Hello, we are currently planning on upgrading to 4.X from 3.X - and as part of this we intend to make changes to the interface. Is it possible to run a both versions on the same cluster and have user switch back and forth in order to test proposed upgrade or will it cause issues for users?
There’s no way to run this with both versions I’m aware of. You should follow the typical sys admin workflow and ensure you have a dev and test system to build the new version on and test before you make the switch for your prod users so that you are aware of any changes and ready for them.
Thanks Travis - so just be clear: running a test version at 4.X and a separate (on a different server/vm) prod at 3.X is not possible? In which case how can you ‘test’ the test version?
I think the key issue here is your definition of ‘cluster’. Travis was indicating you can’t run multiple versions on the same hardware. But you CAN run different versions on different hardware that talk to the same cluster.
For example, at OSC we have 3 distinct clusters (i.e. physical racks of systems, each with their own dedicated login nodes and SLURM schedulers). We also have at least 4 different instances of Open OnDemand running. Each instance can talk to all three of the clusters / schedulers. Each instance of Open OnDemand has a production environment, a test environment, and a development environment. Those environments are each running on their own virtual machines. So in essence we have at least 12 virtual machines, each with their own complete Open OnDemand deployment.
We regularly upgrade just some of those virtual machines to the latest version for testing, while leaving the production versions at an older version.
I think the confusion comes when people don’t utilize VMs, but rather run Open OnDemand on bare metal nodes, in which case you can’t run multiple versions on the same node.
Does this clarify the situation?
Thanks Alan - indeed that does clarify for me!
So do each of those 12 instances of OOD you have have their own URLs? If so, how do you track them and communicate them to users? If now what do you do instead?
So do each of those 12 instances of OOD you have have their own URLs?
Yes.
If so, how do you track them and communicate them to users?
They’re used for specific purposes. For example one is a part of an agreement with a commerical client. Another one is for classrooms, so when the classroom is requested & created then it’s communicated at that time.
Tracking we keep everything in puppet for configurations, but I also have to use nslookup from time to time to figure out which is which.