- reuse the previously called cluster_info, rather than calling it again
- get all the instances from the config atomically, to prevent race conditions
- use a list comprehension, for simplicity

Reviewed-by: iustinp