Stop clearing the wake event at sleep entry, keep force-sync skip until consumed, and make pause loops wait on the resume event so UI start kicks topology immediately.
Co-authored-by: Cursor <cursoragent@cursor.com>
Use savepoint upserts with payload dedupe, finalize jobs on a fresh sibling session, serialize concurrent topology syncs, and reap stale running rows so the scheduler does not keep spawning orphans.
Co-authored-by: Cursor <cursoragent@cursor.com>
Phase-1 docks both topology APIs without writing Fabric; expose a 24h topology_auto_sync task and domain=topology manual sync for later LLDP alignment.
Co-authored-by: Cursor <cursoragent@cursor.com>
Queue fabric edge deletes until Save with undo-safe pending state; hard-delete topology/LLDP placeholders from the canvas; purge fully orphaned fabric nodes after inventory detach, with a periodic reconcile GC.
Co-authored-by: Cursor <cursoragent@cursor.com>
API /metrics and /health/ready now read a worker heartbeat when collectors are split out; WebCRT blocking I/O uses a dedicated executor so session pumps do not starve the default pool.
Co-authored-by: Cursor <cursoragent@cursor.com>
Extract channel/session modules and CLI guard, move UME sidebands out of main, and default NETX_RUN_INLINE_SCHEDULERS off so ops run python -m netx_api.worker.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add capability RBAC, Alembic bootstrap, optional worker schedulers, and close public docs by default so lab installs stay usable without shared signing keys.
Co-authored-by: Cursor <cursoragent@cursor.com>
