, as otopi's packager isn't enough for it.

I'd like to mention another point. As explained above, when the engine adds a host, it copies to it a bundle (tarfile). This bundle is not rpm/yum/dnf specific - it should work also on Debian. Normally, ovirt-host-deploy (the rpm package) is installed only on the engine machine. So if you do not care about the engine side for now, you should be able to try adding a Debian host to your engine already - just configure offline packager. This might be easier to debug then by manually running ovirt-host-deploy.

Well, my aim is to run oVirt *on* Debian, similar to how it is run and used by users on say Fedora or RHEL/CentOS. So that's what is guiding me. If your suggestions need to revised in the light of that, do let me know.

- Warm regards Leni Kadali Mutungi

-- Didi

---------- Forwarded message ---------- From: Barak Korren <bkorren@redhat.com> Date: Tue, 16 May 2017 10:06:58 +0300 Subject: Re: [ovirt-users] Building ovirt-host-deploy gives `configure: error: otopi-devtools required but missing` To: Yedidyah Bar David <didi@redhat.com> Cc: Leni Kadali Mutungi <lenikmutungi@gmail.com>, users <users@ovirt.org>

On 16 May 2017 at 09:51, Yedidyah Bar David <didi@redhat.com> wrote:

...

Re (1.), you should also note that currently, the engine is not even packaged properly for Fedora/CentOS. I know this sounds weird, as everyone uses the packages we provide, but it's true - these packages are not compliant with Fedora's packaging guidelines. Why? Because we use maven for building the engine, and package also many dependencies simply as a result of maven getting them from maven central. See also:

https://bugzilla.redhat.com/showdependencytree.cgi?id=1168605&hide_resolved=0

I guess we could walk the middle ground here for DEBs just like we do for RMPs no? Which means - have java dependency bundle packages and release from resources.ovirt.org...

I would love to see a discussion on what it would take to get DEBs through the CI system. I guess we would need some sort of 'build-artifacts.sh.deb' (Or deb8? dep9? not sure what is the convention for Debian version shorthands), some Debian build slaves and Debian equivalents of mock_runner.sh and repoan.

The Foreman project has nice out of tree packages: http://deb.theforeman.org/ that are built via a CI system so having a look at their setup might make sense. If you're using jenkins for CI the jenkins-debian-glue package might be worth a look. Cheers, -- Guido