Current published controller release. Existing opted-in healthy hosts use a natural all-configured-rooms-empty window. Publication, queued delivery and activation are separate facts.

This controller release adds supported maintenance announcements and draining to the existing running controller. Operators can request an empty-room drain or an explicit one-map warning, inspect progress, and cancel by operation ID. A confirmed warning and real match boundary precede any occupied one-map closure. Empty idle rooms close through their independently checked empty-room path. Late joins postpone closure; ordinary automatic updates never choose occupied maintenance.

Warning refusals keep gameplay running. Unknown warning delivery is not replayed. Uncertain closes require review instead of another close attempt. Cancellation is rechecked at the final roster lock. Already closed rooms are never silently reopened.

Provider process startup validates its working directory before admission and sends large private arguments through deadline-bound IPC after a small bootstrap. Failed children are cleaned up with pinned process identities. Positively undispatched lock contention is retryable; invalid authority, uncertain dispatched mutations and corrupt recovery remain strict failures.

Consecutive stopped controller upgrades can retire positively closed launch hints only through exact committed immutable history, including approved older workers. Closed parked aliases are accepted without rewriting ownership evidence. Archived receipt directories retain service-account readability under restrictive umasks. The updater can reuse a recognized stopped native timer launcher across releases; custom or active services are refused.

Status/support show installed and running controller versions, individual worker versions and ownership, queued targets/blockers, maintenance progress and concrete next actions. Lifetime counters and stale observations remain distinct from current faults. No credential or arbitrary remote-command surface is added.

All existing features, tuning, settings, OAuth/Discord configuration, full catalogs, enrollment, ratings, player history and explicit sharing/update/enablement choices are retained. No storage migration or dependency change is required. The v7.1.7 live startup/refresh timeout fix and v7.1.6 prompt feedback repair remain included.

Install the current ZIP asset and read MAINTENANCE_OPERATOR_GUIDE.md. Existing opted-in running hosts update naturally only when all configured rooms are safely empty. Compatible worker-only releases still update each room independently. Stopped or recovery-required hosts need their local supported recovery/start flow; publication cannot remotely start them. Older controllers must first install this release before using its new maintenance commands.

Validation includes frozen provider/administration/native regressions, real worker RPC, full-catalog gameplay/maintenance fixtures, isolated root/service-UID direct and consecutive upgrades, explicit opt-out/history preservation and timer guards. Unchanged whole-repository and donor suites are not repeated. Actual production room interruption is excluded from qualification. The older community refresh archive proves a successful replacement creation followed by failed confirmation and service teardown. Local cooldown explains a reproduced failure path, but the original raw request ledger was absent, so its exact refusal remains unproven.

Current installer and repair downloads · Hosting and operating · Operator commands · Release history