VM runtime: How to convert a Docker image into a Golem image?
Since you're interested in running your own dockerized apps on Golem, we're going to assume you're more or less familiar with Docker.
Obviously, you'll need Docker itself installed on your machine. If you don't have docker, visit: https://www.docker.com/products/docker-desktop and download+install it.
Our gvmkit-build tool requires a decently new Python version to run so please ensure you're running at least 3.6.1.
This is our custom tool that prepares GVM images based on docker images.
To install it:
pip install gvmkit-build
This will install gvmkit-build itself and all its dependencies.
There's just one command that suffices to build your Golem image. We're use our own example here but you can use any Docker image here:
gvmkit-build golemfactory/blender:demo
This will pull the image if needed and then proceed with re-packing it into Golem's custom image format.
If this command fails for you, please ensure that your Docker installation is functioning correctly and that your user has permissions to run Docker by e.g. running:
docker run hello-world
Once the image is built, it can be placed in our the repository.
Currently, for the sake of simplicity, we're using a freely-accessible repository that everybody can push into without any special requirements to make the image available to all providers. This is likely to change in the future when we add an appropriate whitelisting mechanism.
To push the image:
gvmkit-build golemfactory/blender:demo --push
Once the image is pushed, the tool will output its hash, e.g.:
success. hash link 1a72390cbb08117b2d77373185e43701a747a3f7eb9a552c19aa5041
Please note down that hash as you'll need it in the definition of the payload you'll want to run on Golem.
Important: if you use a newly-pushed image in your task, you'll need to give providers some additional time to pull those images before they'll be able to publish offers that are compatible with your demand.
In other words, if you're running a task using a newly-pushed image, always specify a much longer task timeout on the first run.
