Google cloud and AWS both support spot instances which offer significant savings. Rancher is almost there in terms of supporting spot instance automation.
Right now I can setup an autoscale group in Google cloud and define a startup script that auto joins rancher. I can then setup a shutdown script that uses the new API to deactivate and delete an instance. Awesome I can save thousands a month processing millions of tasks a day!
The only problem that remains is having containers deploy to these new machines. Scheduling one container per host doesn’t give me the scale I need. Scheduling x amount of instances doesn’t rebalance when I add new hosts.
I’d have to do something dirty right now to make this work. Like scaling down to 1 and back up. This features is a game changer if implemented.