forked from rook/rook
If a node is gone, we try to remove crash collector deployments so they don't stay stale or pending. We must ensure the node is gone and not just return an error even if we successfully deployment the orphan deployment. Signed-off-by: Sébastien Han <seb@redhat.com>