EC2 instance with predefined IP and SubnetId: network interface stays around after instance termination

[wandhydrant]

Hello, and many thanks for Moto!

I ran Moto against some use cases and found a few discrepancies between AWS and Moto. This is my first issue. I hope to have found a real bug, and not to have made a mistake.

When deploying EC2 instances where an IP address has been manually specified, and DeleteOnTermination is set to True, I found that Moto keeps the Network Interface around even after termination. When deploying a new EC2 instance with the same IP, this results in an exception "IP already in use". Here is example code to demonstrate this behaviour, ran with AlmaLinux 8 / Python 3.9.19 in a venv with only moto[server] 5.0.11:

import boto3
from moto.server import ThreadedMotoServer

server = ThreadedMotoServer()
server.start()
server._server.passthrough_errors = True  # otherwise we just see a 500

ec2_client = boto3.client("ec2", endpoint_url="http://localhost:5000")

existing_subnet_id = ec2_client.describe_subnets()["Subnets"][0]["SubnetId"]

def create_instance():
    instance = ec2_client.run_instances(
        MaxCount=1,
        MinCount=1,
        ImageId="ami-bb9a6bc2",
        InstanceType="t3a.small",
        NetworkInterfaces=[{
            "DeleteOnTermination": True,
            "PrivateIpAddress": "172.31.0.5",
            "DeviceIndex": 0,
            "SubnetId": existing_subnet_id
        }]
    )["Instances"][0]    
    ec2_client.get_waiter("instance_running").wait(InstanceIds=[instance["InstanceId"]])
    return instance

def get_interfaces():
    return ec2_client.describe_network_interfaces(
        Filters=[{"Name": "subnet-id", "Values": [existing_subnet_id]}]
    )["NetworkInterfaces"]
    
assert len(get_interfaces()) == 0

first_instance = create_instance()
assert len(get_interfaces()) == 1
    
ec2_client.terminate_instances(InstanceIds=[first_instance["InstanceId"]])
ec2_client.get_waiter("instance_terminated").wait(InstanceIds=[first_instance["InstanceId"]])

print("After termination:")
print(get_interfaces())
# => a network interface still exists, with "DeleteOnTermination": False

create_instance()
# => Exception("IP already in use")

server.stop()

This happens with the existing subnet (as in the code above) or with a newly created subnet. However, if I do not specify the SubnetId in the run_instances() call, the same existing subnet is used, the network interface is still visible after termination, it still shows DeleteOnTermination=False, but the second run_instances() succeeds and there is no duplicate IP exception.

[bblommers]

Hi @wandhydrant, thanks for raising this and for providing the code sample! I've opened a PR with a fix - NIC's (with DeleteOnTermination=True) are now deleted after an instance is terminated.