# mvirt 0.24.13

migrations: no
wire: unchanged
rev: d6fa8acf418c6cd76fa8df430c9b133184ede8a1

mvirt 0.24.13

On a platform that hosts its objects remotely (ADR-0076 remote mode),
workload and infra logs reach object storage again. The gateway never
proxied the control plane's internal requests, so every log segment
landed on a native store that does not exist in that mode and was
refused. The failing retries also stalled the leader's raft replication
every five seconds.

- mvirt-s3: an internal request with a verified token is proxied when
  its bucket is a proxied bucket of the request's project; everything
  else stays on the native path as before (#361).
- mvirt-s3-proxy: a platform entry point scoped to one project's
  buckets (#361).
- cplane: log segments are encoded off the async runtime, and a failing
  buffer backs off from 5 s to 5 min instead of retrying every tick
  (#361).


## 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.