You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Orphaned agents (parent_id NULL but role=leaf/relay) are now promoted to zone leader instead of being silently stranded. When the mesh tree saturates under churn — a zone leader's stream flapping, a wave of children re-registering, a parent with no remaining capacity — RequestPeers and reassignOrphans used to clear the agent's parent_id and move on. The agent kept its non-ZL role but had no route, which meant broadcast queries (dirq cve, dirq exec WHERE …, etc.) fanned out through zone-leader streams and silently missed those orphans. Both paths now route the orphan into the same escape hatch assignRole already had: set role=zone_leader, signal the agent via a new PeerResponse.new_role field (in-band on RequestPeers) or PeerUpdate(NewRole=ZONE_LEADER, NewParentAddr="") (best-effort on the live stream from reassignOrphans), and reconnect to the server directly. assignRole preserves an existing zone_leader+NULL-parent assignment on re-register so the promotion isn't undone the next time the agent reconnects.
dirq cve and dirq errata no longer label every unreachable host as "non-RHEL". The count was computed as online_agents - assessedHosts and reported as N not assessed (non-RHEL), which conflated three different cases: actually non-RHEL hosts, RHEL hosts that timed out, and RHEL hosts the mesh couldn't reach. The line now splits the count using the agent record's os field — output reads N vulnerable, N patched[, N non-RHEL][, N RHEL did not respond], so a non-zero "did not respond" surfaces as the fleet-health signal it actually is, instead of being silently absorbed into a misleading "non-RHEL" total.
Added
DIRQ_REGISTRATION_JITTER_SECONDS agent setting (config key: registration_jitter_seconds) caps a random startup delay applied before the first Register call. Smooths thundering-herd boot scenarios: multi-VH emulation auto-picks a sensible default (N/4 seconds, clamped to 5–60s), production fleets can opt in for rack-reboot or post-maintenance fleet restarts. Zero disables jitter, preserving previous behavior. The AWS test-fleet script sets registration_jitter_seconds: 30 when DIRQ_REPLICAS_PER_VM > 1.