Barn 0.7.0: simpler test labs and automatic recovery

Barn 0.7.0 streamlines first-run startup and recovery for local test labs: re-running barn up finishes interrupted work, retries guest bootstrap stages, and updates guest helper scripts in place without restarting running VMs.

What changed

• Shorter first run. Interactive up can generate the default inventory, install missing host tools, and bring an intact inactive Barn network back online. Fast checks stay quiet while longer tasks display progress and a compact summary.
• Independent guest stages. Data disks, host shares, guest hostnames, node-to-node SSH, and private networking initialize independently. If an optional stage fails, Barn reports the specific limitation while keeping management SSH available, and subsequent up runs skip stages that are already healthy.
• Automatic test-disk recovery. Healthy filesystems are mounted as-is; unrecognized or confirmed-corrupt filesystems on test data disks are reformatted and remounted with an explicit data-loss notice. Transient probe failures, missing devices, busy mounts, and host I/O errors never trigger formatting.
• Fewer manual fixes. Unwritable host shares fall back to read-only and recover on the next up once permissions are fixed. Occupied automatic SSH forward ports are reassigned for stopped guests. Interrupted image downloads resume automatically and fail over across official repositories while enforcing SHA-256 digest verification.
• Clean output and backward-compatible state. Guest warnings are stored in a disposable cache outside the schema-2 VM documents and node directories, so older Barn binaries can still inspect and manage the deployment.

Upgrade notes

• Re-running up performs guest recovery (there is no separate repair command), whereas start powers on existing nodes without applying inventory changes. Passing --no-wait skips readiness and recovery checks. See the storage guide for data-disk recovery rules.
• A guest with working management SSH and optional stage warnings returns exit code 0. Automation that requires every optional feature should inspect nodes[].warnings and nodes[].repairs in --json or --yaml output.
• If an earlier 0.6.0 bootstrap failed after deleting the staged control-node private key, up restores management SSH and reports a control-ssh warning; run barn plan and barn recreate on the control node if it needs peer-to-peer SSH.

Install or upgrade

curl -fLO https://github.com/pgsty/barn/releases/download/v0.7.0/install.sh
chmod +x install.sh
BARN_VERSION=0.7.0 ./install.sh
barn version
barn up

The release includes macOS and Linux (amd64/arm64) archives, Linux DEB/RPM packages, the installer script, checksums, SBOMs, and a Homebrew formula.

Validation

Release commit 9c6d4896d93733d1cb60a7e5d8591e9a06659c9d passed source CI, the packaging snapshot, and the tag workflow, producing 20 verified release assets. Public macOS arm64 and Linux amd64 installer runs verified binary delivery for 0.7.0, and macOS validation covered CLI smoke checks and cross-compilation.

On Ubuntu 26.04 amd64 with KVM/QEMU 10.2.1 and Ubuntu 24.04 guests, recovery tests exercised ext4/XFS corruption, retained persistent disks, failed probes, busy mounts, read-only share fallback, and repeated healthy up runs without restarting the QEMU process:

• An interrupted 0.6.0 bootstrap (stopped at the management-egress probe) resumed under 0.7.0 in 3.3 s while preserving its data disk and UUID, and remained manageable by the 0.6.0 binary (status, stop, start, destroy).
• A deliberately corrupted disposable ext4 data disk was reformatted and remounted in 3.2 s with an explicit data-loss notice while keeping the VM process running.