Release v2.3.0 · Holo795/CBMCoolifyBackup

CBM 2.3.0 adds two-factor authentication, agent settings managed from CBM, and smarter handling of disk space on the agent hosts.

Upgrade notes

Update both the controller and the agents (re-run the install command on each host, or pull ghcr.io/holo795/cbm-agent:2.3.0). Migrations 0029 and 0030 run automatically when the controller starts (new TwoFactor table, new columns on users, agents and settings).
Agents still on 2.2.x keep working with a 2.3.0 controller, but ignore the settings set in CBM and keep the old disk behaviour until they're updated.
Settings already set by environment variable on a host (AGENT_CONCURRENCY, AGENT_MIN_FREE_MB, LOG_LEVEL) keep winning: they show as locked in CBM. Remove them from the agent container to manage them from CBM instead.
Nothing changes for anyone until they turn two-factor on, or an admin requires it.
Restoring CBM from a recovery file turns two-factor off for every account (its secrets are tied to the old install's BETTER_AUTH_SECRET); people set it up again from Profile, and accounts the policy covers are asked at sign-in.

Authenticator app + backup codes. From Profile → Two-factor authentication: scan a QR code (or type the key) in 1Password, Bitwarden, Google Authenticator, Aegis…, confirm with a first code, and keep the 10 single-use backup codes. New backup codes or turning it off are on the same card.
Asked after every kind of sign-in — password, and also GitHub / Google / GitLab: no session exists until the code is checked, so a compromised provider account isn't enough. Trust this device for 30 days is available; wrong codes lock the account for a few minutes and a code can't be reused.
Admin policy. Settings → Two-factor authentication: optional, required for admins, or required for everyone. A covered account that hasn't set it up is held on a setup page until it does, and can't turn it off. The section shows how many accounts don't use it yet.
Admin reset. On Users, members with two-factor carry a 2FA badge; … → Reset two-factor helps someone who lost their phone (their app, backup codes, trusted devices and sessions stop working).
Accounts that only sign in with a social provider can set it up without a password. API tokens (MCP) and agents are not affected.

Agents → Default settings sets, for every agent: jobs at once, free space kept on the host, copy mode and log level. The gear on an agent's row overrides them for that host only; an empty field keeps the default.
Changes apply on the agent's next heartbeat — no reinstall, no restart.
A value set by environment variable on the host still wins; the field is then locked and shows the host's value. The agents list shows each host's settings in effect and flags custom ones.

Before copying a volume or host folder, the agent now measures it and compares it with the free space on its work dir, before freezing anything.
Copy mode for tar destinations (new, per agent or by default; AGENT_STAGING_MODE): Auto copies through the host when it fits and otherwise sends the volume straight to the destination; Local always copies through the host and fails cleanly, with the sizes, when it doesn't fit; Direct always sends straight to the destination. A direct send keeps the containers frozen until the upload ends, and the backup log says so.
Encrypted backups no longer need twice the room: the copy is checked, hashed and encrypted in a single pass.
The restic engine always needs the local copy: with too little room it now fails cleanly with the sizes instead of filling the disk.
A failed backup no longer leaves its partial copy in the work dir, and a failed copy no longer leaves a helper container running.

Social sign-up from an invitation. The invitation page now shows Continue with … for each configured provider; an invitee could only sign up with a password before.

Published to the GitHub Container Registry — also tagged :2.3 and :latest:

docker pull ghcr.io/holo795/cbm-controller:2.3.0
docker pull ghcr.io/holo795/cbm-agent:2.3.0
docker pull ghcr.io/holo795/cbm-mcp:2.3.0