template with autoscaling group gets stuck and never completes

[Miguel Grinberg]

The following template creates an autoscaling group plus a separate server. It assigns the "current_size" attribute of the asg to the other server as metadata. When running the template, the autoscaling group completes, but the other server never starts. The same template without the metadata variable runs through completion just fine.

Here is the template:

  heat_template_version: 2013-05-23

  parameters:
    image:
      type: string
      label: Image name or ID
      default: cirros
    flavor:
      type: string
      label: Flavor
      default: m1.tiny
    network:
      type: string
      label: Network
      default: private

  resources:
    asg:
      type: OS::Heat::AutoScalingGroup
      properties:
        min_size: 1
        desired_capacity: 3
        max_size: 5
        resource:
          type: OS::Nova::Server
          properties:
            image: { get_param: image }
            flavor: { get_param: flavor }
            networks:
              - network: { get_param: network }

    lb:
      type: OS::Nova::Server
      properties:
        image: { get_param: image }
        flavor: { get_param: flavor }
        networks:
          - network: { get_param: network }
        metadata:
          sz: { get_attr: [asg, current_size] }

[Miguel Grinberg (miguelgrinberg)]

Further investigation revealed that the stack gets stuck when an exception is raised when evaluating attributes. Testing on latest Kilo release i was unable to reproduce the problem with the template shown above, but I can reproduce when I replace the last line with an invalid attribute, for example:

        sz: { get_attr: [asg, bad_attr] }

I was working with Juno last week when I found this bug. I'm thinking that maybe I mistyped "current_size" and that is why it hung for me on that property.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/168203
Committed: https://git.openstack.org/cgit/openstack/heat/commit/?id=29365bda1cab48aad7b98907af521130321e526a
Submitter: Jenkins
Branch: master

commit 29365bda1cab48aad7b98907af521130321e526a
Author: Angus Salkeld <email address hidden>
Date: Fri Mar 27 12:00:42 2015 +1000

    Set state to failed if we get a base exception

    The scheduler only raises ExceptionGroup if aggregate_exceptions is
    True. This change adds a catch-all exception handler that sets the stack
    to failed on any exceptions.

    Co-Authored-By: Miguel Grinberg <email address hidden>
    Change-Id: I76a6e87924416335921115ec3147a991659cfb2e
    Closes-Bug: #1436128