In our normal operations, we are not using PYTHONPATH but using buildout
to generate wrappers. We are still using python setup.py test to run
tests, which sets PYTHONPATH to the eggs used by this python and this
cause issues when this python starts subprocesses, especially when it
starts another version of python which is supposed to use a different
set of eggs.

On python3.6 this raises:

 sqlite3.OperationalError: cannot VACUUM from within a transaction

https://www.python.org/dev/peps/pep-0508 is less verbose.

This reverts commit 03d0be06.

In some cases the debian distribution version string can be returned uppercase.

  vacuum as little as possible, as it is a heavy operation to perform.

  To allow more concurrent access to the sqlite

actually ansible uses /usr/bin/python as defaut interpreter. Use this
temporary workaround until we can set properly the ansible python
interpreter. See https://docs.ansible.com/ansible/latest/reference_appendices/python_3_support.html

In general we should only setGroupReference when the generation matches, as the builder can generate multiples SPLs at once.

/reviewed-on !183

…t makes no sense to test it

/reviewed-on !182

Our condition was only checking executables, which missed some non
executable libraries.

Following up on !172 (comment 96033)

/reviewed-on !180

@luke , As we discussed. Merging dierctly.

Tests passing in
https://nexedi.erp5.net/test_result_module/20200117-9F5C0733

/reviewed-on !181

Slave will display state of the Software Instance which it is attached to.

The reported state is informed by the partition giving user more accurate view on the state, so it can know if the partition was processed or not.

  This allow us define if the instance is failing to change state (start/stop) or failing after started or stopped

  Invoke activities are not expected on this test, as there are
  others partitions that can allocate before we call the API we
  would like to test.

/reviewed-on !179

During teardown, we sometimes have errors that processes are still
running after being stopped, something like:
    
    Fault: <Fault 91: 'STILL_RUNNING: TestTelegraf-0'>

This is because we only retried `slapos node report` two times, but
`slapos node report` tells supervisor to stop processes without
waiting and when processes takes time to stop, this is sometimes not
enough. Increase to retry up to 10 times by default.

/reviewed-on nexedi/slapos.core!178