https://rancher.com/ logo
Join Slack
Powered by
This message was deleted.
# k3s
a
This message was deleted.
c
Is the image 
<http://my.registry.com/devops/external-docker/nginx:v0.111|my.registry.com/devops/external-docker/nginx:v0.111>
available on one of your nodes? How exactly did you load it? Can you confirm that you see that image in 
ctr -n <http://k8s.io|k8s.io> image ls
on one of the nodes?
If you start k3s with --debug and check the k3s logs in journald, you should see various events from spegel when the image is loaded and pulled.
g
Hi! Thanks for your answer. I have only one node. The image was loaded with 
ctr -n <http://k8s.io|k8s.io> image import image.tar
. I can confirm that I see it with 
ctr -n <http://k8s.io|k8s.io> image ls
Copy code
<http://my.registry.com/stable/devops/external-docker/nginx:v0.111|my.registry.com/stable/devops/external-docker/nginx:v0.111>             application/vnd.docker.distribution.manifest.v2+json sha256:113ffbd0c92490728547fe48740bb084f187c57aa92f65d4351bd5018aeed93a 88.9 MiB  linux/amd64 io.cri-containerd.image=managed
I will try with the debug flag and report back.
One strange thing : if I try to start a simple test pod manually (with kubectl apply), I can load an image tagged with "docker.io" like 
<http://docker.io/library/alpine|docker.io/library/alpine>
that I loaded with ctr like the rest but did not re-tag with "my.registry.com". It seems the issue is limited to the images with "my.registry.com" at the beginning. For context, here is my 
registries.yaml
file: ```mirrors: docker.io: registry.k8s.io: my.registry.com:```
Here is the spegel debug message when trying to pull the image:
Feb 29 074236 ip-10-0-10-10 k3s[1797514]: time="2024-02-29T074236Z" level=info msg="spegel 2024/02/29 074236 \"msg\"=\"\" \"error\"=\"mirror resolve retries exhausted for key: my.registry.com/devops/external-docker/nginx:v0.111\" \"path\"=\"/v2/devops/external-docker/nginx/manifests/v0.111\" \"status\"=500 \"method\"=\"HEAD\" \"latency\"=\"389.43µs\" \"ip\"=\"127.0.0.1\" \"handler\"=\"mirror\""
c
If you have only one node then what do you need this for? You just need to use airgap tarballs to preload things into containerd. The whole point of this is to allow multiple nodes to share their copies of images.
g
Oh, I thought it was a pre-requisite for being able to use airgapped images. So loading the images with 
ctr -n <http://k8s.io|k8s.io> image import
is all I need? If so, why can I start a pod using an image prefixed "docker.io" and not "my.registry.com" when they are both loaded from the tarballs?
c
You could load them manually, or just drop the tarball in the images dir before starting k3s.
What are they tagged as in the tarball? How do they show up in containerd after being imported? What is the Image pull policy in your pod spec?
g
Tagged as : 
<http://my.registry.com/devops/external-docker/nginx:v0.111|my.registry.com/devops/external-docker/nginx:v0.111>
(for this image at lease, there are others) They show up the same after being imported (if 
ctr
is what I should use to get that info) The pull policy if 
IfNotPresent
Oh no... I'm so sorry... I just realised that there is a path part missing in the image names... Oh what a noob! I think it does work and that was just a keyboard/chair issue 😕 Thanks for your help ! I now know that I don't need the embedded registry/mirror 🙂
25 Views
Open in Slack
PreviousNext
Powered by