# mvirt 0.24.2

migrations: no
wire: unchanged
rev: e153d330f92a6bf4fe4bdb9508b5f157fcd650e7

mvirt 0.24.2

A patch release with two fixes on top of 0.24.1 (which carried #340: a raft
leadership change no longer restarts the control plane on two hosts).

What changed

- #356: `GET /v1/logs?resource=<pod>&source=container` — the documented
  pod-wide query — returned nothing for a pod whose container was printing,
  because the pod -> containers widening existed only on the tenant feed.
  Every path that reads `resource` now widens a pod to its containers: the
  platform feed, the audit feed, the tenant feed, `/v1/logs/stats` and the
  SQL query endpoint.
- #342: a rebooted node's clock is now stepped, not slewed. The nixpkgs
  default `initstepslew` could never resolve its hostnames before DNS came
  up; the node image now sets `makestep 1 3`, and mvirt-cplane waits (at
  most 60 s, never blocking) for chrony before it starts, so a replica that
  may become leader does not stamp wall time 20 s off.

Upgrade: ordinary host-by-host roll (release.md §C1). No store migration.


## 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 <control-plane host>
nix run .#colmena -- apply boot --reboot --on <node>   # 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.