1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
|
=============
Microversions
=============
As shade rolls out support for consuming microversions, it will do so on a
call by call basis as needed. Just like with major versions, shade should have
logic to handle each microversion for a given REST call it makes, with the
following rules in mind:
* If an activity shade performs can be done differently or more efficiently
with a new microversion, the support should be added to shade.
* shade should always attempt to use the latest microversion it is aware of
for a given call, unless a microversion removes important data.
* Microversion selection should under no circumstances be exposed to the user,
except in the case of missing feature error messages.
* If a feature is only exposed for a given microversion and cannot be simulated
for older clouds without that microversion, it is ok to add it to shade but
a clear error message should be given to the user that the given feature is
not available on their cloud. (A message such as "This cloud only supports
a maximum microversion of XXX for service YYY and this feature only exists
on clouds with microversion ZZZ. Please contact your cloud provider for
information about when this feature might be available")
* When adding a feature to shade that only exists behind a new microversion,
every effort should be made to figure out how to provide the same
functionality if at all possible, even if doing so is inefficient. If an
inefficient workaround is employed, a warning should be provided to the
user. (the user's workaround to skip the inefficient behavior would be to
stop using that shade API call)
* If shade is aware of logic for more than one microversion, it should always
attempt to use the latest version available for the service for that call.
* Objects returned from shade should always go through normalization and thus
should always conform to shade's documented data model and should never look
different to the shade user regardless of the microversion used for the REST
call.
* If a microversion adds new fields to an object, those fields should be
added to shade's data model contract for that object and the data should
either be filled in by performing additional REST calls if the data is
available that way, or the field should have a default value of None which
the user can be expected to test for when attempting to use the new value.
* If a microversion removes fields from an object that are part of shade's
existing data model contract, care should be taken to not use the new
microversion for that call unless forced to by lack of availablity of the
old microversion on the cloud in question. In the case where an old
microversion is no longer available, care must be taken to either find the
data from another source and fill it in, or to put a value of None into the
field and document for the user that on some clouds the value may not exist.
* If a microversion removes a field and the outcome is particularly intractable
and impossible to work around without fundamentally breaking shade's users,
an issue should be raised with the service team in question. Hopefully a
resolution can be found during the period while clouds still have the old
microversion.
* As new calls or objects are added to shade, it is important to check in with
the service team in question on the expected stability of the object. If
there are known changes expected in the future, even if they may be a few
years off, shade should take care to not add committments to its data model
for those fields/features. It is ok for shade to not have something.
..note::
shade does not currently have any sort of "experimental" opt-in API that
would allow a shade to expose things to a user that may not be supportable
under shade's normal compatibility contract. If a conflict arises in the
future where there is a strong desire for a feature but also a lack of
certainty about its stability over time, an experimental API may want to
be explored ... but concrete use cases should arise before such a thing
is started.
|
