50 was too much now that we keep snapshots in tests.

The supervisord binary in some environments (like webrunner), seems to not
follow the automatic way to find the configuration file, so instead of relying
on it, just explicitly provide the file.

/reviewed-on !168

Now that slapos node instance outputs summary of problems, no need to
pollute the logs so much

destroying instance is supposed to remove these, but when there are
issues removing instances we sometimes have supervisor config left from
previous instances.

Reformating now delete theses after logging the problem.

maybe this is something slapos node instance should do instead ...

When having more than one read_only_shared_part_list, the template was
first rendered with indentation and then dedented, but since dedent only
remove the common indentation, we ended up with slapos.cfg like this:

```
$ cat /data/slappart11_testnode/cqg/inst/test0-9/tmp/etc/slapos.cfg

            [slapos]
            software_root = /data/slappart11_testnode/cqg/inst/test0-9/tmp/soft
            instance_root = /data/slappart11_testnode/cqg/inst/test0-9/tmp/inst
            shared_part_list =
              /srv/slapgrid/slappart11/srv/shared
/srv/slapgrid/slappart11/srv/testnode/cqg/shared
              /data/slappart11_testnode/cqg/inst/test0-9/tmp/shared
            master_url = http://10.0.12.198:46312
            computer_id = local
            root_check = False
```

First dedent, then render the template to prevent this issue. Apply this to
other uses of textwrap.dedent for consistency.

/reviewed-on !140

connection_dict generated by client can be different in details from the
server side, so pass it thorugh a way how it is treat on the server.

Also, as we are going to be compatible with py3, calculate hash from
sorted items of a dict, instead of relying on side effect of py2 "ordered dict"

https://portingguide.readthedocs.io/en/latest/dicts.html#changed-key-order

Test are focused on client side, and the tricky cases are covered, they
somehow contain the protocol of client <--> server comparision.

All tools from the ecosystem shall use the same way to convert between python's
dict and XML.

Additionally stabilise generated XMLs by sorting the dictionary.

A set of utility classes to run a local slapos computer and control
softwares/instances on this computer

Set proper limit.

This is follow-up of !127

/reviewed-on !128

  This save us to call a jio_get later per line in a loop

  The API must relies on Hateoas API based on JIO only.

  Ensure that path is correct if hateoas_url don't finish with '/'

  Now a user can get a token from a configured slapos client or slapos console.

  This might evolve into a full jio api to written in python

  Avoid to excessively post to master if the state is the same as the once expected,
  this decreases significantly the number of posts to master.

proxy was just removing the software from database when destruction was
requested. Instead, keep it as requested_state = "destroyed" and return
this information to the node, so that nodes properly destroy the
software.

checked with:
  pylint --disable all --enable spelling --spelling-ignore-words=URI,SSL,API,XXX,uninstall,initialize,slapgrid,slapformat,kvm,instantiation,IComputer,IException,Vifib,IRequester,ISupply,SocketError,virtio,boolean,traceback,http,mysoftwareproduct,foo,buildout --spelling-dict en_US slapos/slap/interface/slap.py

/reviewed-on !66

  The API that notify availability and building from Computer Partition is unecessary and it's
  implementation overuse resources and cause flaky effect on the state of the Computer Partition.

  At the Software Release, the states can be available/building/error, and this state tens to be
  immutable once the software release is finished (it will be available, where building and error are
  transitory states).

  At Computer Partition, we intent to have started/stopped/destroyed instances, where available and
  building were transitory, but they have no practical function as a second report would come quickly
  after. So we may consider that started is actually same as available, in this case, it is a duplicated
  meaning, as you cannot be available w/o been started/stopped.

  computer_partition.building were never used, as the master knew that a partition is been processed,
  so this API were never used.

  Report transitory states are prevent us to determinate the actuall state, considering that the latest
  actuall state is more relevant (started/stopped/destroyed).

slapos/slap/slap.py slap.initializeConnection: Cache master node's Hateoas URL and do not ask many times. Communication to master node is expensive.

slapos/slap/slap.py: Fix ComputerPartition.setConnectionDict. Declare connection_parameter_hash explicitly, UnboundLocalError may occur.

Buildout cannot request destroy sub instances when it is removed from
instance parameters by the user. Ex: request a cluster of KVM, with
kvm1 and kvm2 instances, edit parameters and remove kvm2 in the list.
Buildout will just ignore that instance and it will be removed only when
the hosting subscription will be destroyed.

                 -- kvm1
root instance --|
                 -X- kvm2

So when processing a partition, we send to master the list of requested sub
instance(s). If the previous list was ['kvm1', 'kvm2'], then after remove kvm2
the list will be ['kvm1'] (for the root instance). Commit this list to the
master will unlink between root instance and kvm2 (predecessor categorie).

A new alarm will search for unlinked instances: Instance which has no parent,
and destroy them (also destroy the sub tree).