We are using the EC2 plugin to automatically launch instances on spot, and this generally works OK. One of the biggest problems, however, is that if AWS terminates the instance half way through a build due to spot price change or they need the capacity back, the pipeline spits out an error along the lines of "The channel is closing down or has closed down," and sits there forever, never recognising that the instance has gone and it will never complete. This means that you manually have to stop the build, and re-run it.

In an ideal world, the job would be put back on the queue to run on another instance, however I'd settle for it simply dying peacefully and the job failing or getting cancelled automatically.

I've done a search on here but can't find any issues that match this description.