powerdns uses a socket named pdns.controlsocket in socket-dir (
https://doc.powerdns.com/authoritative/running.html#control-socket )
We sometimes see errors on test nodes that this path is too long:

     Aug 13 20:54:10 Unable to open controlsocket, path '/srv/slapgrid/slappart13/srv/testnode/cvr/inst/test0-0/tmp/inst/pdns1/var/run/pdns-socket/pdns.controlsocket' is not a valid UNIX socket path.

Instead of using "var/run/pdns-socket", put the socket in "var/run",
which should be short enough.

As cpu-count and ram-size can be setup freely by the user, the cpu-max-count
and ram-max-size shall follow it.

Otherwise it generates hard to track issues with starting VMs.

The image-url-list parameter is passed-thru only if it's present in the
original request, as the default instance differentiates correctly between
parameter existence or emptiness (or being None).

Downloadable images are enabled when key image-url-list (described in
instance-kvm-input-schema.json) is present.

Images are downloaded outside of partition processing, as this can take a lot
of time by template/image-download-controller.py

Configuration is checked and cleaned up by
template/image-download-config-creator.py

Promises are used for:

 * checking if the current configuration from the request has been processed
 * checking status of configuration generation, image download process and
   checksum validity

Details about errors are exposed by using monitor stack provided HTTP server,
so that user is able to take informed decision about how to fix the problem.

Wait even longer, as in some environments downloading big files takes a lot
of time which can lead to unstable test results.

Note: On fast environments the max won't be reached, so in such case the test
      speed won't be affected.

Interesting HTTP server from https://h2o.examp1e.net/

Some tests are checking deeply that bootstrap script works correctly inside
if prepared VM, but those prepared VMs are big chunks of data to be fetched,
so it's required to wait quite long.

   This contains data for running tests on Upgrade

This reverts 80af2841 and b4f83dd0.

We will force to compile gcc only in the machine preparing the source but not in the OBS machines actually compiling the packages.

The PYTHON variable is used in Makefile (https://gitlab.gnome.org/GNOME/glib/-/blob/2.58.3/glib/Makefile.am#L463) to generate the shebang for script gtester-report. With component/defaults.cfg, we have PYTHON=python2.7 so the shebang become '#!python2.7' which is an error for building a package on Fedora/SUSE:

[ 3269s] *** ERROR: ./opt/slapos/parts/glib/bin/gtester-report has shebang which doesn't start with '/' (python2.7)

The simplest solution is to remove this tester script as it is not needed and have been removed in newer version of glib already. This patch can thus be reverted when upgrading glib version.

Introduce an [editorconfig] configuration file to tell developer editors that we want an empty line at end of file.

After making softwares and component uses a buildout.hash.cfg file, run [eclint] fix to apply the style.

eclint is also added in the commit hooks (that one can install with `npm install` from the root of this repository)

[editorconfig]: https://editorconfig.org/
[eclint]: https://github.com/jedmao/eclint

See merge request nexedi/slapos!757

It seems `eclint fix folder/file` sometimes dos not work, but `eclint
fix folder/` worked.

done with:

     npm install
     git ls-files | grep -v '\/unstable\/' | xargs ./node_modules/.bin/eclint fix
     git ls-files | grep buildout.hash.cfg | xargs ./update-hash

So that we all edit files in a consistent style

Sometimes, gitlab and qemu are reported by software release tests to link against libcap-ng and libzma from system. This happens when the development packages are installed on the system.

Approach here is to build with slapos version of the libraries.

See merge request nexedi/slapos!789

Using an script to set the environment is easier to use, but the main
reason is that if we want to be able to use shared part from outer
slapos (like when using in theia or slaprunner), the working directory
for test must be contained in the instance directory of the outer
slapos, so that when we run slapos node prune in the outer slapos this
command can see that shared parts are used.

This was using zlib and lzma from system if they are available, so we
tell bundler to use the slapos version.

We have to use both in --with-cflags for nokogiri itself and as CFLAGS
environ for when nokogiri build its included libxml2

don't let configure autodetect and build something different depending
on the presence of this library on the host system

Explicitly build with libcap-ng instead of depending on presence of the
development library on the host.

Also fix the way of disabling tinfo, there was warning at the end of
configure saying this option is not used

Backend logs are exposed as usual access and error logs.

By using rsyslogd templates and regex filtering, the rsyslogd reacts itself
and creates needed files per each slave which accesses it. Thanks to this, it's
configuration is static from point of view of SlapOS profiles, and can be
generated once.

As the rsyslogd configuration became fully special to backend-haproxy, the
rsyslogd template filename and its references has been correctly renamed.