Skip to content

v0.50.2 — Roster

Choose a tag to compare

@agit8or1 agit8or1 released this 14 Nov 23:58
01ed077

The "Upgrade to full agent" button says where the upgrade is, and running a reinstall link clears a queued reinstall.

DEWEY-WKS2022 showed Upgrade requested — waiting for check-in while the Devices list showed it online three minutes
earlier. Its audit trail:

When (UTC) What happened
2026-11-13 05:09 The upgrade to the signed agent was requested
05:11 It was handed out and downloaded
Nov 13 → Nov 14 The computer went silent
2026-11-14 21:49 Reinstall agent (at next check-in) was queued
23:17:13 The 0.48.2 reinstall link was run on the machine (ENDPOINT_CLAIMED)
23:17:21 First heartbeat: the migration was handed out again, and so was the queued reinstall
23:20:04 AGENT_MIGRATION_COMPLETED

Two things were wrong:

  • The button's label never changed. It printed one constant, UPGRADE_PENDING_LABEL, for every open migration
    state, so a request that was handed out, downloaded or not confirmed (handed out more than two hours ago with no
    signed agent reporting) still read "waiting for check-in". The server's state text was right, but only in the
    tooltip and the header line. Now the button reads:

    • waiting for check-in (pending);
    • on hold (held);
    • handed out — installing (dispatched);
    • downloaded — waiting for the full agent to report (downloaded);
    • not confirmed — check the computer (not_confirmed).

    UpgradeView carries the state as step, and the button carries it as data-step. The pending wording is
    unchanged, which e2e-ui.mjs checks.

  • A queued reinstall outlived the reinstall. claimEndpoint now clears reinstallRequestedAt, reinstallRequestedBy
    and reinstallDispatchedAt when it rebinds the node. A claim is the installer having just run, so the request is done.
    Before, the fresh agent was handed the reinstall on its first heartbeats and ran the installer again. It would stop
    only once a heartbeat reported this server's agent version or a signed update check after the hand-out. Clearing a
    request only narrows what happens. The claim is still audited as ENDPOINT_CLAIMED.