Description
I'm trying to implement a simple operator release workflow with the new FBC (1 operator, 1 bundle, 1 index - all with the same version). I'd like to perform as much of it locally before pushing the results. What I currently have (all wrapped in Make rules) is roughly this:
${CONTAINER_TOOL} build -f operator.Dockerfile -t ${OPERATOR_IMG} .
# ... generate the bundle and build bundle image
${CONTAINER_TOOL} build -f bundle.Dockerfile -t ${BUNDLE_IMG} .
# add bundle to index
$(OPM) render $(BUNDLE_IMG) --output=yaml >> my-operators-index/my-operator/bundles.yaml
# add bundle to stable channel
yq -i e '.entries += {"name": "my-operator.v$(VERSION)", "skips": .entries[-1].skips + .entries[-1].name}' my-operators-index/my-operator/stable-channel.yaml
# validate index
$(OPM) validate my-operators-index
# commit and tag the result
git commit ...
# build index
${CONTAINER_TOOL} build -f my-operators-index.Dockerfile -t ${INDEX_IMG} .
# publish all images
${CONTAINER_TOOL} push ${OPERATOR_IMG}
${CONTAINER_TOOL} push ${BUNDLE_IMG}
${CONTAINER_TOOL} push ${INDEX_IMG}
This would alloe me to keep building of the components and publication separate and make the process up to the commit idempotent and easy to separate into a step that can be performed locally and another step that requires registry write credentials.
Unfortunatelyopm render accesses the registry and it is not possible to reconfigure this behavior. Am I not using opm render correctly or this there another way to extract the bundle? The usage of opm render was inspired by the OpenShift 4.12 FBC docs: https://docs.openshift.com/container-platform/4.11/operators/understanding/olm-packaging-format.html#olm-fb-catalogs-example_olm-packaging-format