Why
Hospital/enterprise reality: the researcher gets a temporary admin window (or an IT admin runs the install), and elevation is revoked afterwards. Today Docker Desktop only auto-adds the installing user to docker-users, autostart is a printed Hint, and the WSL memory budget is a printed Hint — so the researcher's standard account can end up with a broken Docker, a cluster that doesn't come back after reboot, and a 50%-of-RAM VM that OOMs training. We hand-fix this per site today (net localgroup docker-users <user> /add, autostart, .wslconfig); the installer runs elevated and should just do it.
Where
scripts/install-k8s.ps1 (v1.9.6): autostart hint L2138, memory hints L2351–2356; no docker-users handling exists. Related: the Linux prepare-host admin step (#381) and its CLI wrapper (tracebloc/cli#394) — this is the Windows peer, see tracebloc/backend#1178 / #1179.
Scope (all during the elevated run, with consent)
- Ask for / accept the daily username when the elevated user differs (parameter + prompt);
net localgroup docker-users <user> /add.
- Enable Docker Desktop autostart for that user ("start when you sign in") — settings.json
autoStart or the per-user Run key.
- Write
%UserProfile%\.wslconfig [wsl2] memory=<sized from the #393 training budget, capped at physical − 4 GB> for the daily user + orchestrate wsl --shutdown and a Docker restart.
- Print a "configured for " summary so IT can review what changed.
Acceptance criteria
- Standard (non-admin) user signs in the next day: Docker starts, cluster auto-recovers, Docker budget is training-capable — zero admin actions needed.
Why
Hospital/enterprise reality: the researcher gets a temporary admin window (or an IT admin runs the install), and elevation is revoked afterwards. Today Docker Desktop only auto-adds the installing user to
docker-users, autostart is a printed Hint, and the WSL memory budget is a printed Hint — so the researcher's standard account can end up with a broken Docker, a cluster that doesn't come back after reboot, and a 50%-of-RAM VM that OOMs training. We hand-fix this per site today (net localgroup docker-users <user> /add, autostart,.wslconfig); the installer runs elevated and should just do it.Where
scripts/install-k8s.ps1(v1.9.6): autostart hint L2138, memory hints L2351–2356; no docker-users handling exists. Related: the Linuxprepare-hostadmin step (#381) and its CLI wrapper (tracebloc/cli#394) — this is the Windows peer, see tracebloc/backend#1178 / #1179.Scope (all during the elevated run, with consent)
net localgroup docker-users <user> /add.autoStartor the per-user Run key.%UserProfile%\.wslconfig[wsl2] memory=<sized from the #393 training budget, capped at physical − 4 GB>for the daily user + orchestratewsl --shutdownand a Docker restart.Acceptance criteria