Automatic/Daily builds for OSGeoLive

[kalxas]

Having completed the new build system for OSGeoLive 6.0 (build_chroot.sh and inchroot.sh) we are now in a position to create a simple cron job to make automatic builds. This can be assigned to a base Xubuntu VM as described in ​http://wiki.osgeo.org/wiki/Live_GIS_Build#How_to_do_development_.2F_debugging_with_the_current_build_method

A last issue that we need to tackle is that after each build the base VM needs a reboot to correctly unmount the /dev folder. This problem is discussed here: ​http://osgeo-org.1560.n6.nabble.com/Build-Process-Filesystems-td5024288.html with a possible fix included.

[kalxas]

Closed #778 as duplicate

[kalxas]

Possible fix from mailing list committed in [9529] but unfortunately did not work as seen in ​http://aiolos.survey.ntua.gr/gisvm/dev/osgeo-live-debug-build9529-log.tar.gz

Any ideas?

[hamish]

Replying to kalxas:

but unfortunately did not work as seen in ​http://aiolos.survey.ntua.gr/gisvm/dev/osgeo-live-debug-build9529-log.tar.gz

Any ideas?

Finished chroot part
======================================
umount: /home/user/livecdtmp/edit/proc: device is busy.
        (In some cases useful info about processes that use
         the device is found by lsof(8) or fuser(1))

maybe follow the advice and put a lsof &/or fuser in the build script to see what it does?

where is the script run from? if the pwd is in the tmp dir, is it still attached even if you 'cd ..' out of it within the script?

Hamish

[kalxas]

Hi Hamish,

The script is executed from /livecdtmp which is top level directory. I will double check.

Angelos

[kalxas]

not enough time to tackle now.

[kalxas]

Good news, the error message after the chroot part is no longer there in Xenial, which means that the build process does not teardown the build host anymore (which required a manual reboot until now).

I have prepared a small bash script to trigger a git pull, an iso build, log post-processing and finally an automated upload of the resulting files to a remote server, and it worked flawlessly, leaving the build host in a perfect state.

I think the best way forward would be to replace the existing bootstrap.sh script with a build.sh which will automate the whole build process. (Actually the bootstrap.sh will remain and still be used internally)

What is now missing is some git magic to figure out when a cron-job should trigger a build. After that we can have a completely automated build process.

I also think this is a good time to re-organize the installers. We had some discussion with Brian some months ago, we should discuss this during a project meeting.

[kalxas]

The chroot problem has been finally fixed with this commit:

​https://github.com/kalxas/OSGeoLive/commit/d11466972812d3c4025b79a4b4b61c099c606444

Build process is now fully automated.