Images are themselves manifests, and reference a config blob #79
Conversation
openshift-ci-robot
added
the
size/XL
smarterclayton
force-pushed
the
correct_image_layer_name
branch
from
8ddd177 to
df28b03
Compare
image/v1/types.go
Outdated
| // imagesWithoutMetadata contains information about all images referenced by an image | ||
| // stream whose metadata could not be loaded. This can be due to an administrator or | ||
| // other automated process deleting the image (in which case the image cannot be pulled) | ||
| // or be a temporary condition because the image has not yet been uploaded. |
There was a problem hiding this comment.
@smarterclayton when can we have a reference to the image that has not yet been uploaded?
There was a problem hiding this comment.
If someone writes an integration that uses the API but does so without verifying the references. We don't allow it directly, but I want to leave this open so that a consumer of the API knows that this is possible even if we're confident none of our existing code paths allow it.
There was a problem hiding this comment.
I can change the godoc if you'd like, but I do want to call out that this API is dependent on both the object references and the object metadata.
There was a problem hiding this comment.
is it better to have this as a separate field, or have a missing attribute of ImageBlobReferences that can only be true if config and layers are unset? looking up an image in two places is a pain
There was a problem hiding this comment.
I could do either.
There was a problem hiding this comment.
@liggitt @smarterclayton I prefer a missing attribute on ImageBlobReferences. I don't care for parallel maps.
|
@deads2k since liggitt is out - this is a must for 3.11 (since the field is new). Jordan and I had verbal agreement on a separate map for tracking missing references, but we didn't talk naming or structure. |
|
one question on the separate map vs additional field on existing ImageBlobReferences type. LGTM otherwise |
The naming of the nested field was wrong (this known as the image "config") and we need to correct it. An OpenShift image is always a manifest.
smarterclayton
force-pushed
the
correct_image_layer_name
branch
from
df28b03 to
a6b8d79
Compare
|
Updated |
|
Need to regenerate proto. We've been encouraging https://github.com/openshift/api#pull-request-process. I don't think this will cause problems, but something to keep in mind. /lgtm |
|
Ok, will do it now. Didn't see that before. |
Communicates when an image stream references an image api object that has been deleted by an administrator.
smarterclayton
force-pushed
the
correct_image_layer_name
branch
from
a6b8d79 to
1ba94e5
Compare
|
New changes are detected. LGTM label has been removed. |
openshift-ci-robot
added
size/L
The naming of the nested field was wrong (this known as the image "config") and we need to correct it. An OpenShift image is always a manifest. This is naming only.
Second change: Allow the API to return info about images whose references are not resolvable
Administrators may delete an image intentionally or accidentally, and the layers API needs to be able to inform the registry that the image is referenced, but no layer data is available, so that the registry can more accurately return an error to the client.
@liggitt as discussed. The first is an error @bparees and I found when updating api.ci. The second is our discussion about references.