[Windows] Docker VMM guest never boots: silent freeze upstream of init, 0-byte sailor pool file, no VmCompute events (DD 4.94.0, Ryzen 3700X, Win 11 26200) #740

Environment

• Docker Desktop: 4.94.0 (241994) — stable channel, confirmed latest by docker desktop update ("already the latest version")
• Engine: 29.8.2 Linux; all-users (exe) install on C:\Program Files\Docker\Docker
• Host: Windows 11, build 10.0.26200 (x64), AMD Ryzen 7 3700X
• Virtualization stack: Windows Hypervisor Platform + VirtualMachinePlatform + Microsoft-Hyper-V-All all Enabled (verified elevated), hypervisorlaunchtype=Auto, BIOS virtualization on
• Live proof the hypervisor works on this host: the classic Hyper-V LinuxKit backend (same DD install) is running normally right now with 29 containers — so WHP is enabled and a hypervisor is running while the VMM attempts fail

What happens (reproducible across 6 attempts over two days)

1. Switch to Docker VMM via settings-store.json ("UseLibkrun": true; also tested with the dotted key "vm.hypervisor.windows": "docker-vmm" present — the dotted key alone with UseLibkrun:false routes to Hyper-V and is not honored, so UseLibkrun is the effective router on 4.94.0)
2. Start Docker Desktop → log shows engine linux/sailor hypervisor process up after 7-11ms and VM started
3. Guest never reaches Linux init: %LOCALAPPDATA%\Docker\vm-data\00000002.000007cf (sailor vsock pool file) is created but stays 0 bytes for the boot's entire life; %LOCALAPPDATA%\Docker\log\vm\init.log is not written for the new boot (no guest-side output at all)
4. Backend loops forever: dockerd HEAD /_ping … context deadline exceeded (1.0004s per attempt, back-to-back), engine state starting, docker daemon not ready 5m30s after start, still waiting
5. com.docker.sailor.exe stays alive with near-zero CPU (~0.09–0.25 s over minutes)
6. Windows Event Log Hyper-V-Worker / VmCompute: no errors at all — the failure is invisible host-side

One earlier attempt before updating to 4.94 did briefly serve real dockerd traffic (~90 s of exec/stats/info the backend had claimed — though server-side attribution was uncertain) before freezing the same way. All other boots were silent from birth.

The exact dialog DD's checker would show here, if it fired

Only three VMM-start-failure templates exist in the 4.94.0 backend binary:

1. generic "Docker VMM could not start… collect diagnostics and submit an issue" ← this is our case
2. "Windows Hypervisor Platform is unavailable" — eliminated (feature enabled + in-use)
3. "WHP installed but no hypervisor is running" — eliminated (Hyper-V engine boots and serves traffic concurrently)

We attempted the flip via settings-store edits rather than the UI, so no modal was displayed; the generic template applies.

Already tried and ineffective

• Fresh pool files (delete 00000002.* before every boot)
• memoryMiB: 4096 override (default tier >4 GB required for VMM) → same silent freeze
• Coherent settings (WslEngineEnabled:false, memory/disk/cpu consistent), SettingsVersion:46
• Relaunch cycles with clean stops (/ forced kills after a wedged stop), service restarted, empty vm-data recreate
• Full docker desktop update re-check (4.94.0 is already the latest version)

Diagnostics

Uploaded with the hyper-v engine running (includes the 6 failed-boot histories from this machine):

Diagnostics ID: 5DC45752-61D3-4672-978D-04B08FCC7D0B/20261012043929

Key files in the bundle: Docker\log\host\com.docker.backend.exe.log* (sailor boots at 2026-10-12T00:38Z, 02:28Z, 02:35Z, plus tonight's), Docker\log\vm\ (empty-of-new-content rotations), vm-data timestamps.

Ask

Is Docker VMM on Windows known-dead on AM4/Ryzen 3000 + build 26200? The guest dying upstream of init.log with a 0-byte pool file, zero vmcompute events, and no checker verdict looks like a backend/sailor bug, not a host-misconfiguration. Any known fix ETA would save us further local attempts (4.95 / GA was announced for late Oct 2026). Happy to test a dev build or provide the local hyper-v backup receipts on request.