Working directories

Local backend: A single working directory per job

When fractal-server executes a task, it has to read/write several execution-related files (note: these are files used internally by fractal-server, and not the actual scientific data being processed). These files are stored in a working directory which is unique for each workflow execution, with a name like workflow_000001_job_000001 (and within the parent directory specified in FRACTAL_RUNNER_WORKING_BASE_DIR). All files in this folder are owned by the user running fractal-server, and the permission set of each folder like workflow_000001_job_000001 is set to 755. Note that the folder is created as part of the submit_workflow function.

SLURM backend: Server-side and user-side working directories

When using the SLURM backend, parts of the execution-related files are written by the user who submitted the workflow for execution, who is impersonated through the sudo -u command. Such user has no access to the server-side working directory (the one in FRACTAL_RUNNER_WORKING_BASE_DIR), and therefore uses a local (i.e. user-side) working directory. By default, as a parent directory we use the cache_dir attribute of the current User. Since the user running fractal-server may not have access to this user-side directory, several operations are run through sudo -u (e.g. checking whether a file exists, or copying its content).