Previous gitlab version was 8.17.0, this commit upgrade gitlab to version 9.5.10,
and update all configuration to work with this version according to the installation procedure,
see: https://gitlab.com/gitlab-org/gitlab-foss/-/blob/v9.5.10/doc/install/installation.md

If pgdata already exists, then postgres recipe will not update or install anything when buildout will run. This result to no configuration or wrapper updated if configuration changes.

For example this change:  29d1ed68  the path of `posgtres bin` is now in shared folder, then impossible to run postgres service anymore because path was not updated by buildout in wrapper.

/reviewed-on nexedi/slapos!703

Since 10.4, it seems mysql_install_db does not take anymore
BASEDIR/lib/plugin as default value for plugin_dir.

Shamefully copied from neoppod/stress-testing SR.

Only TestSlave class and its subclasses are tested, as those are enough
complex scenarios: master partition, kedifa partition and frontend partition.

This allows to use monitor-setup-url correctly.

Instead of forcing to set monitor port in some cases, just generate them,
so it's possible to correctly instantiate caddy-frontend on one partition
scenario like in webrunner or tests.

This avoids duplication of information.

The cluc from https://docs.trafficserver.apache.org/en/7.1.x/admin-guide/logging/formatting.en.html#cache-details
allows to track the log entry per the originated request, which basically
makes debugging of ATS behaviour possible.

/reviewed-on !689

/reviewed-on nexedi/slapos!633

This reverts commit 2cc61477.

Now we now, without "futures" egg, gunicorn refuses to start:

    Error: class uri 'gunicorn.workers.gthread.ThreadWorker' invalid or not found:

    [Traceback (most recent call last):
      File "/srv/slapgrid/slappart14/srv/testnode/cqc/inst/test0-0/tmp/soft/8c22f5a21e7f016f38916f098ab2339b/eggs/gunicorn-19.7.1-py2.7.egg/gunicorn/util.py", line 134, in load_class
	mod = import_module('.'.join(components))
      File "/srv/slapgrid/slappart14/srv/testnode/cqc/inst/test0-0/tmp/shared/python2.7/8d8826883d34c5a7ec7a5b8c187d89c9/lib/python2.7/importlib/__init__.py", line 37, in import_module
	__import__(name)
      File "/srv/slapgrid/slappart14/srv/testnode/cqc/inst/test0-0/tmp/soft/8c22f5a21e7f016f38916f098ab2339b/eggs/gunicorn-19.7.1-py2.7.egg/gunicorn/workers/gthread.py", line 37, in <module>
	""")
    RuntimeError:
	You need to install the 'futures' package to use this worker with this
	Python version.

There have been a few fixes since last time we update.

nexedi/cloudooo@0ff799eb...67e233f2

/reviewed-on nexedi/slapos!698

Yet another non-obvious change that was committed without explanation.
Let's revert and see what breaks...

- lsmod is wrong because kvm_intel may be built inside the kernel
- kvm_intel is wrong because it only works for Intel CPUs
  (for exemple, there exists kvm_amd)
- /dev/kvm may have wrong permissions

check-maximum-elapsed-time and failure-amount are added to support
monitoring distant backends with some margin on the failures.

This fixes errors like:

    File "python3.6/lib/python3.6/re.py", line 142, in <module>
      class RegexFlag(enum.IntFlag):
  AttributeError: module 'enum' has no attribute 'IntFlag

hexagonit.recipe.download-1.7.post4 (Nexedi's patch) is not compatible
with Python 3 (yet?).

In order to avoid the very long #!/.../python path, call surykatka with
explicit interpreter.

By default one failure is considered as bad thing, but allows to select sites
which shall be treated with some tolerance.

Allows to configure maximum time for which the site is considered correct.

For each class (the maximum elapsed time value) separate surykatka process
is started, in order to not impact fast sites checking by long timeouts on
slow sites.