We have quite a few ec2 slave configurations. Recently we've added a configuration which uses spot instances.

When a job is run which triggers the request to launch the spot instance, and so long as the instance is not launched or deleted, other jobs triggering the launch of the non-spot instances will hang forever and their instances will not be attempted to be launched.

As soon as I delete the spot instance (which is in the down state since it hasn't called back yet), the non-spot instance is provisioned and and the hung job is executed.