Provisioning spot instance can leave dangling spot request

[jimfleming]

Using docker-machine to provision a spot instance opens a spot request and waits for it to be fulfilled. If this fails, docker-machine exits but the spot request remains open and may be fulfilled later. This results in a dangling spot request with an active instance.

[alexanderkjeldaas]

This seems to happen in many other cases as well.

$ ./docker-machine-setup.sh  # Run docker-machine create with spot instance..
Running pre-create checks...
Creating machine...
(aws01) Launching instance...
(aws01) Waiting for spot instance...
^C
[0]  ~$ docker-machine rm aws01
About to remove aws01
Are you sure? (y/n): y
Error removing host "aws01": unknown instance
MissingParameter: The request must contain the parameter KeyName
        status code: 400, request id:

Similarly if the spot request is canceled in the console. Same thing happens.

[alexanderkjeldaas]

Also:

$ docker-machine ls
NAME    ACTIVE   DRIVER      STATE   URL   SWARM   DOCKER    ERRORS
aws01   -        amazonec2   Error                 Unknown   MissingParameter: The request must contain the parameter InstanceId
        status code: 400, request id:

It seems like the docker-machine state machine is completely missing a state - the fact that this machine has not been allocated yet?!! Not having allocated a spot instance is not an 'Error' state.

[chillosu]

This plagues our CI runners constantly with spot instances.

[peternijssen]

Same issue here. Someone figured a work-around?

[nolith]

I did a patch for a proper cleanup of dangling resource in case of AWS machine creation failure. #4317