Timeout when importing torch

[sudoprime]

This example app.py always times out when import torch is included, but works without the import. The dockerfile works locally just fine. I have no explanation of the behavior. Whether I pip3 install torch or install from requirements.txt, the result is the same.

app.py

import json
import torch
import sys

def handler(event, context):
    return 'Hello from AWS Lambda using Python' + sys.version + '!'

Dockerfile

FROM public.ecr.aws/lambda/python:3.7

RUN pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu --target "${LAMBDA_TASK_ROOT}"

COPY app.py ${LAMBDA_TASK_ROOT}

CMD ["app.handler"]

[leandrodamascena]

FROM public.ecr.aws/lambda/python:3.7

RUN pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu --target "${LAMBDA_TASK_ROOT}"

COPY app.py ${LAMBDA_TASK_ROOT}

CMD ["app.handler"]

I tried this with Python 3.11 and it seems that this library requires some additional libraries that the base Lambda image does not contain:

{"errorMessage": "libcublas.so.*[0-9] not found in the system path ['/var/task', '/opt/python/lib/python3.11/site-packages', '/opt/python', '/var/lang/lib/python3.11/site-packages', '/var/runtime', '/var/lang/lib/python311.zip', '/var/lang/lib/python3.11', '/var/lang/lib/python3.11/lib-dynload', '/var/lang/lib/python3.11/site-packages', '/opt/python/lib/python3.11/site-packages', '/opt/python']", "errorType": "ValueError", "requestId": "", "stackTrace": [" File \"/var/lang/lib/python3.11/importlib/__init__.py\", line 126, in import_module\n return _bootstrap._gcd_import(name[level:], package, level)\n", " File \"<frozen importlib._bootstrap>\", line 1204, in _gcd_import\n", " File \"<frozen importlib._bootstrap>\", line 1176, in _find_and_load\n", " File \"<frozen importlib._bootstrap>\", line 1147, in _find_and_load_unlocked\n", " File \"<frozen importlib._bootstrap>\", line 690, in _load_unlocked\n", " File \"<frozen importlib._bootstrap_external>\", line 940, in exec_module\n", " File \"<frozen importlib._bootstrap>\", line 241, in _call_with_frames_removed\n", " File \"/var/task/app.py\", line 2, in <module>\n import torch\n", " File \"/var/task/torch/__init__.py\", line 366, in <module>\n _load_global_deps()\n", " File \"/var/task/torch/__init__.py\", line 324, in _load_global_deps\n _preload_cuda_deps(lib_folder, lib_name)\n", " File \"/var/task/torch/__init__.py\", line 284, in _preload_cuda_deps\n raise ValueError(f\"{lib_name} not found in the system path {sys.path}\")\n"]}

The base Lambda image is built on top of a minimal Amazon Linux container image, so I don't know if you will be able to run torch with that. You may have to create your own image to run on Lambda. Check how to do that here.

Also, can you consider upgrading Python to at least 3.9? Python 3.7 is EOL since 2023.

Thanks.