In this final section we describe three advanced job features that may be of interest.
At the start of this tutorial we mentioned that jobs run in isolation. In certain scenarious it is not practical to make a copy of the project directory for each job (e.g. you have very large files).
To overcome this issue you can start a new job using the flag --file-access=shared. The job will then access the host directory directly.
When using this flag, rememeber that the job may be affected if you start making changes in the project directory while it is running. Similarly, if the job creates temporary files or outputs, then it will all be immediately visible in your project directory.
If you want to interactively explore the job’s files without copying them back to the host directory, you can use the command
$ cjr job:exec $ID bash
Since this may be a common usecase, we create the following convienient shorthand
$ cjr job:shell $ID
It is also possible to start a Jupyter server that has access to the a job’s files (and not the project directory). This can be useful for interactivly viewing or post-processing results before copying them back to your project directory.
To start a jupyter server in a job, run the command
$ cjr job:jupyter:start ID # ID represents the ID of the parent job.
To stop the jupyter server you can run
$ cjr job:jupyter:stop ID
To list all jupyter servers that are running inside job file volumes use
$ cjr job:jupyter:ls