# mvirt 0.14.14 migrations: no wire: unchanged rev: 26ea0f484ffb14bf1d20d2a6b00fbe4d43a94992 mvirt 0.14.14 — a leader that lost its controller reconciled nothing, forever A 1.2 s quorum-freshness flicker aborted the one reconcile Controller a cplane process would ever have; when freshness returned 300 ms later, the replica kept serving as leader — REST answering, nodes Online, raft healthy — while reconciling nothing, indefinitely. On the production fleet this held for 2 d 17 h until a VM created in the console sat in Pending with no node. Changed: - cplane: the Controller is rebuilt on every leadership gain, like every other leader loop — taking leadership and running the reconcilers are one transition (#217). The tunnel-connect receiver lives in a shared slot each controller generation locks for its lifetime; deposition releases it, and a successor parks until its predecessor is gone, so two generations can never converge concurrently. The parser-binding table is shared across generations. - Regression test replays the incident: an aborted generation's successor must converge a VM created only after the handover. ## Upgrading Point your flake's `mvirt` input at this release and redeploy the control-plane host first, then each node in turn: ``` nix run .#colmena -- apply boot --reboot --on nix run .#colmena -- apply boot --reboot --on # one at a time ``` This release adds no migrations, so the control-plane host rolls back the same way a node does: boot the previous NixOS generation.