whats the best patterns for configuring providers for resources which are sometimes not available

[RonnyPfannschmidt]

im having a dishka configuration where im composing a larger number of providers

some of those providers are unable to provide their objects in certain environments
for example the provider for kafka consumers/producers is only able to provide them when either the environment provide by k8s provides them as secrets or a alternative is added to a local configuration

so if a configruation is avaliable (which is known by a resource instead of at provider creation time) im able to create the client, else im not

im currently using something like

# typed in gh form as illustation
from myapp import kafka

from dishka import Provider

from myapp.depenency_inject import Configuration, KafkaConfiguration
type MaybeKafkaConfiguration = KafkaConfigruation | None
type MaybeKafkaProduer = kafka.Producer | None 


class KafkaProduerProvider(Provider):
    @provide
    def kafka_config(self, config: Configuration):
         if configration.kafka: return configration.kafka
         else: return None

   @provide
    def kafka_consumer(config: MaybeKafkaConfiguration):
 		if config is None:
 		   return None
        return kafka.Producer(config)
         
      
    ```

using optional and wiring handling them trough the provider configuration seems like a problematic antipattern

instead i'd like to be able to raise a exception from the first dependency so that dishka is enabled to give a proper explanation on why something is not available
  

[Tishka17]

Sorry, I didn't get the problem. Isn't that what you expect?

class Config:
    pass


class Client:
    def __init__(self, config: Config):
        pass


class MyProvider(Provider):
    @provide(scope=Scope.APP)
    def get_config(self) -> Config:
        raise ValueError("Config is not available")

    client = provide(Client, scope=Scope.REQUEST)


container = make_container(MyProvider())
with container() as c:
    c.get(Client)

Error:

Traceback (most recent call last):
  File "/home/tishka17/src/dishka/tmp/not_available.py", line 23, in <module>
    c.get(Client)
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 152, in get
    return self._get_unlocked(key)
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 167, in _get_unlocked
    return compiled(self._get_unlocked, self._exits, self._cache)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "__dishka_factory_130361435959280", line 3, in get
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 163, in _get_unlocked
    return self.parent_container.get(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 152, in get
    return self._get_unlocked(key)
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 163, in _get_unlocked
    return self.parent_container.get(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 154, in get
    return self._get_unlocked(key)
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 167, in _get_unlocked
    return compiled(self._get_unlocked, self._exits, self._cache)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "__dishka_factory_130361435959168", line 3, in get
  File "/home/tishka17/src/dishka/tmp/not_available.py", line 16, in get_config
    raise ValueError("Config is not available")
ValueError: Config is not available

[Tishka17]

What if we do something like this:

    @provide(scope=Scope.APP)
    def get_config(self) -> Config:
        raise CannotProvideError("No config is available")

...very long trace here...
...
  File "/home/tishka17/src/dishka/src/dishka/container.py", line 167, in _get_unlocked
    return compiled(self._get_unlocked, self._exits, self._cache)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "__dishka_factory_133126958739440", line 3, in get
  File "/home/tishka17/src/dishka/tmp/not_available.py", line 20, in get_config
    raise CannotProvideError("No config is available")
Cannot provide dependency: No config is available
      ~~~ component='', Scope.REQUEST ~~~
   ↓  __main__.Interactor   
   ↓  __main__.Client       
      ~~~ component='', Scope.APP ~~~
   ↓  __main__.Config       MyProvider.get_config
    → ...                   ???

dishka.exceptions.CannotProvideError

[RonnyPfannschmidt]

looks like a good start - i think it would be a good idea to cut the traceback short (the dishka internals and the compiled factories are loads of line noise

[Tishka17]

I do not really want to reraise any error with different stack as I am not sure that it is safe to copy any Exception value. But I can do it for one specific exception

[Tishka17]

Can you please provide more info why it is important for you to see origin of container.get close to the error call here? The provider behavior is not dependent on who requests object

[RonnyPfannschmidt]

for debugging with the traceback, none of the dozens of frames part of the inner working of Container help - the key points one can use for debugging are:

• site of request
• site of providing
• direct chain of dependencies between request and the provider with a exception