Some CRDs may have a status element at their root. Thus we need a corresponding field

Read only fields were meant to be read only. However sometimes these fields must be set (in the case of an operator managing these resources for example).

We should add a mean to edit these fields with a context manager for example:

with pykorm.fields.UnlockReadOnly:
  obj.read_only = 42

The method would compare the uid and each individual field with the other model

Pykorm is rather slow when dealing with a large number of objects. In my case, at ~850 objects, an .all() query takes around 20s !

>>> a = time.time() ; len(list(MyCRD.query.all())); print(time.time() - a)
851
21.512641668319702

in comparison, kubectl takes only 1s:

$ time kl get MyCRD.p6k.ch -A 
real	0m1.392s
user	0m0.536s
sys	0m0.182s

A watch function implemented as a generator would be really nice to have

May be related with @YiannisGkoufas issue

AnyISalIn's MR #11 introduced nested fields. It wasn't merged in #13 because some tests didn't pass and the change would've been too big.

However the idea was really good and they should be inttroduced into pykorm

Very useful tool!
Was wondering whether it would be useful to have a utility that takes a CRD definition in yaml format and produces the necessary pykorm models.

I'm currently looking into using pykorm for a small project that I'm building with kopf. So far I've been using pykube but I'm not really happy with it's API.
pykorm looks very promising as it allows to properly map my CRD & core k8s resource to python objects.

One thing that are not clear to me or maybe might be missing are default models within pykorm for all the existing core resources.
Based on the example below i assume that i would need to build the models on my own?

If that is the case I'd propose to build these models & then create a PR to merge them into this lib

In most cases the status field in the root is a subresource and can't be updated with the same api as spec or metadata. There are separate APIs like create_namespaced_pod_status, patch_namespaced_custom_object_status, etc.

I've worked around this by creating a mixin that I add to all model classes and provides a method that calls the relevant *_status variant of the APIs but it would be nice if this could be supported by pykorm directly. Either by calling both on save()/apply() if there is a Status field in the model or even a separate way (pk.status().save(obj), pk.save_status(obj), ...).

We should add the ability to choose a specific kubeconfig (when connecting to different cluster for example).

pk = pykorm.Pykorm(kubeconfig_file='foo/bar.yaml')

When creating aand saving a model with some empty (non mandatory) fields, it will fail because of the default value which will be None.

We should skip this attribute.

Models using DictNestedField fields are wrongly inited and can't be used for writing:

Example (from XXX):

class Score(pykorm.models.Nested):
    exterior: int = pykorm.fields.DataField('exterior')
    delicious: int = pykorm.fields.DataField('delicious', 10)


@pykorm.k8s_custom_object('pykorm.infomaniak.com', 'v1', 'peaches')
class Peach(pykorm.NamespacedModel):
    variety: str = pykorm.fields.Spec('variety', default='default-variety')
    price: str = pykorm.fields.Spec('price', default=1)
    colours: list = pykorm.fields.Spec('colours', default=[])
    score: Score = pykorm.fields.DictNestedField(Score, path=['spec', 'score'])

Results in the following error when writing to p.score.delicious:

>       p.score.exterior = 42
E       AttributeError: 'NoneType' object has no attribute 'exterior'

CRDs fields can have default values:

@pykorm.k8s_custom_object('pykorm.infomaniak.com', 'v1', 'foo')
class Foo(ScoreMixin, pykorm.NamespacedModel):
    colours: list = pykorm.fields.Spec('colours', [])

However, the default value of the field is not copied on object instantiation. This results in bugs like these:

f1 = Foo(...)
f1.colours = ['foo']

f2 = Foo(...)
print(f2.colours)  # -> ['foo'] . Should return [] !

The source code should be further documented in order to generate an autodoc

We should add a Relationship field whose role is to reference another resource.

However, how to deal with references to other namespaces ? We could support:

• ClusterResource → ClusterResource
• NamespacedResource(ns n) → NamespacedResource(ns n)
• NamespacedResource(ns n) → Cluster resource

We can speedup the filter_by method on Namespaced objects when filtering by a specific namespace instead of crawling all the namespaces.

Pykorm supports CRDs, as well as core objects and apps/v1 objects. But many are missing (e.g. batch/v1 for example). We should add them too