+1 agreed. Also it's not like we can just kick off a new pipeline run while the other one is running -- if the initial pipeline run that was kicked off takes more than 10 minutes, the next scheduled one will get cancelled due to timeout waiting in the queue, which means we need to pretty much set an alarm for 10 minutes and come back and kick it off again.

The issue with cancelling pipelines is state management.

Suppose you're running a pipeline that modifies 3 different containers. If you stop half-way, what happens? It's easy to say "Well, roll back." -- but there's just so much risk of error there. One of the difficulties in building Cycle is walking the line between giving as much flexibility as possible while also working to prevent situations that are extremely error-prone.

If we made pipelines cancelable prior to running (like if they're in the queued state), would that help?