Computational integrations¶

On top of exposing a web API, a Fractal instance must be integrated to at least one computational resource which is used mainly for two goals:

• Setting up Python environments for task groups - see more details.
• Executing scientific tasks - see more details.

Supported integrations¶

The configuration variable FRACTAL_RUNNER_BACKEND determines which one of the three following modality is in-place:

1. In a local instance, every computational operation (setting up task environments and executing tasks) is run by the machine user who is running fractal-server, and the instance typically only has a single user. Note: this integration is mostly supported for testing and development.

2. A SLURM/sudo instance requires access to a SLURM cluster, with some additional assumptions - notably:

   • The user that runs fractal-server also needs sufficient permissions to impersonate other users for running jobs (e.g. via sudo -u some-user sbatch /some/submission-script.sh);
   • There must be a shared filesystem which both the user running fractal-server and other users have access to.

3. A SLURM/SSH instance requires access to a SLURM cluster through SSH, by impersonating one or several service users.

The specific configuration for each computational resource is defined in the Resource database table, with the following creation schemas:

• Local resource
• SLURM/sudo resource
• SLURM/SSH resource

For each resource, there may be one or many computational profiles, with the following creation schemas:

• Local profile
• SLURM/sudo profile
• SLURM/SSH profile

Here are some minimal examples of how to configure resources and profiles in the three different cases:

--8← "docs/assets/resource_and_profile/snippet.md"