Hmm. So many responses. All of them on the order of "you did something wrong."
Newsflash: AWS doesn't always work correctly. I've used AWS for a very long time. I've personally witnessed instances that do not start, instances that do not stop, disk corruptions on deployed instances and network failures on running instances.
I've never seen a case where a public IP was not created. Until this morning. Now I can add that to the list.
For the record - here's what I verified :)
- Three identical instances in the cluster
- All instances are in the same availability zone
- All instances have same VPC
- VPC DNS settings are correct (resolution / hostnames enabled)
- All instances have same subnet
- Subnet has: a) public routing table; b) option enabled to create public IP
- Plenty of IP space available in the subnet
Two of the three instances receive a public IP. The third does not.
So for any others in the future getting to this post: No, you are not insane. Yes, it is possible that AWS screws up.
In our case, manually terminating the problem instance and issuing a new cluster up..."fixed" the problem.
And - I upvoted the answer that indicated a "launch more like this" from a STOPPED instance had an impact on public IP. Not because it is the correct answer (it is not) but because it demonstrates an admirable response to an otherwise inexplicable situation: trial and error / experimentation. The good old "Gee, what happens if I try this...". As cloud professionals: If all other standard troubleshooting steps fail and the only alternative is to blow away the instance (or subnet, or Lambda function, or DynamoDb, or SNS queue; whatever the failing resource) then it's wise to think outside the box and try other actions.
In other words: keep an open mind.