Repository navigation
Support for custom executors #689
Description
Gitlab has a nice feature where you can easily create a custom executor, it's documented here https://docs.gitlab.com/runner/executors/custom.html.
This allows one to easily support custom container runtimes such as podman, plain runc, singularity, or whatever container runtime you can think of.
At its core, gitlab gives you a script.sh with all the commands for a job, and you are supposed to make it execute somewhere.
Is such a feature possible with github as well? It would be incredibly useful for clusters where you would have e.g. slurm + rootless containers instead of kubernetes/docker.
Activity
ad-m commentedon Nov 22, 2020
Using your own executors would be a very interesting solution. I see that this would allow to create a fully secure worker based on ephemeral virtual machines (like hosted one). Currently, the worker relies on trust as previous tasks may interfere with subsequent tasks and only safe code execution should be executed.
This is important restriction, especially for public repositories and open-source projects that cannot guarantee the execution of only safe code. Some open source project – such as Apache Airflow – require additional computing power that cannot be provided by hosted GitHub Actions workers too.
It allows to execute jobs on AWS Fargate for serverless jobs: https://about.gitlab.com/blog/2020/05/11/introducing-autoscaling-gitlab-runners-on-aws-fargate/
If it is possible to guarantee that the workflow is small (especially based only on JavaScript actions), it could allow extremely fast execution of jobs in Azure Functions. I have a few workflows that regularly only execute a little JavaScript script. Creating an entire Virtual Machine for them is not effective.
vsoch commentedon Sep 16, 2022
Hi! Is this not supported by? https://docs.github.com/en/actions/hosting-your-own-runners/about-self-hosted-runners
vsoch commentedon Sep 16, 2022
We need to be able to run https://ecp-ci.gitlab.io/docs/admin/jacamar/auth.html#authorization-flow