Also need to consider how replication controllers produce IDs for the pods they make, right now I think they just copy the possibly-blank ID field in their pod template.

If Pod ID is required, and a ContainerManifest ID is also required, should the pod controller duplicate the pod id into the container manifest id if the container manifest id is not specified? Then the replication controller should also generate a unique container manifest ID in the template on creation if one is not specified.

I will make that change in #253 if so

I think so. Re3plication controller should add a short random
prefix/suffix to each ID

On Mon, Jun 30, 2014 at 12:39 PM, Clayton Coleman notifications@github.com
wrote:

If Pod ID is required, and a ContainerManifest ID is also required, should
the pod controller duplicate the pod id into the container manifest id if
the container manifest id is not specified? Then the replication controller
should also generate a unique container manifest ID in the template on
creation if one is not specified.

I will make that change in #253
#253 if so

Reply to this email directly or view it on GitHub
#294 (comment)
.

With #846 the need for a manifest ID on the primary source may go away, once we can isolate the Kubelet from the etcd schema. Probably still need a source of uniqueness on config from other sources, but it could be source specific instead of being inside the source.

Addressed by implementation of #1225?

EDIT: Yes, will add this to #1519

@smarterclayton Should we expect that this has been addressed in v1beta1 by the BoundPod conversion functions in pkg/api/conversion.go? If I try to create a pod without the manifest version and id, it fails, since those fields are required. However, the conversion functions appear to trash the user-provided id value, and there's no choice for version. Example:

apiVersion: v1beta1
kind: Pod
id: www
desiredState:
  manifest:
    version: v1beta1
    id: www-manifest
    containers:
      - name: nginx
        image: dockerfile/nginx

result:

cluster/kubectl.sh get -o json pod www
{
    "metadata": {
        "name": "www",
        "namespace": "default",
        "selfLink": "/api/v1beta1/pods/www",
        "resourceVersion": "12",
        "creationTimestamp": "2014-11-04T06:03:29Z"
    },
    "desiredState": {
        "manifest": {
            "version": "v1beta1",
            "id": "www",
            "uuid": "4c6e7bf1-63e8-11e4-a40f-42010af03d18",
            "volumes": null,
            "containers": [
                {
                    "name": "nginx",
                    "image": "dockerfile/nginx",
                    "imagePullPolicy": ""
                }
            ],
            "restartPolicy": {
                "always": {}
            }
        },
        "status": "Waiting",
        "host": ""
    },

BoundPod gets half of it. Once we land the internal pod refactor the validator can stop validating version and id (they become optional and ignored). We will still need to support the backward compatible validator for old ContainerManifests on disk though.

----- Original Message -----

@smarterclayton Should we expect that this has been addressed in v1beta1 by
the BoundPod conversion functions in pkg/api/conversion.go? If I try to
create a pod without the manifest version and id, it fails, since those
fields are required. However, the conversion functions appear to trash the
user-provided id value, and there's no choice for version. Example:

apiVersion: v1beta1
kind: Pod
id: www
desiredState:
  manifest:
    version: v1beta1
    id: www-manifest
    containers:
      - name: nginx
        image: dockerfile/nginx

result:

cluster/kubectl.sh get -o json pod www
{
    "metadata": {
        "name": "www",
        "namespace": "default",
        "selfLink": "/api/v1beta1/pods/www",
        "resourceVersion": "12",
        "creationTimestamp": "2014-11-04T06:03:29Z"
    },
    "desiredState": {
        "manifest": {
            "version": "v1beta1",
            "id": "www",
            "uuid": "4c6e7bf1-63e8-11e4-a40f-42010af03d18",
            "volumes": null,
            "containers": [
                {
                    "name": "nginx",
                    "image": "dockerfile/nginx",
                    "imagePullPolicy": ""
                }
            ],
            "restartPolicy": {
                "always": {}
            }
        },
        "status": "Waiting",
        "host": ""
    },

---

Reply to this email directly or view it on GitHub:
#294 (comment)

I'm going to close this. IMO, we should focus on v1beta3.