document RASA_DUCKLING_HTTP_URL env variable usage

[N-Olbert]

Rasa version: 1.10.7

Python version: 3.7

Operating system (windows, osx, ...): Windows 10 1909

Issue: It seems like the Duckling-URL becomes a hard-coded part of the NLU-Model. Changing the Duckling-URL after training seems to have no effect.

Repro:

1. Train a model using f. e. http://localhost:8000 as Duckling-URL
2. Change the URL to f. e. http://localhost:55555 (but DONT retrain the model)
3. Start the Duckling-server on http://localhost:55555
4. Run rasa run
5. See failure after parsing any message

See also: https://forum.rasa.com/t/does-the-duckling-url-become-a-hard-coded-part-of-the-trained-model/31001

Error (including full traceback):

Failed to connect to duckling http server. Make sure the duckling server is running/healthy/not stale and the proper host and port are set in the configuration. More information on how to run the server can be found on github: https://github.com/facebook/duckling#quickstart Error: HTTPConnectionPool(host='localhost', port=8000): Max retries exceeded with url: /parse (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x000002BE546C5D48>: Failed to establish a new connection: [WinError 10061] Es konnte keine Verbindung hergestellt werden, da der Zielcomputer die Verbindung verweigerte'))

Note: Port 8000 is still used.

Content of configuration file (config.yml) (if relevant): Before

language: de
 pipeline:

*   name: WhitespaceTokenizer
*   name: RegexFeaturizer
*   name: LexicalSyntacticFeaturizer
*   name: CountVectorsFeaturizer
*   name: CountVectorsFeaturizer
     analyzer: "char_wb"
     min_ngram: 1
     max_ngram: 4
*   name: DIETClassifier
     epochs: 100
*   name: EntitySynonymMapper
*   name: ResponseSelector
     epochs: 100
*   name: "DucklingHTTPExtractor"
     url: "http://localhost:8000"
     dimensions: ["time"]
     locale: "de_DE"
     timezone: "UTC"
     timeout : 3

After

language: de
 pipeline:

*   name: WhitespaceTokenizer
*   name: RegexFeaturizer
*   name: LexicalSyntacticFeaturizer
*   name: CountVectorsFeaturizer
*   name: CountVectorsFeaturizer
     analyzer: "char_wb"
     min_ngram: 1
     max_ngram: 4
*   name: DIETClassifier
     epochs: 100
*   name: EntitySynonymMapper
*   name: ResponseSelector
     epochs: 100
*   name: "DucklingHTTPExtractor"
     url: "http://localhost:55555" #This is the only change
     dimensions: ["time"]
     locale: "de_DE"
     timezone: "UTC"
     timeout : 3

[N-Olbert]

Exalate commented:

N-Olbert commented:

According to the forum, this is defined functionality (see https://forum.rasa.com/t/does-the-duckling-url-become-a-hard-coded-part-of-the-trained-model/31001), so please consider this as an enhancement request rather than a bug.

Not being able to change the port/uri after training is bad because if a customer has already used the Duckling-port which was provided during training there is no way to change this port except to train a new model (which takes some time).

[sara-tagger]

Exalate commented:

sara-tagger commented:

Thanks for raising this issue, @Ghostvv will get back to you about it soon

✨

Please also check out the docs and the forum in case your issue was raised there too

🤗

[wochinge]

Exalate commented:

wochinge commented:

You could use an environment variable instead of a hardcoded URL in your configuration, e.g.

language: de
 pipeline:

*   name: WhitespaceTokenizer
*   name: RegexFeaturizer
*   name: LexicalSyntacticFeaturizer
*   name: CountVectorsFeaturizer
*   name: CountVectorsFeaturizer
     analyzer: "char_wb"
     min_ngram: 1
     max_ngram: 4
*   name: DIETClassifier
     epochs: 100
*   name: EntitySynonymMapper
*   name: ResponseSelector
     epochs: 100
*   name: "DucklingHTTPExtractor"
     url: $ {DUCKLING_URL} 

     #This is the only change
     dimensions: ["time"]
     locale: "de_DE"
     timezone: "UTC"
     timeout : 3

I see that this isn't ideal in terms of usability, but otherwise we'd need some way to pass in the duckling url during loading the model. I think using an env variable still seems the best way to do this. Wdyt?

[N-Olbert]

Exalate commented:

N-Olbert commented:

@wochinge Thanks for that suggestion. This is a clever trick, but doesn't seem to work: The metadata.json within the nlu -folder of the resulting model contains already the resolved value of the environment variable.

Anyways I found a solution: There is a hard coded env var within the DucklingHTTPExtractor which always wins:

rasa/rasa/nlu/extractors/duckling_http_extractor.py

Lines 99 to 104 in b70e9f4

	def _url(self) -> Optional[Text]:
	"""Return url of the duckling service. Environment var will override."""
	if os.environ.get("RASA_DUCKLING_HTTP_URL"):
	return os.environ["RASA_DUCKLING_HTTP_URL"]
	return self.component_config.get("url")

(This issue can be closed form my side)

[wochinge]

Exalate commented:

wochinge commented:

Ah, good catch. Completely forgot about RASA_DUCKLING_HTTP_URL

🙈

(This issue can be closed form my side)

I'll convert this to a documentation issue as RASA_DUCKLING_HTTP_URL should be documented in https://rasa.com/docs/rasa/nlu/components/#ducklinghttpextractor

[sync-by-unito]

➤ Maxime Verger commented:

💡 Heads up! We're moving issues to Jira: https://rasa-open-source.atlassian.net/browse/OSS.

From now on, this Jira board is the place where you can browse (without an account) and create issues (you'll need a free Jira account for that). This GitHub issue has already been migrated to Jira and will be closed on January 9th, 2023. Do not forget to subscribe to the corresponding Jira issue!

➡️ More information in the forum: https://forum.rasa.com/t/migration-of-rasa-oss-issues-to-jira/56569.

[RheagalFire]

This still hasn't been documented.