Rasa NLU Server: "error": "Unable to initialize persistor"

Rasa Open Source
1

Hi all

I would like to have 1 instance of Rasa NLU server (through Docker) to handle multiple projects.

Here is my dir:

.
├── components
│   ├── __init__.py
│   ├── __pycache__
│   │   └── __init__.cpython-36.pyc
│   └── stanford
│       ├── com_stanford_nlp.py
│       ├── __init__.py
│       └── __pycache__
│           ├── com_stanford_nlp.cpython-36.pyc
│           └── __init__.cpython-36.pyc
├── log
│   ├── rasa_nlu_log-20181113-100055-1.log
│   └── stanford_nlp.log
├── models
│   ├── company_a
│   │   └── en
│   │       ├── metadata.json
│   │       └── training_data.json
│   └── model_config_for_server.yml
└── projects
    └── company_a
        ├── config
        │   └── config.yml
        └── data
            └── data.json

The command to run Rasa NLU server using Docker:

docker \
    run \
        --name rasa_nlu_server \
        -p 5001:5001 \
        -v $(pwd)/components/:/app/components \
        -v $(pwd)/models/:/app/models \
        -v $(pwd)/log/:/app/log \
        rasa/rasa_nlu:latest-tensorflow \
    run \
        python -m rasa_nlu.server \
            --port 5001 \
            --token 12345 \
            --write log/server.log \
            --response_log log \
            --path models \
            --pre_load all \
            --config models/model_config_for_server.yml

By the way, the content of model_config_for_server.yml is:

language: "en"
pipeline:
  - name: "components.stanford.com_stanford_nlp.Stanford_NLP"

When I run:

curl 'http://my_rasa_nlu_server:5001/parse?token=12345&q=hi&project=company_a'

Here is the error I got:

WARNING  rasa_nlu.project  - Using default interpreter, couldn't fetch model: Unable to initialize persistor
ERROR    __main__  - Unable to initialize persistor
Traceback (most recent call last):
  File "/app/rasa_nlu/server.py", line 245, in parse
    self.data_router.parse, data))
  File "/usr/local/lib/python3.6/site-packages/twisted/python/threadpool.py", line 250, in inContext
    result = inContext.theWork()
  File "/usr/local/lib/python3.6/site-packages/twisted/python/threadpool.py", line 266, in <lambda>
    inContext.theWork = lambda: context.call(ctx, func, *args, **kw)
  File "/usr/local/lib/python3.6/site-packages/twisted/python/context.py", line 122, in callWithContext
    return self.currentContext().callWithContext(ctx, func, *args, **kw)
  File "/usr/local/lib/python3.6/site-packages/twisted/python/context.py", line 85, in callWithContext
    return func(*args,**kw)
  File "/app/rasa_nlu/data_router.py", line 271, in parse
    model)
  File "/app/rasa_nlu/project.py", line 261, in parse
    interpreter = self._interpreter_for_model(model_name)
  File "/app/rasa_nlu/project.py", line 366, in _interpreter_for_model
    metadata = self._read_model_metadata(model_name, model_dir)
  File "/app/rasa_nlu/project.py", line 383, in _read_model_metadata
    self._load_model_from_cloud(model_name, path)
  File "/app/rasa_nlu/project.py", line 422, in _load_model_from_cloud
    raise RuntimeError("Unable to initialize persistor")
RuntimeError: Unable to initialize persistor

Can anyone please tell me how to fix it?

Thank you very much in advance for all your help.

2

Hello, you need to specify in the request the model you want to use. For example:

curl -XPOST http://my_rasa_nlu_server:5001/parse -d '{"q":"hello there", "model": "current"}'

Hope it helps.

1 Like
3

Thanks @jeanmetz, it works!

To people who has this error, my full request is this:

curl -X POST http://my_rasa_nlu_server:5001/parse?token=12345 -d '{"q":"hi", "project":"company_a", "model":"en"}'

Here is the result I got:

{
  "intent": {
    "name": null,
    "confidence": 0.0
  },
  "entities": [],
  "text": "hi",
  "project": "company_a",
  "model": "en"
}

The intent null is expected as I am just testing the server. :smiley:

1 Like
4

Glad to help!

5

Hi, I’m in a similar situation but I can’t manage to load the models (I have my own NLU models trained with keras. I have no problems to load the models with Agent.load, but now I can’t pass to the server). My dir is (i was based in an example with this architecture):

|--nlu_utils
|   -glove_utils.py
|   -intent_classifier.py
|   -entity_extractor.py
|--nlu_models
|   -intent_model.h5
|   -entity_model.h5
|   -glove_model.h5
|-nlu_config.yml

where my nlu_config is:

language: "en_core_web_md"

pipeline:
- name: "nlu_utils.glove_utis.GloveNLP"
- name: "nlu_utils.int_proba.IntentProba"
- name: "nlu_utils.ner_dl.DLEntityExtractor"

and if I run python -m rasa_nlu.server --config ./nlu_config.yml --path nlu_models/ --pre_load all it starts the server but not loads any model.

If I do curl localhost:5000/status the answer is:

{
  "max_training_processes": 1,
  "current_training_processes": 0,
  "available_projects": {
    "entity_model.h5": {
      "status": "ready",
      "current_training_processes": 0,
      "available_models": [
        "fallback"
      ],
      "loaded_models": [
        "fallback"
      ]
    },
    "intent_model.h5": {
      "status": "ready",
      "current_training_processes": 0,
      "available_models": [
        "fallback"
      ],
      "loaded_models": [
        "fallback"
      ]
    },
    "glove_model.h5": {
      "status": "ready",
      "current_training_processes": 0,
      "available_models": [
        "fallback"
      ],
      "loaded_models": [
        "fallback"
      ]
    },
   
    }

It’s not looking where the models are, could anyone help me please? I’m lost

6

I’ve checked this behaviour and actually could reproduce the same problem. So, indeed --pre_load all is not working as expected. I’ll investigate further for a solution. If anyone out there knows the solution, please help us :slight_smile:

7

@jeanmetz you may try --pre_load [all], I remembered that somewhere in your doc it use [all] instead of all (without square-bracket)

8

Hi @alucard001, --pre_load [all] isn’t accepted as argument by the argument parser. On the other hand, if the argument is given as --pre_load '[all]' the parser will accept as a valid value, but the loader will not find any project with the name [all] and will then not load the model available.

The response to GET on /status resource is:

{
  "max_training_processes": 1,
  "current_training_processes": 0,
  "available_projects": {
    "current": {
      "status": "ready",
      "current_training_processes": 0,
      "available_models": [
        "nlu-a",
        "nlu-b"
      ],
      "loaded_models": []
    }
  }
}

Conclusion: the problem still occurs.

9

I did some digging about the --pre_load option for rasa_nlu.server. The functionality we want is not supported yet. The issue is actually ambiguity in the rasa nlu documentation, which stats the following:

--pre_load PRE_LOAD [PRE_LOAD ...]
                        Preload models into memory before starting the server.
                        If given `all` as input all the models will be loaded.
                        Else you can specify a list of specific project names.
                        Eg: python -m rasa_nlu.server --pre_load project1
                        --path projects -c config.yaml

It actually refers to projects and not models.

There is an open pull request to solve this issue: https://github.com/RasaHQ/rasa_nlu/pull/1410

hopefully it gets approved soon.

10

I face with this problem too, and I solved with change models name folder. Suppose we have two bots in path projects with this folder:

base_folder   
└───projects
│   └───bot1
│      └───model_20190303-234432
│   └───bot2
│      └───model_20190303-224432

So we can run server using python -m rasa_nlu.server --path projects.

Note : folder inside model folder must follow format model_YYYYMMDD-hhmmss.

Hope it help!

11

Hi

I have tried out in similar fashion but not able to run multiple bots I am getting error as “error”: “No project found with name ‘default’.”

12

When you inference it, model which you load is bot1, so it will be load models on that folder