• v1.17.7 81c4ff4e33

    Morphit v1.17.7
    All checks were successful
    morphit-ci / Supply-chain audit gate (fail on new high/critical) (push) Successful in 33s
    morphit-ci / TypeScript typecheck (sweep all workspaces) (push) Successful in 1m0s
    morphit-ci / apps/web svelte-check (svelte-kit sync + svelte-aware tsc) (push) Successful in 53s
    morphit-ci / Integration tests (real Postgres 16) (push) Successful in 1m56s
    morphit-ci / ansible-lint (playbook quality gate) (push) Successful in 1m0s
    morphit-ci / Smoke suite (run-smokes.sh, triple-pulse) (push) Successful in 31m10s
    morphit-release / Build + publish release tarball (push) Successful in 33m6s
    Stable

    agorise released this 2026-09-13 22:21:49 +00:00 | 6 commits to main since this release

    Signed by agorise
    GPG key ID: 53524E1F1017EB9C

    Morphit v1.17.7

    Finishes what v1.17.6 started: the snapshot publisher now works when run on a
    schedule, and operator alerts stop repeating themselves. No protocol or consensus
    change.

    Fixed

    • The snapshot publisher works on a timer. It talks to IPFS by switching to
      the IPFS account first — and on a hardened server, both of the standard tools
      for doing that are blocked: one is refused outright by the service's security
      settings, the other is not permitted to change accounts at all. The job is
      already running with full privileges and does not need to switch accounts: the
      IPFS command line only reads a small file and then talks to the running service
      over a local connection. It now tries talking directly first, falls back to
      each account-switching tool in turn, and uses whichever actually works —
      reporting which one it chose. Guessing which tool a given server permits is
      what made this fail repeatedly.
    • Alerts stop repeating. A unit that stayed broken produced a CRITICAL alert
      every five minutes, over and over, for the same unchanged fact. Alerts now fire
      when something changes: once when a unit fails, once when it recovers, and
      again if it fails afresh. A repeat failure is never swallowed, and if the
      bookkeeping cannot be written the behaviour falls back to alerting every scan
      — never to silence.

    Added

    • The alert monitor is now genuinely run during testing, driven through a
      whole failure → recovery → failure-again cycle, because "alerts once and then
      stays quiet" is a property you can only observe by running it repeatedly.
    • The publish test now reproduces a server where both account-switching tools
      are blocked
      , which is the exact situation that broke the live publisher.

    Notes

    • No protocol/consensus change. Everything here is operator-facing.
    • If you disabled the publish timer to stop the alerts, re-enable it after
      upgrading: sudo systemctl enable --now morphit-snapshot-publish.timer.
    • OPERATIONS.md §52 covers the snapshot pipeline and how to check it.
    Downloads