docker: update infra_image to new registry #23927
Conversation
The gcr.io container registry is shutting down in March. Update the default `image_image` for Docker's "pause" containers to point to the new location hosted by the k8s project. Fixes: #23911 Ref: https://hashicorp.atlassian.net/browse/NET-10942
tgross
force-pushed
the
infra-image-registry
branch
from
34ca05e
to
deb9410
Compare
There was a problem hiding this comment.
lgtm!
Regarding restrictive networks, I suspect (without evidence) that folks in such environments likely already set infra_image to some internal mirror, which would make this a no-op for them.
But that's just conjecture, so I agree with you on waiting to backport it until after <some time> after an announcement, since the alternative would be to suggest folks on older versions set this new value in their client config manually, which doesn't seem as user-friendly to me. I.e. the default should "just work" for our currently-supported versions.
There was a problem hiding this comment.
LGTM. I don't think preannouncing is necessary, and we should get this into backports ASAP. There's often a weeks-to-months long lag between when a user selects a target version to upgrade their cluster to before they actually roll that version out to every environment. I'd hate for them to select the announcement version only to discover mid-rollout they need to upgrade again or manually update their infra_image on every client.
If users upgrade before reading upgrade notes, then preannouncing doesn't help anyway. I vote we rip the bandaid.
Co-authored-by: Michael Schurter <[email protected]>
tgross
added
backport/ent/1.6.x+ent
The gcr.io container registry is shutting down in March. Update the default `image_image` for Docker's "pause" containers to point to the new location hosted by the k8s project. Fixes: #23911 Ref: https://hashicorp.atlassian.net/browse/NET-10942
The gcr.io container registry is shutting down in March. Update the default
infra_imagefor Docker's "pause" containers to point to the new location hosted by the k8s project.Fixes: #23911
Ref: https://hashicorp.atlassian.net/browse/NET-10942
Question/note for reviewers: I said in #23911 that this is a breaking change so it should go in 1.9.0, but in March when the registry move happens our 1.8 LTS will still be in its support window (as will 1.7 until ~May or so, assuming that's when 1.10 goes out). Anyone on these older versions will have new allocations fail unless they update their pause container. Maybe we should pre-announce backporting this to 1.8 and 1.7 and then do so after 1.9 goes GA to give folks some time to deal with it if they have restrictive firewall environments?