/usr/bin/bash: cannot execute binary file
I'm trying to set up custom template with my own container image. When I try to set up SSH according to docs (https://docs.runpod.io/pods/configuration/use-ssh) I get errors on container start:
2024-02-18T17:40:55.341955276Z /usr/bin/bash: /usr/bin/bash: cannot execute binary file
2024-02-18T17:41:11.745017647Z /usr/bin/bash: /usr/bin/bash: cannot execute binary file
2024-02-18T17:41:28.059617609Z /usr/bin/bash: /usr/bin/bash: cannot execute binary file
2024-02-18T17:41:44.387377159Z /usr/bin/bash: /usr/bin/bash: cannot execute binary file
2024-02-18T17:42:00.764592909Z /usr/bin/bash: /usr/bin/bash: cannot execute binary file
2024-02-18T17:42:17.092566318Z /usr/bin/bash: /usr/bin/bash: cannot execute binary file
Use SSH | RunPod Documentation
The basic terminal SSH access that RunPod exposes is not a full SSH connection and, therefore, does not support commands like SCP. If you want to have full SSH capabilities, then you will need to rent an instance that has public IP support and run a full SSH daemon in your Pod.
13 Replies
What kind of image is this? Looks like it has an issue with bash for some reason.
Here's Dockerfile. It works fine in vast.ai
FROM nvidia/cuda:12.2.2-devel-ubuntu22.04
ENV DEBIAN_FRONTEND=noninteractive \
PYTHONUNBUFFERED=1 \
PIP_NO_CACHE_DIR=on \
SHELL=/bin/bash \
NVIDIA_VISIBLE_DEVICES=all \
NVIDIA_DRIVER_CAPABILITIES=compute,utility \
PYTORCH_CUDA_ALLOC_CONF=backend:cudaMallocAsync
SHELL ["/bin/bash", "-o", "pipefail", "-c"]
# Install dependencies
RUN apt-get update && apt-get install -y python3-pip python3-dev python3-venv git wget vim tmux gettext-base && \
wget https://go.dev/dl/go1.21.6.linux-amd64.tar.gz && \
tar -C /usr/local -xzf go1.21.6.linux-amd64.tar.gz && \
rm -r go1.21.6.linux-amd64.tar.gz /var/lib/apt/lists/*
ENV PATH="${PATH}:/usr/local/go/bin"
# Install hfdownloader
RUN git clone https://github.com/bodaay/HuggingFaceModelDownloader /HuggingFaceModelDownloader && \
cd /HuggingFaceModelDownloader/ && ./BuildLinuxAmd64.sh \
chmod +x /HuggingFaceModelDownloader/output/hfdownloader_linux_amd64_1.2.9 && \
mv /HuggingFaceModelDownloader/output/hfdownloader_linux_amd64_1.2.9 /usr/local/bin/hfdownloader && \
mkdir /models
# Install tabbyAPI
RUN git clone https://github.com/theroyallab/tabbyAPI /app/tabbyAPI && \
rm -r /app/tabbyAPI/models && ln -s /models /app/tabbyAPI/models && \
cd /app/tabbyAPI && python3 -m venv venv \
&& . venv/bin/activate \
&& pip3 install -U -r requirements.txt
WORKDIR /app/tabbyAPI
COPY config_template.yml config_template.yml
COPY run_tabby.sh run_tabby.sh
RUN chmod +x run_tabby.sh
# Expose ports for exllama
EXPOSE 5000
ENTRYPOINT ["/bin/bash"]its bc
ur tabby dockerfile
doesnt have openssh
i recommend
ur FROM should use runpod pytorch template
cause its built with pytorch + cuda + ubuntu too
so ur bash is trying to set up openssh but it doesnt exist
also looks like ur setting up ssh:
can read this
https://discord.com/channels/912829806415085598/1194711850223415348
Solution
@Arahizzz remove ENTRYPOINT ["/bin/bash"] from image and rebuild image
oh i see.. their bash script is apt-get installing openssh 👁️
nope
ENTRYPOINT ["/bin/bash"]
has higher priority than docker start command
Oh wow learnt something new didnt realize the interaction between cmd/run/entrypoint huh; great to know
Chatgpt says maybe can add a —entrypoint flag to run command too, to override the entrypoint on run 👁️ maybe can add that before rebuilding the image too with entrypoint removed ; tho rebuilding shouldnt be bad either cause everything would be cached
nope as it's high level argument
dang
just remove ENTRYPOINT ["/bin/bash"] from docker image and it should work
enjoy photo of it admin destroying docker container
Good night
Yeah I also had to chage my Docker images back from ENTRYPOINT to CMD so that people can override the docker command.
I removed ENTRYPOINT and it worked. Thank you, all.