better handling when 'ctr image pull' fails
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Charmed Kubernetes Testing |
Triaged
|
Medium
|
Unassigned |
Bug Description
I've noticed cdk-addons failures across all supported releases for what seems to be network errors:
https:/
sudo lxc exec image-processor -- ctr image tag docker.
ctr: image "docker.
https:/
sudo lxc exec image-processor -- ctr image tag quay.io/
ctr: image "quay.io/
https:/
sudo lxc exec image-processor -- ctr image tag docker.
ctr: image "docker.
These upstream images do indeed exist, and there doesn't seem to be any pattern about which image causes the failure.
We always continue on an "image not found" error on the pull. Our assumption was that this error would only happen if an unsupported arch was requested, so we'd just move on to the next image:
However, now it seems like there's a different pull error and the above "grep not found" is not breaking out of our loop. This leads to us trying to tag a nonexistent image a few lines later:
We'll need to figure out why the pull failure is occurring -- perhaps retrying the pull until we get something, or die after a sane number of tries.
summary: |
- retry ctr image pull on failure + better handling when 'ctr image pull' fails |
Changed in charmed-kubernetes-testing: | |
importance: | Undecided → Medium |
status: | New → Triaged |