Reset a Hub

Reset a Hub#

Some hubs are used “seasonally”, and need to be “reset” after the end of the workshop / event. This helps save on disk space as well as keep the admin interface clean.

This is a destructive action, so should only be done in consultation with the community champion. Let them know what’s going to happen, and await confirmation before performing the action. You can use the following template.

Note

We are going to “reset” the hub. This would do the following:

  1. Remove all user home directories. The shared directories would not be touched, but you can clean them up yourself if you need.

  2. Remove all the users from the admin view, except 2i2c engineers and those explicitly marked as admins via our configuration. Those are:

    {{ list of admin users from the config.yaml file for this hub}}

This will also forcibly log out and stop the server of any user running at that time.

Let us know if this sounds ok, and we will proceed!

Once they respond affirmatively, you can begin.

Delete all home directories#

This clears up space and reduces their cloud costs.

  1. Get a shell with the home directories mounted

    deployer exec homes <cluster-name> <hub-name>

    This may take 30s-1minute.

  2. Look for directories with the word shared in them. Normally you would only find _shared-public and _shared, but check just in case there’s another one. We don’t want to accidentally delete anything else!

    cd /home
ls | grep shared

    If you find other directories named shared, escalate to the engineering slack channel to help figure out what to do next.

  3. Delete everything other than the shared directories. This may take a while (upto a hour) depending on the size of the home directories.

    ls /home | grep -v -P "_shared|_shared-public" | xargs rm -rf

  4. Exit out of the pod, as you’re done now!

    exit

Clear out existing list of users#

This keeps the admin interface clean, which is quite helpful for workshop hubs.

  1. Make sure that there are no user servers running. If they are, stop them.

    deployer use-cluster-credentials <cluster-name>
kubectl -n <hub-name> delete pod -l component=singleuser-server

  2. The easiest way to do this is to simply clean out the sqlite db that JupyterHub uses to store information. Get a shell on the hub pod for this hub:

    deployer exec hub <cluster> <hub>

  3. Rename the jupyterhub.sqlite file to a different name. This way, we preserve the db if we need it for some reason in the future, but also force JupyterHub to create one from scratch.

    mv jupyterhub.sqlite jupyterhub.sqlite.$(date --iso-8601)
exit # we are done, get out.

    This will create a jupyterhub.sqlite.<current-date> file.

  4. We now need to restart the JupyterHub process. The easiest way to do that is to kill the JupyterHub pod, so kubernetes will bring up a new one.

    deployer use-cluster-credentials <cluster-name>
kubectl -n <hub-name> delete pod -l component=hub

    The hub pod should restart, and everything should be fine now.

Notify the community#

Once this is done, let the community know the process has been completed.

Contents