On 16 May 2016 at 13:03, Luís Matallui <matallui@gmail.com> wrote:
> First of all, thanks so much for the (really fast) reply.
>
> About adding devices, I got the formats now (yaml, jinja2).
> However, I still couldn't figure out how to actually add those files into
> the database.

The files do not live in the database, they live in the codebase and
are referenced by entries in the database. So create a device_type in
the database as normal, create a device in the database (and mark it
as a pipeline device), then that entry will be used with the files.

https://staging.validation.linaro.org/static/docs/v2/pipeline-admin.html#adding-support-for-a-device-of-a-known-type
https://staging.validation.linaro.org/static/docs/v2/pipeline-admin.html#create-entry-known-type

> In the web administration panel, if I try to add a new device, it doesn't
> allow me to add a jinja or yaml file, even though I can see those options
> under advanced options.

Correct. The database is no place for arbitrarily long lumps of text /
templates. The contents of the device dictionary are displayed in the
device detail page in the admin interface for reference.

The mapping is:

Database: device_type:name -> filename.jinja2 (file in
/etc/lava-server/dispatcher-config/device-types/

The device dictionary is best in a file (for version control reasons)
and is updated in the database using the command line.

> Nevertheless, about the whole process... I guess I'm still trying to
> understand how the whole Lava workflow goes.
> I get the part about syslinux, and you're right, if the target system boots
> syslinux (no matter where that came from), that's our bootloader, which Lava
> doesn't support.
>
> Just so I understand the difference between "deploy" and "boot".
> In this scenario, is "deploy" the part where syslinux gets deployed into the
> system, and "boot" the part where the system would boot once syslinux
> commands select what to boot? Also, if syslinux triggers an installer, is
> that still part of boot?

Can be - it depends on the use case. For simplicity, you can simply
have a very long boot where the installer operates and then the end of
the boot is determined by the installer getting to a login shell. If
you want to track messages from the installer or particularly to track
the booting of the device into the installed system as a boot action,
then that becomes installer-specific.

> From what I got from Neil, I could simply consider that as "boot" process,
> and add a long boot timeout.

You can, yes. It does change how you view the details of that install
but it depends what is your priority - are you primarily testing the
installer or the installed system? Remember: if you say both, that
needs to be TWO separate jobs, not one. Test one element at a time.

> One thing that I haven't been able to get (yet another thing!): I've
> heard/read in a few places about the "Lava Overlay".
> Can I ask what this is? How would I fit this into the picture?

The LAVA overlay is a tarball containing the lava helper scripts and
your lava test shell definitions. e.g.
https://staging.validation.linaro.org/scheduler/job/146127/complete_log?debug=on&section=deploy#R10.1

In order to run the lava test shell scripts, this overlay needs to be
unpacked into the relevant location. If you are testing the installed
system, then there needs to be some way of getting that into the
installed system. That may involve code changes for installers other
than Debian.

> Thanks in advance,
> matallui
>
> On 16 May 2016 at 11:32, Neil Williams <neil.williams@linaro.org> wrote:
>>
>> On 16 May 2016 at 11:25, Matt Hart <matthew.hart@linaro.org> wrote:
>> > On 16 May 2016 at 11:17, Neil Williams <neil.williams@linaro.org> wrote:
>> >> On 14 May 2016 at 01:32, Luís Matallui <matallui@gmail.com> wrote:
>> >>> Hi,
>> >>>
>> >>> For the last couple of days I’ve been trying to setup a device in our
>> >>> lab for deployment and installation of a Linux OS.
>> >>> However, I’ve came across a few issues and lots of questions and
>> >>> hopefully I can get some answers here.
>> >>>
>> >>> Considerations:
>> >>>
>> >>> 1. I’ve decided to give it a go with LAVA v2 and try using the
>> >>> pipeline model
>> >>> 2. I’m not testing any “boards”, so it’s hard to find a similar
>> >>> example
>> >>
>> >> This is a common misconception. Anything which runs tests as part of
>> >> LAVA is a board/device. A DUT - device under test, even if it is a
>> >> virtual machine running on a dispatcher.
>> >>
>> >>> 3. I am not deploying any Linaro images (not even debian based)
>> >>
>> >> For that to be true, you would be testing only the bootloader or
>> >> firmware, e.g. UEFI, Grub etc. There is a system deployed to that
>> >> device - something that looks like POSIX. That system can be
>> >> considered an image (it could be slowly copied off using dd) and it
>> >> can be re-deployed as such when things go wrong. If you get to a login
>> >> prompt and a POSIX shell once the device has booted, that device has
>> >> had an image deployed.
>> >>
>> >> LAVA V2 does away with the concept of a "Linaro" image - if the booted
>> >> device provides a POSIX shell and it is on a machine other than the
>> >> dispatcher - it's a deployment.
>> >>
>> >>> 4. My team develops and supports a reference Linux OS, based on RHEL,
>> >>> so that means we have total control of our images, kickstarter scripts, etc.
>> >>
>> >> RPM based testing is not common in LAVA and the support scripts are
>> >> not routinely tested. This may mean that you need to use some wrapper
>> >> scripts to ensure that the correct dependencies are installed but you
>> >> are still running tests on a deployment after booting a device. You
>> >> have control over the images, so presumably those images are updated
>> >> from time to time - that is an image deployment that LAVA can do.
>> >>
>> >>> 5. We already have a PXE server in our network, which is where our
>> >>> servers (the targets in this context) get booted from by default
>> >>> 6. Once booted from PXE, we get a PXE menu, where we can either select
>> >>> an option, or press ESC and add a custom command (at this syslinux is
>> >>> running)
>> >>> 7. We have access to the serial console of every device via serial
>> >>> port server (telnet <ipaddr> <port>)
>> >>> 8. We have power control over every device via IPMI
>> >>>
>> >>> Issues:
>> >>>
>> >>> 1. I couldn’t find the documentation for how to add devices and device
>> >>> types to the server (which location to add the file, which format -
>> >>> yaml/conf)
>> >>
>> >> In LAVA V2, all configuration lives on the server. The device type
>> >> template is a jinja2 file which is parsed to output YAML.
>> >>
>> >>
>> >> https://git.linaro.org/lava/lava-server.git/blob/HEAD:/lava_scheduler_app/tests/device-types/x86.jinja2
>> >>
>> >> Adding devices to V2:
>> >>
>> >> https://staging.validation.linaro.org/static/docs/v2/pipeline-admin.html#
>> >>
>> >>> 2. In the above described environment, I suppose we would skip the
>> >>> deployment phase, since the devices already boot into syslinux from PXE (is
>> >>> this correct?).
>> >>
>> >> Not necessarily.
>> >>
>> >>
>> >> https://staging.validation.linaro.org/static/docs/v2/dispatcher-design.html#persistence
>> >>
>> >> You are better considering if you should actually deploy a fresh copy
>> >> of the image each time to ensure you know what you are actually
>> >> testing. Hint: only ever change one element at a time - it may sound
>> >> obvious but you need to be obsessive about this.
>> >>
>> >>> Either way, it would be nice to be able to run ‘ipmitool chassis
>> >>> booted pxe’ before rebooting the system.
>> >>
>> >> That is effectively your hard_reset_command in the Device Dictionary.
>> >>
>> >>
>> >> https://staging.validation.linaro.org/static/docs/v2/glossary.html#term-device-dictionary
>> >>
>> >> https://staging.validation.linaro.org/static/docs/v2/pipeline-admin.html#create-device-dictionary
>> >>
>> >> https://staging.validation.linaro.org/static/docs/v2/pipeline-admin.html#
>> >>
>> >> There are examples of x86 device dictionary entries but those
>> >> particular x86 devices are rather specialised and the device
>> >> dictionary includes lots of potentially confusing elements.
>> >>
>> >>> 3.  Either way (via boot or deploy), how can I make sure to detect the
>> >>> PXE (syslinux) menu, send the ESC key, and send the command I need to
>> >>> trigger the kickstart installation?
>> >>
>> >> LAVA V2 has iPXE support.
>> >>
>> >>
>> >> https://git.linaro.org/lava/lava-dispatcher.git/blob/HEAD:/lava_dispatcher/pipeline/test/sample_jobs/ipxe-ramdisk.yaml
>> >
>> > Can I just jump in here and state, like I do rather often, that the
>> > fact this board uses PXE to load syslinux, has nothing to do
>> > whatsoever with iPXE support.
>> > PXE != iPXE, iPXE is just another bootloader. How the board gets to
>> > that bootloader is largely irrelevant.
>> > The real support needed here is "syslinux", but V2 doesn't yet have a
>> > generic bootloader support which would mean this could be driven by
>> > just config files.
>>
>> True - I end up using syslinux for other projects and it can be a pain
>> to drive properly. Grub (specifically grub2) is a lot easier - if you
>> are open to that level of change in your tests, it is worth doing.
>>
>> > You could of course use iPXE to boot your installer, replacing
>> > syslinux, but I would imagine that's part of the things they are
>> > trying to test.
>>
>> >>> To sum-up, the Workflow I’m trying to achieve after having completed
>> >>> the whole setup sort of goes like this:
>> >>>
>> >>> 1. Reboot target device into PXE (the PXE itself will download and
>> >>> start syslinux here)
>> >>> 2. Wait for PXE menu (expect some sort of string)
>> >>> 3. Send ESC key (to get into the boot shell)
>> >>> 4. Send command (this will trigger our kickstarter script and the
>> >>> installation will take around 5 minutes). The images that are needed are
>> >>> fetched automatically from our sftp server.
>> >>
>> >> This installation *is* a deployment. LAVA V2 has support for the
>> >> Debian Installer (because most LAVA instances are based on Debian) but
>> >> that support isn't going to work for a Linux OS installer which is not
>> >> Debian-based. You might be able to support this with a simple (long)
>> >> timeout or you could look at the Debian Installer support and
>> >> implement something related for your installer.
>> >>
>> >>> 5. Wait for boot prompt after system installation succeeds
>> >>> 6. Login using credentials (username, password)
>> >>
>> >> auto_login:
>> >>
>> >>
>> >> https://git.linaro.org/lava/lava-dispatcher.git/blob/HEAD:/lava_dispatcher/pipeline/test/sample_jobs/bbb-nfs-url.yaml
>> >>
>> >>> 7. Run whatever tests we need from here
>> >>>
>> >>> Any help here would be much appreciated.
>> >>> Thanks in advance!
>> >>
>> >> Documentation for these steps is currently being updated and reviewed.
>> >>
>> >> Currently, LAVA (V1 and V2) require a Debian based system - are you
>> >> implementing such support or is your team interested in porting the
>> >> Debian packaging to RHEL /RPM support? Note that there will be a *lot*
>> >> of dependencies which would also need to be packaged before starting
>> >> on the LAVA codebase. You will also need to maintain those
>> >> dependencies, update them and keep them inline with the LAVA software
>> >> which is tested on a Debian system.
>> >>
>> >> Your use case is supportable but it will need work at your end as we
>> >> are testing on a different environment.
>> >>
>> >> --
>> >>
>> >> Neil Williams
>> >> =============
>> >> neil.williams@linaro.org
>> >> http://www.linux.codehelp.co.uk/
>> >> _______________________________________________
>> >> Lava-users mailing list
>> >> Lava-users@lists.linaro.org
>> >> https://lists.linaro.org/mailman/listinfo/lava-users
>>
>>
>>
>> --
>>
>> Neil Williams
>> =============
>> neil.williams@linaro.org
>> http://www.linux.codehelp.co.uk/
>
>



--

Neil Williams
=============
neil.williams@linaro.org
http://www.linux.codehelp.co.uk/