Skip to content

Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure

TypeScript

test(branch-profile): give each kubeconfig refusal its own token (R4-L8)

Alfredo Lopez7a69327f17ee
diff --git a/scripts/tests/test-branch-profile-lifecycle.sh b/scripts/tests/test-branch-profile-lifecycle.sh
index 19ed86e70..e3939c86e 100755
--- a/scripts/tests/test-branch-profile-lifecycle.sh
+++ b/scripts/tests/test-branch-profile-lifecycle.sh
@@ -161,11 +161,17 @@ if [[ "${1:-}" == config ]]; then
exit 0
;;
# kube-contexts holds one "<context> <server>" line per kubeconfig context.
+ # get-contexts-fails makes the kubeconfig unreadable, as a malformed file
+ # or one kubectl cannot open does.
get-contexts)
[[ "$*" == 'config get-contexts -o name' ]] || {
printf 'kubectl stub: unsupported config call: %s\n' "$*" >&2
exit 64
}
+ [[ ! -e "${state}/get-contexts-fails" ]] || {
+ printf 'error: stub: unable to read the kubeconfig\n' >&2
+ exit 1
+ }
awk '{ print $1 }' "${state}/kube-contexts"
exit 0
;;
@@ -184,6 +190,17 @@ if [[ "${1:-}" == config ]]; then
printf 'error: context was not found for specified context: %s\n' "${wanted}" >&2
exit 1
}
+ # Two server markers model a dangling context: @missing-cluster names a
+ # cluster the kubeconfig does not define, which `view --minify` refuses
+ # as client-go does; @empty-server is a cluster entry with no server,
+ # for which the jsonpath prints nothing.
+ case "${server}" in
+ @missing-cluster)
+ printf 'error: cannot locate cluster %s\n' "${wanted}" >&2
+ exit 1
+ ;;
+ @empty-server) exit 0 ;;
+ esac
printf '%s' "${server}"
exit 0
;;
@@ -357,7 +374,7 @@ reset_state() {
profiles/external-rest-api rpc-proxy/rpc-proxy >"${state}/services"
: >"${state}/reachable"
rm -f "${state}/fail-get-deploy" "${state}/curl-fail" "${state}/start-switches-context" \
- "${state}/use-context-fails" "${state}/minikube-start-fails" \
+ "${state}/use-context-fails" "${state}/minikube-start-fails" "${state}/get-contexts-fails" \
"${state}/pf-ignores-term" "${state}/minikube-stop-fails" "${state}/minikube-delete-fails" \
"${state}/node-label" "${state}/minikube-profile-list-fails" "${state}/docker-info-fails"
# By default the branch profile's context is this local Minikube, minikube
@@ -622,6 +639,49 @@ for action in start delete; do
assert_log_lacks 'kubectl --context=' "${action} of a profile whose context is a remote DNS name must not address the cluster"
done
+# Each refusal of the kubeconfig read names its own cause. An unreadable
+# kubeconfig and a dangling context (a context whose cluster is not defined,
+# or whose cluster has no server) are not remote endpoints: reporting them as
+# BRANCH_PROFILE_REMOTE_CONTEXT sends the operator after a cluster that does
+# not exist. The remote loops above are the witness that a non-local endpoint
+# still refuses as BRANCH_PROFILE_REMOTE_CONTEXT.
+for action in start status pf pf-health health stop setup delete; do
+ reset_state
+ : >"${state}/get-contexts-fails"
+ bp "kubeconfig-unreadable-${action}" "${action}" "CONFIRM_DELETE=${profile}" "CONFIRM_PROFILE=${profile}"
+ assert_rc 1 "${action} of a profile whose kubeconfig cannot be read"
+ assert_output_has 'BRANCH_PROFILE_KUBECONFIG_UNREADABLE: unable to read the kubeconfig contexts' \
+ "${action} of a profile whose kubeconfig cannot be read"
+ assert_output_lacks 'BRANCH_PROFILE_REMOTE_CONTEXT' \
+ "${action} of a profile whose kubeconfig cannot be read is not a remote context"
+ assert_log_has 'kubectl config get-contexts -o name' "${action} tried to read the kubeconfig contexts"
+ assert_log_lacks 'kubectl config view' "${action} stopped at the unreadable context list"
+ assert_no_calls_to minikube "${action} of a profile whose kubeconfig cannot be read must not call minikube"
+ assert_log_lacks 'kubectl --context=' "${action} of a profile whose kubeconfig cannot be read must not address the cluster"
+ assert_no_calls_to docker "${action} of a profile whose kubeconfig cannot be read must not reach Docker"
+
+ for dangling in missing-cluster empty-server; do
+ reset_state
+ write_kube_contexts "${profile}" "@${dangling}"
+ bp "context-${dangling}-${action}" "${action}" "CONFIRM_DELETE=${profile}" "CONFIRM_PROFILE=${profile}"
+ assert_rc 1 "${action} of a profile whose context is dangling (${dangling})"
+ if [[ "${dangling}" == missing-cluster ]]; then
+ assert_output_has "BRANCH_PROFILE_CONTEXT_DANGLING: kube context ${profile} is in the kubeconfig, but kubectl could not resolve its cluster" \
+ "${action} of a profile whose context names an undefined cluster"
+ else
+ assert_output_has "BRANCH_PROFILE_CONTEXT_DANGLING: kube context ${profile} has no API server" \
+ "${action} of a profile whose context has an empty server"
+ fi
+ assert_output_lacks 'BRANCH_PROFILE_REMOTE_CONTEXT' \
+ "${action} of a profile whose context is dangling (${dangling}) is not a remote context"
+ assert_log_has "kubectl config view --raw --minify --context=${profile}" \
+ "${action} read the dangling context (${dangling}) from the local kubeconfig"
+ assert_no_calls_to minikube "${action} of a profile whose context is dangling (${dangling}) must not call minikube"
+ assert_log_lacks 'kubectl --context=' "${action} of a profile whose context is dangling (${dangling}) must not address the cluster"
+ assert_no_calls_to docker "${action} of a profile whose context is dangling (${dangling}) must not reach Docker"
+ done
+done
+
# The reported shape: an adopted clerum-* profile named like a remote cluster's
# context. delete must refuse without running `minikube -p clerum-prd delete`.
write_owned_profile_copy clerum-prd