ImportError: cannot import name 'UJSONResponse'

[creatorrr]

Version

cli version: 0.35.0

Description

Using starlette.responses.Response throws error on local debug run

Configuration

relevant cortex.yaml:

- name: chitchat
  kind: RealtimeAPI
  handler:
    type: python
    path: chitchat_service/__init__.py
    models:
      path: s3://cortex-examples/sklearn/mpg-estimator/linreg/

Stack traces

(error output from cortex logs <api name>)

Traceback (most recent call last):
  File "/src/cortex/serve/start/server.py", line 46, in <module>
    main()
  File "/src/cortex/serve/start/server.py", line 40, in main
    log_config=log_config,
  File "/opt/conda/envs/env/lib/python3.6/site-packages/uvicorn/main.py", line 362, in run
    server.run()
  File "/opt/conda/envs/env/lib/python3.6/site-packages/uvicorn/main.py", line 390, in run
    loop.run_until_complete(self.serve(sockets=sockets))
  File "uvloop/loop.pyx", line 1456, in uvloop.loop.Loop.run_until_complete
  File "/opt/conda/envs/env/lib/python3.6/site-packages/uvicorn/main.py", line 397, in serve
    config.load()
  File "/opt/conda/envs/env/lib/python3.6/site-packages/uvicorn/config.py", line 278, in load
    self.loaded_app = import_from_string(self.app)
  File "/opt/conda/envs/env/lib/python3.6/site-packages/uvicorn/importer.py", line 23, in import_from_string
    raise exc from None
  File "/opt/conda/envs/env/lib/python3.6/site-packages/uvicorn/importer.py", line 20, in import_from_string
    module = importlib.import_module(module_str)
  File "/opt/conda/envs/env/lib/python3.6/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 994, in _gcd_import
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load
  File "<frozen importlib._bootstrap>", line 955, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 665, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 678, in exec_module
  File "/opt/conda/envs/env/lib/python3.6/site-packages/cortex_internal/serve/wsgi.py", line 15, in <module>
    from cortex_internal.serve.serve import start
  File "/opt/conda/envs/env/lib/python3.6/site-packages/cortex_internal/serve/serve.py", line 30, in <module>
    from fastapi import FastAPI
  File "/opt/conda/envs/env/lib/python3.6/site-packages/fastapi/__init__.py", line 23, in <module>
    from .responses import Response
  File "/opt/conda/envs/env/lib/python3.6/site-packages/fastapi/responses.py", line 10, in <module>
    from starlette.responses import UJSONResponse  # noqa
ImportError: cannot import name 'UJSONResponse'

Suggested solution

This bug is due to an older version of starlette. Was fixed in tiangolo/fastapi#2342

[deliahu]

@creatorrr thanks for opening this issue.

I was unable to recreate the error. I tried a simple handler like this:

from starlette.responses import Response

class Handler:
    def __init__(self, config):
        pass

    def handle_post(self, payload, query_params):
        return Response("ok")

Do you mind creating a minimal example which reproduces the error? Perhaps it has to do with the contents of your requirements.txt?

[creatorrr]

It’s happening with starlette==0.14.2 but goes away with starlette==0.14.0

[deliahu]

Yes, and it also seems to work by default if starlette is not specified in requirements.txt (the pre-installed version is 0.13.6). It is generally not recommended to change the version of cortex's internal dependencies (such as starlette); is there a feature that you need access to in starlette 0.14?

[creatorrr]

Using a library that depends on ^0.14.2, I was able to pin to a lower version but it'd be great if it was possible to update cortex deps.

[deliahu]

Yes, that makes sense. Have you attempted pinning startlette and fastapi to the appropriate versions? E.g.

# requirements.txt

fastap==0.65.1
starlette==0.14.2

[creatorrr]

Yes that works (pinning starlette==0.14.0). It’d be great to have it documented somewhere if possible.

[deliahu]

Awesome! Yes, we will look into documenting this.