If a cluster launch is interrupted before AWS can even return a list of instances, we hit a part of the code where cluster_instances is not defined. We should protect against that.
Additionally, when Flintrock comes across a broken cluster (e.g. missing tags) left behind by an interrupted launch, we should:
Display a warning from inside _get_cluster_master_slaves().
Return all the untagged instances as slaves.
Triggered by the discussion with @douglaz in #179.
If a cluster launch is interrupted before AWS can even return a list of instances, we hit a part of the code where
cluster_instancesis not defined. We should protect against that.Additionally, when Flintrock comes across a broken cluster (e.g. missing tags) left behind by an interrupted launch, we should:
_get_cluster_master_slaves().Triggered by the discussion with @douglaz in #179.