Worker PID is not available yet

[viratbiz]

🐛 Describe the bug

I am using python version 3.9. There is error coming that "Worker pid is not available yet." This is due to the model_service_worker.py in ts. The error is name: "Integer" is not defined. I have also attached the image for reference.

Error logs

2023-05-29T13:36:37,550 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - Backend worker process died.
2023-05-29T13:36:37,551 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - Traceback (most recent call last):
2023-05-29T13:36:37,551 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - File "/Users/virat.s/miniconda3/envs/python_39/lib/python3.9/site-packages/ts/model_service_worker.py", line 250, in
2023-05-29T13:36:37,551 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - worker = TorchModelServiceWorker(
2023-05-29T13:36:37,551 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - File "/Users/virat.s/miniconda3/envs/python_39/lib/python3.9/site-packages/ts/model_service_worker.py", line 69, in init
2023-05-29T13:36:37,551 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - self.port = Integer.pa + LOCAL_RANK
2023-05-29T13:36:37,551 [INFO ] W-9001-midjourney_1.0-stdout MODEL_LOG - NameError: name 'Integer' is not defined
2023-05-29T13:36:37,554 [INFO ] W-9005-midjourney_1.0-stdout MODEL_LOG - Backend worker process died.

Installation instructions

pip install torch-model-archiver torchserve

Model Packaing

Standard sytable_diffusion_handler.py

config.properties

No response

Versions

Python version 3.9

Repro instructions

torch-model-archiver --model-name midjourney --version 1.0 --handler midjourney_handler.py --extra-files model.zip -r requirements.txt

requirements.txt:
transformers
diffusers==0.16.1
numpy

torchserve --start --ts-config config.properties

Possible Solution

No response

[msaroufim]

Seems related to #2349

[lxning]

@viratbiz could you provide your config.properties?

BTW, the handler implementation has to avoid multiple process unzip the same zip file concurrently if you pass zip file (ie." --extra-files model.zip ). In fact, you don't need pass zip file. you can just pass "--extra-file model_dir --archive-format tgz", then archiver will generate tgz file for you.

[viratbiz]

Using the default config properties given here: https://github.com/pytorch/serve/blob/master/examples/diffusers/config.properties

[lxning]

@viratbiz according to the above error log, tcp is used. However, the config.properties uses the socket. there is a log starting with "Worker cmdline:" in logs/ts_log.log. could you provide it so that i can know which protocol is used? Thanks,