Separate user accounts - now or later

[Zarquan]

In the current Ansible based deployment all the notebooks are run using the same Linux user account, fedora, regardless of which Zeppelin account the user is logged in as. Each of us also has an individual Linux user account, which we use to login to the server via ssh. The mismatch between these two patterns causes some unexpected behaviour to do with file permissions and access.

• Everything that Zeppelin does, including everything done in a %sh paragraph is done as the fedora user.
• In order to allow Zeppelin notebooks to access data in our /user directories, the permissions on our /user directories have to read/write to all.
• Setting the permissions on our /user directories to read/write to all means everyone can overwrite everyone else's data (very easy to do when we are sharing notebooks).
• Our user Linux accounts don't have a /home directory where they can store secrets.

Long term plan, we will want to isolate Zeppelin notebook execution on a per user basis, so each person has a separate Linux user account, notebooks are executed using that person's Linux uid and file ownership and permissions are set in a consistent manner.

We can choose to make that change now, while we only have two users, or we can put this off until later when the transition will be more painful.

What do you think - some pain now or much pain later ?

[Zarquan]

The reason I'm looking at this now is because I'm mid way through setting up a mechanism for our users to import and export data to/from the system. We have been using ssh for this, but I'm also looking into alternatives like using the ownCloud system at Edinburgh as a staging platform.

Whichever solution we use the expected behaviour is that files you upload belong to your account, are stored in a directory with your name on it, and are not publicly accessible unless you take specific action to make them public.
In addition, whichever solution we use relies on having a home directory where users can store credentials, passwords and ssh keys safely.

All of which doesn't sit well with running everyone's notebooks using the same generic Linux account.

[NigelHambly]

I vote pain now: the sooner we address long-term administration issues the better.