Documentation · Releases · Medusa

Medusa

v0.7.12 — Medusa

What landed

• N independently-scored chainweb containers per host — each its own nodes row, its own ServerScore, its own slice CPU/RAM/drive.
• Slice-aware ServerScore math: drive divisor is per-slice, not host-wide. Honest by construction.
• Cgroups-isolated CPU/RAM/IO bands so noisy neighbours can't starve the slice.
• Install gates: ≥ 3 vCPU, ≥ 6 GB RAM, SSD-only data drive per segregated container.
• Per-drive commitment cap — sum of committed_gb across containers stays under capacity minus 5% safety margin.
• Compose spec hash for install reproducibility — identical inputs produce identical containers.
• Recommended chainweb-port allocator avoids collisions across containers on the same host.
• 104 m-series patches polishing tunneling, install wizard, fleet counters, and admin observability.

Operator notes

A single host can now run N independently-scored chainweb containers sharing the host's drives + network + frps role. Each container is its own nodes row with its own slice CPU/RAM, its own data drive, its own ServerScore.

The four operator rules (operator-stated, codified):

1. DNS gate — every server hosting chainweb must have a stable DNS hostname. The install endpoint refuses without one.
2. Full-host minimums — chainweb in normal mode requires ≥ 2 physical cores, ≥ 8 GB RAM, an SSD where the data sits.
3. Segregated minimums — each container in cluster mode reserves ≥ 3 vCPU + ≥ 6 GB RAM. SSD required for the data drive.
4. Per-drive commitment cap — sum of committed_gb across every chainweb on a drive ≤ drive capacity − 5% safety margin.

Audit reference

The slice-aware scoring model, drive-divisor honesty proof, and the 148/148 test sign-off are published in the Medusa audit →

Patch log