v1.22.51
Class-connectivity queries can no longer cache a zero that was really an outage.
_aggregate_class_connectivity used to return [] when a Neo4j or Solr call it depends on failed, and that became {'count': 0} — stored by the Solr result cache and pinned for a month at the v3-cached edge, which is how gamma Kenyon cell downstream connectivity showed "There is no data to display" on 2026-09-04 with 3,886 rows in the backend.
- Backend failures in the class-connectivity family now raise
ConnectivityBackendError; ha_api answers 5xx and no cache layer stores it. True zeros are unchanged. - A result computed with some instances' connectivity missing is tagged
partial: served, never stored in the Solr result cache, kept at the edge for 600 s (VFBQUERY_PARTIAL_EDGE_TTL). /run_querynow sendsX-Accel-Expires: 0+Cache-Control: no-storeon any empty result, so the nginx edge never stores an empty body. No owl_cache change needed.