Skip to content

fix(e2e): deliver image via k3d direct mode (root cause of the tools-node race) #186

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sunib merged 1 commit into from
Jul 2, 2026
Merged

fix(e2e): deliver image via k3d direct mode (root cause of the tools-node race) #186

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 24 additions & 12 deletions hack/e2e/load-image.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -165,21 +165,33 @@ runtime_image_id_in_node() {
import_image() {
local ref="$1"
local attempt node_name pinned
# k3d's tools-node import path can transiently lose the tarball between the
# save and the per-node `ctr image import` ("ctr: open /k3d/images/<tar>:
# no such file or directory") — and k3d still exits 0, logging "Successfully
# imported". The pin step below is what actually detects the miss, so on a
# failed pin re-run the whole import instead of giving up.
# Deliver with `-m direct` (docker save streamed straight into each node's
# `ctr image import` stdin) rather than the default tools-node mode. The
# tools-node path has a race: k3d's exec poll can misread the save-image
# exec as finished before the tarball hits the shared volume, the node
# import then fails ("ctr: open /k3d/images/<tar>: no such file or
# directory") — and k3d logs the error but still exits 0. Direct mode has
# no tarball/volume window and DOES propagate node import failures.
# Caveat: direct + ctr's --all-platforms chokes on multi-arch images saved
# by a containerd-store docker (foreign-platform blobs absent from the
# stream); PROJECT_IMAGE is always the single-arch controller image.
#
# Trust nothing either way: the pin step below verifies the image really
# landed in every node, and any failed attempt (loud or silent) is retried.
for attempt in 1 2 3; do
echo "Importing ${PROJECT_IMAGE} into k3d cluster ${CLUSTER_NAME} (attempt ${attempt}/3)"
"${K3D}" image import -c "${CLUSTER_NAME}" "${PROJECT_IMAGE}"
pinned=1
while IFS= read -r node_name; do
if ! pin_imported_image "${node_name}" "${ref}" "${PROJECT_IMAGE}" "${IMAGE_REPO}" "${IMAGE_TAG}"; then
pinned=0
break
fi
done < <(cluster_node_names)
if "${K3D}" image import -m direct -c "${CLUSTER_NAME}" "${PROJECT_IMAGE}"; then
while IFS= read -r node_name; do
if ! pin_imported_image "${node_name}" "${ref}" "${PROJECT_IMAGE}" "${IMAGE_REPO}" "${IMAGE_TAG}"; then
pinned=0
break
fi
done < <(cluster_node_names)
else
echo "WARN: k3d image import exited nonzero on attempt ${attempt}" >&2
pinned=0
fi
if [[ "${pinned}" == "1" ]]; then
return 0
fi
Expand Down