[Buildroot] [ANNOUNCE] Autobuilder script available, join the autobuilder effort!

[Thomas De Schampheleire]

Hi Thomas,

On Sat, Jun 14, 2014 at 9:23 PM, Thomas Petazzoni
<thomas.petazzoni at free-electrons.com> wrote:
> Hello,
>
> Most likely the majority of the Buildroot developers are aware of the
> Autobuilder mechanism we have, for which the results are available at
> http://autobuild.buildroot.org/ and posted daily on the mailing list.
>
> Until now, both Peter and I were using ad-hoc scripts that were so ugly
> that they were not published, so it was not easy for others to join the
> autobuilder effort.
>
> I'm therefore happy to announce that I've written a completely new
> version of a Python script that allows to run an autobuilder instance
> on a given machine. It's a single script, and its installation is very
> easy (though one must of course pay attention to have the Buildroot
> pre-requisites installed). So it means that from now on, it's a lot
> easier for others than Peter and I to set up an autobuilder instance.
> If you have some spare cycles on a server, or on your workstation, even
> for just a few hours per day, this might be of interest to you!
>
> The script is available at:
>
>   http://git.buildroot.net/buildroot-test/plain/scripts/autobuild-run
>
> (NOTE: I am *not* a Python developer, there are very possibly many
> possible improvements to do in the script. Error handling and logging
> is probably not at its best. Patches welcome!).
>
> Getting started
> ===============
>
>
> To get started, you can simply run it with:
>
>   ./autobuild-run --ninstances 2 --njobs 3
>
> This will run 2 parallel independent Buildroot builds, each using 3
> parallel jobs (i.e BR2_JLEVEL=3). In this configuration, autobuild-run
> will store the results in local tarballs (one for each build), they are
> not submitted to the http://autobuild.buildroot.org website.
>
> This mode is a good way to test if your autobuilder instance is working
> as expected. It is also a way for people to help improve the script
> without necessarily running a full-blown autobuilder instance, and also
> to verify that the changes are
>
> What's going on
> ===============
>
> Each instance runs in its own instance-<number>/ directory, with inside:
>
>  - dl/, the download directory for this instance. In the current
>    Buildroot, we cannot share the same download directory between
>    parallel instances. This is being fixed by Yann.
>
>  - buildroot/, the Buildroot source code.
>
>  - output/, the build output.
>
> To see what's going on while builds are running, there are two level of
> log files:
>
>  - instance-<number>/instance.log, which persists across builds and
>    gives some high-level information on what's going on (build
>    starting, build ended with success or not, etc.)
>
>  - instance-<number>/output/logfile, which is valid only for the
>    duration of the current build, and gives the usual Buildroot output.
>
> Getting results submitted
> =========================
>
> If you want the results of your autobuilder instance to be submitted to
> http://autobuild.buildroot.org, you should first contact me to get some
> credentials to submit those results. Once this is done, you need to
> write a configuration file that looks like this:
>
> [main]
> ninstances = 2
> njobs = 3
> http-login = <the-login>
> http-password = <the-password>
> submitter = Your Name (name of machine)
>
> And start your autobuilder instance using:
>
>   ./autobuild-run -c <configuration-file-path>
>
> And that's it.
>
> Testing coverage
> ================
>
>  * The architecture/toolchain combinations are chosen randomly amongst
>    a set of pre-defined configurations, listed at
>    http://autobuild.buildroot.org/toolchains/configs/ (see the .csv
>    file).
>
>  * The packages are chosen randomly using 'randpackageconfig', and
>    there are some exceptions encoded in the script to avoid known
>    problematic configurations.
>
> Current deployment
> ==================
>
> The script is currently deployed on two machines:
>
>  - The Free Electrons build server, on which it replaces the old crap I
>    was using to do the autobuilds.
>
>  - Another private server of mine, which is the one actually hosting
>    the http://autobuild.buildroot.org. I'm not sure I will let builds
>    run forever on this machine, but it was useful to have two machines
>    using the script to test in different conditions.
>
> It is worth mentioning that due to this deployment, there are a few
> bogus results on http://autobuild.buildroot.org due to missing packages
> on the new build server, issues in the script or for other reasons.
> This should hopefully settle down in the next few days.
>
> Hopefully, Peter will join the dance and convert gcc10, gcc20 and
> gcc110 to use the script as well.
>
> Contribute
> ==========
>
> Feel free to test the script, help improve it, and send patches. It is
> available in the Git repository at:
> http://git.buildroot.net/buildroot-test/.


First of all thanks a lot for making this available!
I am currently testing the script, and hope to get permission to run
it during the night on one or more work machines.
I had to make one modification currently: since we are behind a
firewall, running 'git' directly is not possible, but I need to use a
wrapper script. I added this:

+try:
+    gitcmd = os.environ['GIT']
+except KeyError:
+    gitcmd = 'git'

at the top of the script, and replaced all direct occurrences of 'git'
with gitcmd.
This allows me to run 'GIT=git-wrapper autobuild-run'.

I used an environment variable because it seemed superfluous to add a
configuration variable for it, but if you prefer I can do that too.

Is there an easy way to get a web interface like that on
autobuild.buildroot.org, for the locally created builds? This would
make it easy to inspect the results while testing (for example to see
that some host package like 'bzr' needed to be installed, ...).

Best regards,
Thomas