Skip to content

Upgrading from 0.5: keep what an installation has, and say what is older - #59

Merged
helightdev merged 13 commits into
reworkfrom
rework-upgrade
Nov 15, 2026
Merged

helightdev merged 13 commits into
reworkfrom
rework-upgrade

Conversation

@helightdev

Copy link
Copy Markdown
Owner

Fixes what a trial upgrade of a 0.5.2 installation turned up.

  • The server recipe updates the container in place and keeps its options (Docker socket, groups, extra hosts); upgrade-server carries log options.
  • A node taken in keeps its SSH limits and root's key login.
  • The client refuses a server of an older API before anything runs, and messages say which side is older: client, server or controller.
  • A kept token is never lost, and berth-remote's settings are taken in on every command.
  • berth serve --web is accepted with a warning, / answers with a pointer, and refusals exit 3.
  • Hosts of 0.5 that berth never joined to the tailnet keep applying without it.
  • A store of an older berth is copied before it is converted.
  • New page "Upgrading from 0.5", and an end-to-end case that opens a 0.5.2 store.

The merge of rework brings the stricter take-over rule for provisioners, so the upgrade page and the install skill now name the adopt action for the ones an apply leaves alone.

Docker puts the daemon's log options on every container, so refusing them refused most real servers. The helper also takes options to set over the container's, for the server recipe.
…gs on every command

The import ran in berth ui alone and removed berth-remote's file even when berth used another server. Now it runs before any command looks for a server, and leaves the file unless berth has its contents. The screens forget the kept token only on sign-out or for one typed in its place; a server that does not answer, or refuses the token, leaves it.
…ntroller is behind

A client refuses a server with an older API for anything but berth version, and the screens say the same on connect. A controller too old to list its actions is said to be old, with the command that updates it, on the host's page, its actions and the overview.
… before anything runs

berth serve takes --web again and says it does nothing, GET / says to use berth login, and a missing tailnet key, secret or ISO is a refusal (412) rather than a bad request.
…p applying without it

berth 0.5 joined a host only with a pasted key or with its switch for keys berth makes, which was off unless set. The tailnet's import marks the other profiles with devbox_tailnet_optional: they join once berth can make the key, and until then an apply goes on without the tailnet.
berth 0.5 cannot read a store once this berth has converted it, so the first open copies a SQLite store beside it with VACUUM INTO, named for the berth and schema it had. A store in PostgreSQL gets a line saying a dump is the way back. The fixture is a store berth 0.5.2 wrote, with made-up hosts and fake keys.
The order of the updates, what is converted, berth-remote's settings, the tailnet and going back, with the old spellings generated from the command line's table. Nodes say what taking in a server keeps, and updates no longer says sessions need berthd API 2.
@helightdev
helightdev merged commit ef934c1 into rework Nov 15, 2026
18 checks passed
@helightdev
helightdev deleted the rework-upgrade branch November 15, 2026 19:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant