Print

[brian]

Hi,
I need to add a form of node provisioning, using your terminology from http://www.jppf.org/doc/5.2/index.php?title=Node_provisioning a facility "which allows dynamically growing or shrinking a JPPF grid based on the workload requirements." In my cloud environment a jppf worker node lives in its own isolated container, I never want >1 running in the same container. This means the ability for a node to spawn another one using process launcher is not helpful. Could this be made pluggable? In the "start a node" hook instead of launching a process I would tell our cloud service to launch a new worker container. On the "shrink the grid" side I was imagining some idle detection built into worker node and it self destructs if it hasn't been asked to do any work lately, but open to other approaches, and not familiar with how you do it today.

Today I added this node provisioning on the client side outside JPPF - the client knows their job's SLA so can ask cloud service to start up appropriate workers before submitting it. But ideally it's done within JPPF. Your thoughts?

[brian]

Hi,
Did you have a chance to think about whether this would make sense as a pluggable component?
Thanks.

[brian]

Hi,
Did you have a chance to think about whether this would make sense as a pluggable component?
Thanks.

[lolo]

Hello,

The problem is that there are so many types of containers: VMs for the many existing hyervisors, Docker images, Cloud images, each wiith different APIs and command-line management utilities. Trying to abstract a common interface for all of them would be like creating a standard to unify all standards: it would be one more standard.

JPPF already has pluggable MBeans, which seem a good fit for this kind of adminstrative operations. This is actually how the existing provisioning facility is implemented. This is really the approach I recommend. I'll be happy to assist if you encounter road blocks on the way, for instance if you have a need for some functionalities of the node to be unlocked or exposed. What do you think?

Thanks,
-Laurent

[brian]

To clarify, you are suggesting I write my own JPPFNodeProvisioningMBean subtype which knows how to start/stop nodes in my private cloud. And plug that in. Right?

[lolo]

It doesn't have to be a subtype of JPPFNodeProvisioningMBean, you can use any interface you wish for the MBean. Other than that, yes, this is what I am suggesting.

-Laurent