techainer / mlchain-python Goto Github PK

Hi there

I found a bug when running with mlchain run on my MBP (OSX 10.15.6 but I suspect this problem can be reproduced on the older versions)

• When configure wrapper: None in mlconfig => Run mlchain run => There is no bug when calling API

• When configure wrapper: gunicorn in mlconfig => Run mlchain run => There is 1 bug that prohibits function called:
  objc[28043]: +[__NSCFConstantString initialize] may have been in progress in another thread when fork() was called. objc[28043]: +[__NSCFConstantString initialize] may have been in progress in another thread when fork() was called. We cannot safely call it or ignore it in the fork() child process. Crashing instead. Set a breakpoint on objc_initializeAfterForkError to debug.

The problem disappears when I run this command before mlchain run
export OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES

As I tested:
On Windows: There is no gunicorn installed. Should find an alternative
On Linux: Everything works fine

As I know, mlchain inherently supports serving with nginx with 8080 port through mlchain serve ...
However, when inspecting code, there is no exposed sock for nginx to use

Please add support for this. Thank you

There are unnecessary + uncompressed Javascript files:

docs/js/chat.js
docs/js/custom.js
docs/js/termynal.js
mlchain/server/templates/swaggerui/swagger-ui-standalone-preset.js

Sometime there are missing Exception message from server call. Will need to investigate this issue further

The css files should be compressed.

docs/css/custom.css
docs/css/style.css
docs/css/termynal.css
mlchain/server/static/Source-Sans-Pro.css
mlchain/server/templates/swaggerui/swagger-ui.css

Hi there, we are experimenting this lib on ARM based hardware (E.g NVIDIA Jetson).
You should cover this with CI CD and future plan

Fix Travis CI build error and add auto deployment to PyPi when a released new version

Currently docs are being build and update manually. We need CI/CD for documentations on release tag

Hello there, as far as I know, mlchain supports many features. However, when using mlchain init I think it should be better.
E.g:

• Mlchain supports mode for exported environment variable
• Note for production and dev environments: wrapper, host (localhost -> 0.0.0.0)
• etc

Some may encounter this at mlchain run

Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/gunicorn/arbiter.py", line 583, in spawn_worker
    worker.init_process()
  File "/usr/local/lib/python3.7/dist-packages/gunicorn/workers/gthread.py", line 92, in init_process
    super().init_process()
  File "/usr/local/lib/python3.7/dist-packages/gunicorn/workers/base.py", line 119, in init_process
    self.load_wsgi()
  File "/usr/local/lib/python3.7/dist-packages/gunicorn/workers/base.py", line 144, in load_wsgi
    self.wsgi = self.app.wsgi()
  File "/usr/local/lib/python3.7/dist-packages/gunicorn/app/base.py", line 67, in wsgi
    self.callable = self.load()
  File "/usr/local/lib/python3.7/dist-packages/mlchain/cli/run.py", line 205, in load
    serve_model = get_model(entry_file, serve_model=True)
  File "/usr/local/lib/python3.7/dist-packages/mlchain/cli/run.py", line 310, in get_model
    module = importlib.import_module(import_name)
  File "/usr/lib/python3.7/importlib/__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1006, in _gcd_import
  File "<frozen importlib._bootstrap>", line 983, in _find_and_load
  File "<frozen importlib._bootstrap>", line 962, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'code.server'; 'code' is not a package

Hello there,

As far as I know, the exported environment variables are considered string in Python
Mlchain has already supported type indicator in function to convert from image binary to numpy array. However, when exporting int8, I get a string in my code

Please update, it should be a nice feature.

Switch CI to Github Action with full x86 matrix build. Including:

• 3 OS: Windows, Mac, Linux
• 4 Python version: 3.6, 3.7, 3.8, 3.9

Priority order:

• ENV Var
• mlchain run arg
• mlconfig.yaml value

Note: default group in mlconfig should be effect by this as well

As reported in: Techainer/mnist-mlchain-examples#1

When using mlchain on Windows with gunicorn warper will encounter this problem:

ModuleNotFoundError: No module named 'fcntl

Due to this line:

Environment variable CUDA_VISIBLE_DEVICES don't have effect when running with MLChain. Will fix this soon cc @vuonghoainam

@vuonghoainam reported a bug when install mlchain 0.1.7 and run with mlchain run

    from .h2 import H2Protocol
  File "/usr/local/lib/python3.7/site-packages/hypercorn/protocol/h2.py", line 4, in <module>
    import h2.connection
  File "/usr/local/lib/python3.7/site-packages/h2/connection.py", line 33, in <module>
    from .frame_buffer import FrameBuffer
  File "/usr/local/lib/python3.7/site-packages/h2/frame_buffer.py", line 9, in <module>
    from hyperframe.exceptions import InvalidFrameError, InvalidDataError
ImportError: cannot import name 'InvalidDataError' from 'hyperframe.exceptions' (/usr/local/lib/python3.7/site-packages/hyperframe/exceptions.py)

Mlchain installation problems with virtual environment (anaconda)

ERROR: Failed building wheel for bottleneck
Failed to build bottleneck
ERROR: Could not build wheels for bottleneck which use PEP 517 and cannot be installed directly

Ubuntu: 18.04.5 LTS
python : 3.6
conda : 4.8.3

One internal project of Techainer discover a bug that when using python3 server.py with a content like this:

from mlchain.base import ServeModel
from model import Model
from mlchain import mlconfig

# mlconfig.load_config('mlconfig.yaml')
model = Model(weight_path=mlconfig.weight,
                     debug=mlconfig.debug)
model = ServeModel(model)

if __name__ == "__main__":
    from mlchain.rpc.server.flask_server import FlaskServer
    FlaskServer(model).run(bind=['127.0.0.1:8004'], gunicorn=True)

Have cause the self.sess.run inside the model class to hang forever. While using mlchain run CLI doesn't.

Noted that this model use Tensorflow 1.14, 1.15 suffer the same problem

This DOES NOT effect production usage since we only use mlchain run but this bug is worth more examination.

When I ran QuartServer. I got an exception.

Traceback (most recent call last):                                                                                                      
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/uvicorn/protocols/http/httptools_impl.py", line 396, in run_asgi       
    result = await app(self.scope, self.receive, self.send)                                                                             
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/uvicorn/middleware/proxy_headers.py", line 45, in __call__             
    return await self.app(scope, receive, send)                                                                                         
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/app.py", line 2117, in __call__                                  
    await self.asgi_app(scope, receive, send)                                                                                           
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/app.py", line 2140, in asgi_app                                  
    await asgi_handler(receive, send)                                                                                                   
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/asgi.py", line 33, in __call__                                   
    _raise_exceptions(done)                                                                                                             
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/asgi.py", line 256, in _raise_exceptions                         
    raise task.exception()                                                                                                              
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/asgi.py", line 84, in handle_request                             
    await asyncio.wait_for(self._send_response(send, response), timeout=timeout)                                                        
  File "/usr/lib/python3.7/asyncio/tasks.py", line 442, in wait_for                                                                     
    return fut.result()                                                                                                                 
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/asgi.py", line 98, in _send_response                             
    async for data in body:                                                                                                             
  File "/mnt/hdd/spaces/miles/.local/lib/python3.7/site-packages/quart/wrappers/response.py", line 129, in _aiter                       
    for data in iterable:  # type: ignore                                                                                               
TypeError: 'coroutine' object is not iterable

My environment:

MlChain 0.1.9
Flask 1.1.2
Quart 0.14.1

I've recently picked up on using this framework. I can't find instruction on how to define the specific URL for the API, other than IP:port. What I want to do is deploying an API to an URL like IP:port/route/to/api. How do I achieve that? Thank you.

Starlette is more stable than Quart, so we are considering remove Quart and integrate Starlette as default server (Now is Flask)

How to scale up system to hundreds of node server :)

• Sentry integration
• Integrate Sentry Performance for Mlchain Client and Task
• Add GPU info when error with Sentry
• Ignore Python Packages when logging Sentry
• Fixed Sentry failed with Gunicorn started

At mlchain 0.1.8rc1, I have encountered the following bug during the initialization phase with mlchain. Please help resolve this. Thank you
This link should be helpful I guess: Stackoverflow
face_detection_1 | Traceback (most recent call last): face_detection_1 | File "/usr/local/bin/mlchain", line 11, in <module> face_detection_1 | sys.exit(main()) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/mlchain/cli/main.py", line 50, in main face_detection_1 | cli.main(args=sys.argv[1:], prog_name="python -m mlchain" if as_module else None) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 782, in main face_detection_1 | rv = self.invoke(ctx) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 1259, in invoke face_detection_1 | return _process_result(sub_ctx.command.invoke(sub_ctx)) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 1066, in invoke face_detection_1 | return ctx.invoke(self.callback, **ctx.params) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/click/core.py", line 610, in invoke face_detection_1 | return callback(*args, **kwargs) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/mlchain/cli/run.py", line 104, in run_command face_detection_1 | config = mlconfig.load_file(config) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/mlchain/config.py", line 152, in load_file face_detection_1 | return load_yaml(path) face_detection_1 | File "/usr/local/lib/python3.7/dist-packages/mlchain/config.py", line 144, in load_yaml face_detection_1 | return yaml.load(f, Loader=yaml.FullLoader) face_detection_1 | AttributeError: module 'yaml' has no attribute 'FullLoader'