fix(installer): report host RAM consistently + achievable memory advice (#417)

[shujaatTracebloc]

What & why

The Windows preflight memory check preferred Docker's WSL2 VM budget over physical RAM, so the same 15 GB laptop reported Memory: 7 GB with Docker up and Memory: 15 GB ✓ with it down — flip-flopping across re-runs mid-install — and told a 15 GB machine to "give Docker >= 16 GB", physically impossible advice. Seen on three machines (7/6/15 GB readings; the 16-on-15 hint).

Changes (scripts/install-k8s.ps1)

• Get-PfMemGb now returns HOST RAM only (physical, via CIM) — identical whether Docker is up or down. The container runtime's smaller VM budget is read separately (Get-PfRuntimeMemGb) and shown as its own labeled line: Memory: 15 GB (Docker's current share: 7 GB).
• Get-PfMemRecommendation caps every suggestion at (host − 2 GB) — never more than the machine physically has; floors at 1 GB.
• Step-1 (Test-Preflight) and Step-2 (Test-PreflightRuntimeMem) now give one consistent, host-aware message, and Step-2's recommendation is capped too.

Acceptance (from #417)

• Always two labeled lines: host RAM + Docker's current budget (when a runtime is up)
• Cap every recommendation at (physical − 2 GB); never recommend more than the machine has
• De-duplicate the Step-1/Step-2 warnings into one consistent message
• Same machine, Docker up or down, reports identical host RAM; recommendations always achievable

Tests

• Get-PfMemRecommendation: cap at host−2, returns desired when it fits, floors at 1, the 16-on-15→13 case.
• Get-PfMemGb reports host RAM regardless of the Docker budget (Windows CIM-mocked =16; cross-platform decoupling from Get-PfRuntimeMemGb).
• Test-PreflightRuntimeMem caps its recommendation at host RAM (4 GB on a 6 GB host, never 8).
• Updated the former "Get-PfMemGb prefers docker" test — it asserted the flip-flop bug.

Local verification

Pester 261/0/9 · check-style pass · drift clean · manifest regenerated · PS parse OK. (No bash changes — bats unaffected.)

Closes#417

---

Note

Low Risk
Warn-only preflight UX and installer messaging in install-k8s.ps1; no auth, cluster, or data-path changes.

Overview
Fixes Windows preflight memory reporting where host RAM flipped between Docker up/down and advice could exceed physical RAM (e.g. “give Docker 16 GB” on a 15 GB machine).

Get-PfMemGb now always returns physical host RAM (CIM only); Docker’s VM budget stays on Get-PfRuntimeMemGb. Show-MemoryStatus centralizes warn/ok text: the label is always host GB (plus optional “Docker’s current share”), while pass/fail uses the effective memory (Docker budget when known, else host) so a throttled Docker cap is never green-OK on a large host. Get-PfMemRecommendation caps hints at host − 2 GB (min 1 GB); when host RAM is unknown, hints use raw targets instead of capping at the current budget (#483).

Test-Preflight and Test-PreflightRuntimeMem both call Show-MemoryStatus so Step 1 and post-Docker checks share the same wording. Pester coverage and manifest.sha256 updated for install-k8s.ps1.

^{Reviewed by Cursor Bugbot for commit dcc5f8e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

👋 Heads-up — Code review queue is at 31 / 30

Above the WIP limit. The team convention is to review existing PRs before opening new work.

Open PRs currently in Code review (oldest first):

• averaging-service#222 — fix(sklearn): unwrap Pipeline to terminal estimator before merge dispatch (Prod: Set kubeconfig context namespace after Helm install for tracebloc  #209) · author: @divyasinghds · no reviewer assigned
• averaging-service#225 — fix(averaging): fail fast + dead-letter on permission-denied weight write (release: promote develop → main (installer hardening — OOM gate, apt/needrestart fixes, drift CI) #221) · author: @divyasinghds · no reviewer assigned
• averaging-service#236 — feat(blob_store): vendor the blob_store seam + settings (#1162) · author: @saadqbal · no reviewer assigned
• backend#1201 — docs(rfc): Data spaces & federated dataset composition (draft) · author: @LukasWodka · reviewer: @saadqbal
• backend#1217 — docs(rfc): Dataset integrity & score reproducibility (draft) · author: @LukasWodka · reviewer: @saadqbal
• backend#1261 — RFC-BACKEND-0008: board & process reset (decision RFC — for markup) · author: @LukasWodka · reviewer: @saadqbal
• backend#1284 — docs(deploy): backend deploy runbook skeleton (#1263) · author: @LukasWodka · no reviewer assigned
• backend#1293 — fix(heartbeat): tolerate unknown experiment_key in training_info (#1287) · author: @divyasinghds · no reviewer assigned
• backend#1332 — feat(check-model): allow KNN/GaussianProcess on a single-edge owned dataset (#923 follow-up) · author: @divyasinghds · no reviewer assigned
• backend#1339 — feat(blob_store): migrate remaining trivial FS exists/delete sites (#1162) · author: @saadqbal · no reviewer assigned

Pull from review before opening new work. (This is a nudge from the kanban WIP check, not a block.)

[saadqbal]

Good fix for the flip-flop itself — the reported number is stable now and the cap logic is right. But I think Step-1 gave up too much: by grading host RAM only, a throttled Docker budget now gets a green tick where develop warned, including budgets below the 5 GB floor. Details inline at 3065 — I ran the ladder on both branches to confirm rather than eyeball it.

Two parity things I'd rather not lose, not blockers:

• Get-PfCpu (2974) still has the exact "prefer runtime, fall back to host" line you deleted from Get-PfMemGb, so the CPU row still flip-flops between Docker up and down. Same bug, same machine, and there's no (Docker's current share: N) label on that row to soften it.
• preflight.sh is untouched, so on macOS/Linux #417 still reproduces as filed: _pf_total_mem_kb (:157) prefers the runtime view, and :248 still says "raise to >= 8 GB (16 GB to train)" with no host cap. Closing #417 on the Windows fix alone leaves the reported repro live on the other platforms — worth either porting the policy or narrowing the issue to Windows.

Also worth a note: check-drift.sh gates ps1<->sh parity for API hosts and download hosts but nothing for preflight resource policy, which is why "drift clean" passed on a change that split the two installers' memory semantics.

Pester 264/0/9 green locally and the manifest hash matches the new file.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 2b2cb5a. Configure here.}