About opening the kube_config.new_client_from_config parameter in the Session initialization function #37
Description
Is your feature request related to a problem? Please describe.
When I needed to connect to a remote k8s cluster, I found that GS had no configurable parameters to do so
In the script session.py, at line 572:
api_client = kube_config.new_client_from_config()Let's look at thenew_client_from_config function:
def new_client_from_config(
config_file=None,
context=None,
persist_config=True):
"""
Loads configuration the same as load_kube_config but returns an ApiClient
to be used with any API object. This will allow the caller to concurrently
talk with multiple clusters.
"""
client_config = type.__call__(Configuration)
load_kube_config(config_file=config_file, context=context,
client_configuration=client_config,
persist_config=persist_config)
client_config.verify_ssl = False
return ApiClient(configuration=client_config)k8s allows developers to pass in the config_file parameter, but GS hides this parameter.
Describe the solution you'd like
I think the parameters of the functionnew_client_from_config can be configured by the developer in the Session class, just like:
api_client = kube_config.new_client_from_config(config_file=None, context=None, persist_config=persist_config)
or
api_client = kube_config.new_client_from_config(kw.pop('k8s_client_config'))This allows us to easily load the k8s configuration:
sess = graphscope.session(kw = {config_file=None, context=None, persist_config=True})
or
sess = graphscope.session(kw ={"k8s_client_config" : {config_file=None, context=None, persist_config=True}})Describe alternatives you've considered
Additional context
My current solution is to configure the environment variables as follows:
os.environ.setdefault(
"KUBECONFIG", os.path.join(Config.BASE_DIR, "config/kube_inner_config")
)
import graphscopeThis way of writing causes that environment variables must be declared before the graphscope is imported, which is very inelegant.