recovery(ananke): avoid unnecessary longhorn sidecar churn

scripts/cluster_power_recovery.sh

@@ -1048,6 +1048,31 @@ remove_longhorn_manager_prepull_sidecar() {
   done <<< "${indexes}"
 }
 
+longhorn_manager_prepull_sidecar_has_pull_failures() {
+  kubectl -n longhorn-system get pods -l app=longhorn-manager -o json \
+    | jq -e '
+      [
+        .items[].status.containerStatuses[]?
+        | select(.name == "pre-pull-share-manager-image")
+        | select(((.state.waiting.reason // "") | test("ImagePullBackOff|ErrImagePull|CreateContainerError|RunContainerError|InvalidImageName")))
+      ]
+      | length > 0' >/dev/null 2>&1
+}
+
+remove_longhorn_manager_prepull_sidecar_if_needed() {
+  if ! harbor_endpoint_is_ready 1; then
+    warn "Removing Longhorn manager pre-pull sidecar because Harbor registry API is unhealthy."
+    remove_longhorn_manager_prepull_sidecar
+    return 0
+  fi
+  if longhorn_manager_prepull_sidecar_has_pull_failures; then
+    warn "Removing Longhorn manager pre-pull sidecar because it is in image/runtime failure."
+    remove_longhorn_manager_prepull_sidecar
+    return 0
+  fi
+  log "longhorn-manager-prepull-sidecar=retained harbor=healthy pull_failures=false"
+}
+
 save_longhorn_unlock_optional_replica_snapshot() {
   if [[ "${EXECUTE}" -eq 0 ]]; then
     log "DRY-RUN: save optional workload snapshot to ${LONGHORN_UNLOCK_REPLICA_SNAPSHOT_FILE}"
@@ -2984,7 +3009,7 @@ longhorn_unlock_flow() {
   REFRESH_BOOTSTRAP_IMAGE_ALIASES=1
   freeze_longhorn_deadlock_automation
   ensure_longhorn_cache_first_policy
-  remove_longhorn_manager_prepull_sidecar
+  remove_longhorn_manager_prepull_sidecar_if_needed
   free_longhorn_instance_manager_headroom
   delete_failed_nonstorage_pods_for_headroom
   repair_longhorn_manager_cache_deadlock || warn "Surgical Longhorn manager cache repair did not complete on every affected node."