Idea #18337
Updated by Peter Amstutz 3 months ago
This epic is about minimizing the friction involved in running Arvados at the single host scale. There are a few versions of this:
* Arvados appliance: Arvados pre-installed on dedicated hardware and shipped to users
* Self-install: users self-installing Arvados on dedicated hardware similar to the appliance
* Desktop install: users run Arvados using a container or VM on their laptop, with convenient desktop integration
* Ephemeral instances: Start a fresh Arvados instance, use it to run computations, export the results, and then the Arvados instance is suspended or wiped out. The exported results should include all of the provenance information.
The goal for all of these is for someone to be able to get an Arvados instance running in a few clicks.
The follow up to this is that once they have access to Arvados, they are not looking at an empty instance, but have easy access to biomedical data and workflows they can use to start doing stuff immediately.
*Previously*
Setting up a new single-node Arvados cluster should look something like
<pre>
# add-apt-repository ...
# apt-get install arvados-server
# arvados-server init [some options]
[answer some questions]
OK, you can now log in to https://workbench2.example.com/
# arvados-server health
OK
</pre>
Scaling up by adding new system nodes should be just as easy.
It is okay if some installation environments (or sysadmin preferences) necessitate additional manual intervention. The idea is to make it extremely easy to start a production-quality cluster on a few suitable types of system that most people can get access to, like an on-prem server or publicly reachable cloud instance running Debian stable.
Things you don't currently get by installing arvados-server-easy on a cloud VM:
* Cloud storage (currently it just stores on local disk by default, in /var/lib/arvados)
* -Diagnostics tool should use system root token from the config file when run on a server node, instead of requiring env vars- #17344