Uploading Custom Docker Images on Glows.ai

This tutorial will guide you step by step on how to upload a local Docker image to Glows.ai, covering image preparation, packaging, uploading, and instance creation.

Creating a Docker Image Tar Package

To ensure your image runs properly on Glows.ai, pay attention to the following:

The image ENTRYPOINT must keep running (otherwise the container will exit automatically).
Install and start the SSH service (for convenient instance access).

Setup Instructions

It is recommended to create a separate .sh script as the ENTRYPOINT, for example run_glowsai.sh, containing startup commands for the container (SSH service or custom services). Example:

#!/bin/bash
# Start SSH service
echo "Starting SSH service..."
/usr/sbin/sshd -E /root/.ssh/sshd_running.log

# Custom Command
echo "hello glowsai" > /hello.txt

# Keep the container running
echo "Container running..."
tail -f /dev/null

Also, in your Dockerfile, include SSH installation and configuration, and set the ENTRYPOINT:

# Install and configure SSH server
RUN apt-get update && apt-get install -y openssh-server && \
    mkdir /var/run/sshd /root/.ssh && sed -ri 's/^#?PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config && \
    sed -ri 's/UsePAM yes/#UsePAM yes/g' /etc/ssh/sshd_config && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

# Copy and set permissions for the start script
COPY run_glowsai.sh /usr/bin/
RUN chmod +x /usr/bin/run_glowsai.sh
EXPOSE 22
# Set entrypoint
ENTRYPOINT ["/usr/bin/run_glowsai.sh"]

Practical Steps

Before uploading your local image, ensure SSH is installed and running inside the image and that the ENTRYPOINT keeps running. Otherwise, follow these steps.

set run_glowsai.sh

Create a directory new_images and inside it, create run_glowsai.sh with the following content:

#!/bin/bash
# Start SSH service
echo "Starting SSH service..."
/usr/sbin/sshd -E /root/.ssh/sshd_running.log

# Custom Command
echo "hello glowsai" > /hello.txt

# Keep the container running
echo "Container running..."
tail -f /dev/null

Note: The # Custom Command section is where you can add any commands to run on container startup. Do not modify other parts to avoid startup failure.

Create a Dockerfile with the following content:

# Set the base image
FROM oldwatch/tmp-images:huggingface-text-embeddings-inference_1_8

# Install and configure SSH server
RUN apt-get update && apt-get install -y openssh-server && \
    mkdir /var/run/sshd /root/.ssh && sed -ri 's/^#?PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config && \
    sed -ri 's/UsePAM yes/#UsePAM yes/g' /etc/ssh/sshd_config && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

# Copy and set permissions for the start script
COPY run_glowsai.sh /usr/bin/
RUN chmod +x /usr/bin/run_glowsai.sh
EXPOSE 22
# Set entrypoint
ENTRYPOINT ["/usr/bin/run_glowsai.sh"]

Build the new image in the directory:

docker build -t new_images:huggingface-text-embeddings-inference_1_8 .

Build successfully

After a successful build, use docker save to package the image as a tar file:

docker save -o <output_file_path> <image_name_or_id>

Pack tar

Uploading the Docker Image Tar Package

Open Glows.ai’s Images interface and click Upload Image.
Enter an image description and click Select Image to upload the tar package.

Choose image

During upload, progress will be displayed. Once completed, the image can be used in Glows.ai.

Upload successfully

Creating an Instance

When creating an instance, select My Images. You will see the uploaded image. Choose it, and if a Datadrive Mount is needed, complete the instance creation process.

Create instance

After the instance starts, manually set the ports you want to expose, e.g., SSH on port 22:
- Click New Port Forwarding
- Set Service Port to 22
- Do not check HTTPS
- Click Create to finish port forwarding

Set port