Only use config server if GDA didn't supply params

[olliesilvester]

Fixes #494

Setting the parameters to None by default and then using an 'after' validator to say 'if None then use config server' wouldn't work due to typing complaints about setting a boolean to None.

Instead, I am using a 'before' validator to check if the parameters were supplied by GDA. I have to instantiate the Feature class as this hasn't been done at the time the 'pre' validator runs

Acceptance criteria

If GDA parameters are supplied for gpu, panda and stub_offsets, these are used. If the parameters are missing, the config server is used instead

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 78.21%. Comparing base (b10420c) to head (42d777a).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #496      +/-   ##
==========================================
+ Coverage   78.15%   78.21%   +0.05%     
==========================================
  Files          92       92              
  Lines        6785     6797      +12     
==========================================
+ Hits         5303     5316      +13     
+ Misses       1482     1481       -1     

Components	Coverage Δ
i24 SSX	57.12% <ø> (ø)
hyperion	96.27% <100.00%> (+0.04%)	⬆️
other	100.00% <ø> (ø)

[dperl-dls]

Should: Only fetch the flags from the server once and only define them in one place.

With something like:

Suggested change

	def set_default_feature_flags(cls, values) -> Any:
	cls.features = FeatureFlags()
	if "use_panda" not in values:
	values["use_panda"] = cls.features.best_effort().use_panda_for_gridscan
	if "use_gpu" not in values:
	values["use_gpu"] = cls.features.best_effort().use_gpu_for_gridscan
	def set_default_feature_flags(cls, values) -> Any:
	cls.features = FeatureFlags.best_effort()
	if "use_panda" in values:
	cls.features.use_panda_for_gridscan = values["use_panda"]
	...

then the new definitions above could be removed?

[olliesilvester]

I agree to only get the flags from the server once, but I think the definitions for, eg use_panda still need to be in GridCommon, just having them in the validator isn't enough for pydantic to know that GridCommon has these attributes

[dperl-dls]

oh, of course, sorry. but use_panda still shouldn't be in GridCommon, the flags should be passed in in features, which is a BaseModel as well already.

in the validator for features you can then start with FeatureFlags.best_effort() and update anything which is passed in

[olliesilvester]

Okay, that works, does that mean we should change the logic of best_effort? Right now it checks the URL and uses Hyperion constants for things it can't find. This PR could change it to:

1. Check if GDA supplied values for each parameter, if yes then use that
2. Check if parameter exists in URL, if yes then use that
3. Use Hyperion constants as a last resort

Is that reasonable?

[dperl-dls]

I don't think you need to change the logic in best_effort to achieve that priority order, just set the defaults in teh FeatureFlags class to the constants. maybe add a new layer to I03Constants to hold defaults for feature flags.

[dperl-dls]

but if you want to add the ability to override stuff in best_effort with kwargs that sounds fine to me

[olliesilvester]

I think we still want GDA parameters to have priority. We use best_effort in our plans so these should also give us the GDA parameters when they're supplied