# mvirt 0.24.4

migrations: no
wire: unchanged
rev: 761da86d743316a2780af96d332bc8d0b0af3ca5

mvirt 0.24.4

A patch release: container output keeps reaching the log feed after its
pod-VM is re-created.

What changed

- #355: after a host reboot, a reschedule or a Task retry, a pod-VM is
  re-created under the same container ids and its capture restarts at
  offset 0. Container lines carry no timestamp, so the cplane's ingest dedup
  compared offsets alone and dropped every new line as a re-delivery until
  the new capture outgrew the old one. mvirt-vmm now stamps each streamed
  record with the pod-VM instance it came from, the node forwards it, and
  the cplane dedups per instance. Additive proto fields; an old vmm or node
  behaves as before.

Also carries everything in 0.24.3 (#358 report consumer owned by the level
pass, #322 seeds once per leadership term, #326 outbound HTTP error causes,

Upgrade: ordinary host-by-host roll (release.md §C1 + §C3). 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.