8000 Re-attaching a VM-backed box whose shell exited is refused instead of registering the box again · Issue #3569 · gominimal/minimal · GitHub
[go: up one dir, main page]

Skip to content

Re-attaching a VM-backed box whose shell exited is refused instead of registering the box again #3569

Closed

Description

@norrietaylor

What happens

On a VM-backed host, when a registered box's shell exits, its connection to the VM host daemon ends and the host withdraws the box's row. The next min session attach (or any command that attaches) is refused by the in-VM daemon with AttachError::BoxHostRowEnded, which prints:

the VM host daemon released this session's network addresses; run the same command again to restore them. If it is refused again, destroy the session with min session destroy <session> and start it again with min session activate

The CLI does nothing with this typed error. It does not register the box with the host again, so the user has to retry by hand, and can end up destroying and re-creating the session.

Expected

When the attach gets the row-ended refusal, the host-side CLI registers the box again through minvmd, using the same path an activation uses, and then retries the attach. The facts for the new registration come from the host side (the creator's expanded spec and the host's address allocation), never from the in-VM daemon.

Once that lands, the stop-gap wording above and the refusal path that stands in for it can be removed.

Note: the spec for box networking (Spec 18) needs a clarification first: a re-registration by the creator after an attachment ends counts as a new attachment, so the "registered before the box's first connection" requirement is met for that attachment.

Where

  • crates/minimal-client/src/box_registration.rs: register_box_for_activation, the registration path to reuse.
  • crates/minimald/src/box_actor/attach.rs: AttachError::BoxHostRowEnded and its Display text.
  • crates/minimald/src/box_actor/host.rs: host_row_lost, and the returns of AttachError::BoxHostRowEnded.
  • The CLI attach path in crates/minimal (no handling of this error today).

Workaround

Run the attach again. If it is refused again, min session destroy <session> and min session activate.

epic: #1437


This is a residual of epic #1437 (NET).

Activity

  1. added
    bugSomething isn't working
    priority:should-haveShould be done. Applied by the inbox-route classifier or a human.
    on Nov 18, 2026
  2. norrietaylor commented on Nov 20, 2026

    @norrietaylor
    MemberAuthor

    Fixed by #3615, merged 2026-11-20: an attach refused with BoxHostRowEnded registers the box again through minvmd and retries the attach.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingpriority:should-haveShould be done. Applied by the inbox-route classifier or a human.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions

      0