The '--hint' option for 'server create' expects a key-value pair like so:

  openstack server create --hint group=245e1dfe-2d0e-4139-80a9-fce124948896 ...

However, the command doesn't complain if this isn't the case, meaning
typos like the below aren't indicated to the user:

  openstack server create --hint 245e1dfe-2d0e-4139-80a9-fce124948896

Due to how we'd implemented this here, this ultimately results in us
POSTing the following as part of the body to 'os-servers':

  {
    ...
    "OS-SCH-HNT:scheduler_hints": {
      "245e1dfe-2d0e-4139-80a9-fce124948896": null
    }
    ...
  }

Which is unfortunately allowed and ignored by nova due to the use of
'additionalProperties' in the schema [1]

Do what we do for loads of other options and explicitly fail on invalid
values. This involves adding a new argparse action since none of those
defined in osc-lib work for us. This is included here to ease
backporting of the fix but will be moved to osc-lib in a future patch.

Conflicts:
	openstackclient/tests/unit/compute/v2/test_server.py

NOTE(stephenfin): Conflicts are due to the absence of tests for changes
Ic06d97b29e51828b29d7ac5172645c288e4ada9e ("Compute: Add description
support for server") and If188c3d96fa506dbe62ef256418f2f9bca1520c2 ("Add
host and hypervisor_hostname to create servers"), neither of which we
want to backport.

[1] https://github.com/openstack/nova/blob/19.0.0/nova/api/openstack/compute/schemas/servers.py#L142-L146

Change-Id: I9e96d2978912c8dfeadae4a782c481a17cd7e348
Signed-off-by: Stephen Finucane <sfinucan@redhat.com>
Story: #2006628
Task: #36840
Related-Bug: #1845322
(cherry picked from commit ea27ebb0f918db9eab2f5751a1b065818faa0e6d)
(cherry picked from commit 576d9ecfd0ac27611ec89af6d34e7378b7a39fb3)

This patch removes using the "name" option for a marker when
--deleted is also used. The find_resource() function
that is being called does not correctly handle using the marker
as the "name" in the search when also using deleted=True.
One simple way to fix this is force the marker to only be an ID
when --deleted is used. This is how the nova client works.

Using the --deleted option is available to users with the admin
role by default. If you're an admin listing --deleted servers
with a marker by name, find_resource() is going to fail to find
it since it doesn't apply the --deleted filter to find_resource().

The find_resource() function is trying to find the marker server
by name if it's not found by id, and to find it by name it's
listing servers with the given marker as the name, but not
applying the --deleted filter so it doesn't get back any results.

In the story it was suggested modifying find_resource to include
the deleted query param when it's specified on the command line but
that didn't work because it still results in something like this:

http://192.168.1.123/compute/v2.1/servers?deleted=True&name=4cecd49f-bc25-4a7e-826e-4aea6f9267d9

It seems like there are bugs in find_resource().

Restricting the marker to be the server ID when listing deleted servers
is probably OK since if you're using --deleted you're an admin and you could
be listing across all projects and if you're filtering by a server across all
projects anyway (not that you have to, I'm just saying if you are), or even
showing a server in another project, you have to do it by id rather than name
because find_resource() won't find the server in another project by name, only ID.

story: 2006761
Task: 37258

The server test case was removed from the cherry-pick as it used
features of the openstack client that are not in stable/stein

Change-Id: Ib878982b1d469212ca3483dcfaf407a8e1d2b417
(cherry picked from commit c42c27aa926e93a2507e686eeaf0c5510a29d245)

Fix various things so the functional tests will work under python3:

- A hashlib.md5() can only be update()d with an encoded string in py3.
- There's no dict.iteritems(), change to dict.items() (which is already
  an iterator).
- Open temp files with 'w+' mode rather than the default 'w+b' (as an
  alternative to encoding all the write and expected-read payloads as
  bytes).
- (This is a weird one) Explicitly raise SkipTest from unittest (rather
  than unittest2, which is where cls.skipException landed). Not sure why
  this is busted, but this moves the ball.

Conflict/issue with raising SkipTest on this branch.

(cherry picked from commit f1d742f32adeb662a3fdf8fa3ef3bc391e71ed81)
(cherry picked from commit b866202f54afddca66a77cd989b082e193a96956)

Includes squash of:
Before writing object data to stdout, re-open it in binary mode

Otherwise, you can hit TypeErrors on Python3.

Closes-Bug: 1775482
(cherry picked from commit 415b48056d9d021e04ec972029040a89a6b13928)

Change-Id: Ic9b2b47848a600e87a3674289ae7ae8c3e091fee

With compute API microversion 2.53, nova-compute services
can only be deleted with the ID as a UUID to uniquely identify
the service in a multi-cell deployment. This documents that for
the "compute service delete <service>" argument. The description
of the "compute service list" command is also updated to mention
that the ID can be retrieved as a UUID using 2.53 or greater.

Change-Id: If7d4a27c0aaef588bcd77dd9edddec1e535fbf31
Story: 2005349
Task: 30302
(cherry picked from commit 1557afb554e908e097abd39081891ea78083e20e)

With compute API microversion 2.53 there is a single
PUT /os-services/{service_id} API which takes the service
id as a UUID. Since the openstack compute service set
command only takes --host and --service (binary) to identify
the service, this change checks if 2.53 or greater is being
used and if so, looks up the service by host and binary and
calls the appropriate methods in novaclient.

If the command cannot uniquely identify a compute service
with the given host and binary, an error is raised. A future
change could add an --id option to be used with 2.53+ to
pass the service id (as UUID) directly to avoid the host/binary
filtering.

Change-Id: I868e0868e8eb17e7e34eef3d2d58dceedd29c2b0
Story: 2005349
Task: 30302
(cherry picked from commit 4bd53dc1090fda86f6ce25b76a079e250c9206d8)

When the --name-lookup-one-by-one option passed to the 'server list'
command, the image and flavor names will be looked up for each
server being listed instead of fetching all image/flavor names.

The current code assumes all servers have an image attribute, but
servers booted from volumes have no image, so the following error is
raised when listing BFV servers with --name-lookup-one-by-one:

  AttributeError: ('unicode'|'str') object has no attribute 'get'

The error occurs when the code attempts server.image.get('id').

This fixes the --name-lookup-one-by-one code not to assume an image
for a server. The unit tests for 'server list' have also been
robustified to feature one BFV server to enhance our test coverage.

Story: #2006063
Task: #34777

Change-Id: I312c971346c7ded93f6fcaa515098554b8580295
(cherry picked from commit bfc34e11b3437506508b3e120accc0e212268ac6)

There is a typo in the endpoint group deletion, due to this you
can't remove endpoint groups once assigned. I am adding also the
unit tests to avoid this kind of issues in the future

Task: 30640
Story: 2005521
Change-Id: Ie938f2c9894bb39b4c0ed1f7aa3a6a751a303058
(cherry picked from commit 04e03b2a1fe34046e2148d3c1d17b9053010c8af)

If a required service is not enabled then we skip the test.
The discovery is done by tests/functional/base.py:is_service_enabled
but this method is broken, credentials are not passed to the
'openstack service show' command so every call will fail and every test
that relies on it will be skipped. This commit fixed that method and
the issues that popped up when re-enabling tests.

Network segment range:
 - issue where we assumed network-segment-range extension is always
   present
 - issue where we compare integers and string representations of numbers

Subnet:
 - issue where we try to deepcopy an uncopyable object in UnsetSubnet

Change-Id: Id3cc907c1ed2a25b49cf6f4a7233e0401a02383a
Story: 2005169
Task: 29908
(cherry picked from commit 7741347041b078ca4d687597897194d7797d202d)