Description
`ec2/spark_ec2.py` does not perform any sanity checks on hostnames when testing connectivity in `is_ssh_available` and descendants.
This causes unexpected behavior when running a cluster in a VPC subnet without public IPs if `-private-ips` is not given. While `-private-ips` should be required in this context, the failure mode currently present is suboptimal.
[ ... ]
All 1 slaves granted
Launched master in us-west-1c, regid = r-redacted
Waiting for AWS to propagate instance metadata...
Waiting for cluster to enter 'ssh-ready' state…………Password:
What has happened here is that the public dns name for the instance is a null string, causing the ssh check later in the script to inadvertently connect to localhost to test connectivity to the cluster. The password prompt here is OS X's sshd asking to auth the user on that connection.
Attachments
Issue Links
- links to