Test fixtures exhaust Docker's default address pools under a full-suite run

[genwave-radio]

🐳 What

A full dotnet test GenWave.sln (or the whole Host suite alone) intermittently fails 1–23 facts with:

docker compose -p genwave-t372b-<hash> -f tests/GenWave.MediaLibrary.Tests/db-compose.yaml up -d --wait failed:
Network genwave-t372b-<hash>_default Creating … Error response from daemon: all predefined address pools have been fully subnetted

Seen 2026-08-30 on the dev box (thor) at the end of phase G-e: solution run 23 Host reds, Host-alone 1–2, the same facts green standalone. docker network ls shows no leaked networks — only the 7 standing project networks — so this is peak concurrency, not a leak: every ephemeral fixture (EphemeralStationDatabase, DatabaseFixture) creates its own compose project + network, and xUnit runs collections in parallel.

🔢 Why it bites

Docker's default default-address-pools are 172.17.0.0/12 in /16 slices (15) + 192.168.0.0/16 in /20 slices (16) ≈ 31 networks. Seven are standing (genwave_*, mrdgenwave_*, worldcuptracker*), so ~24 concurrent fixture projects is the ceiling — a full Host run can exceed it.

🛠️ Remedies (pick one; Dean's call — daemon config is a box change)

1. /etc/docker/daemon.json: "default-address-pools": [{"base":"10.200.0.0/16","size":24}] → 256 networks (needs a daemon restart — not while the dev stack is on air).
2. Fixture-side: one shared compose network (external: true) for all fixture projects, or cap fixture parallelism via xunit.runner.json maxParallelThreads.
3. CI: already runs suites per project, so this is dev-box-only today.

Related: gh-#643 (Story240 under narrow filters — a different cause).

https://claude.ai/code/session_01AmM47neuvsJph3ZT4brTCV