feat(#30): dataset rm — in-cluster teardown (table + PVC) [DRAFT]

[aptracebloc]

Draft — pending team-lead review of the teardown architecture. Implements #30 (the CLI-direct approach); the central-catalog piece is split out as #39.

Adds tracebloc dataset rm <table> — the in-cluster teardown of a previously-pushed dataset, packaging the manual kubectl exec cleanup we used during v0.1 testing.

What it does

1. Discovers the cluster (parent release + shared PVC) — same as push.
2. Shows the teardown plan (the MySQL table + both PVC dirs) + a destructive-action warning.
3. Confirms — reuses the feat(dataset push): interactive guided mode #28 prompter on a TTY; --yes skips; --dry-run previews and stops; refuses (exit 3) off a terminal without --yes.
4. DROPs the table (exec into the mysql pod, using its own $MYSQL_ROOT_PASSWORD so no DB credential transits the CLI) and **rm -rf**s the PVC dirs (exec into the jobs-manager pod, which mounts the shared PVC).

Exit codes mirror push: 2 bad name, 3 kubeconfig/refused, 4 no release/PVC, 7 teardown failed mid-flight.

⚠️ The open question (why this is a draft)

The teardown mechanism is exec-into-existing-pods (the "CLI-direct" path). The alternative is a server-side jobs-manager delete-ingestion endpoint that could also remove the backend catalog entry in one place. See the DESIGN NOTE on push.Teardown. This needs your call before it ships.

It also does not remove the central backend catalog entry — the CLI has no direct line to that backend, so a successfully-ingested dataset torn down this way leaves a stale catalog entry. Tracked as #39.

Assumptions worth confirming: a pod whose name contains mysql exposes $MYSQL_ROOT_PASSWORD, and the jobs-manager pod mounts the shared PVC at /data/shared (push.SharedRoot). Both hold for the current parent chart.

Verification

• make ci green.
• Unit tests: TestPlanTeardown (the artifact set) + TestRunDatasetRm_InvalidTableExitsTwo (the name guard, no cluster needed).
• Live --dry-run on EKS dev — discovers release/PVC, prints the plan, deletes nothing:

  Will delete
    mysql table:   training_test_datasets.demo_rm_preview
    pvc path:      /data/shared/demo_rm_preview
    pvc path:      /data/shared/.tracebloc-staging/demo_rm_preview

• The destructive path was NOT run — that needs your nod + a throwaway table first.

Refs #30, #31, #39.

🤖 Generated with Claude Code

---

Note

High Risk
Destructive, irreversible deletes against live cluster MySQL and shared PVC; relies on chart-specific pod naming and exec assumptions, and leaves backend catalog entries stale until #39.

Overview
Adds tracebloc dataset rm <table> to tear down artifacts from a prior dataset push: the MySQL table in training_test_datasets and the final + staging dirs on the shared PVC.

The command follows the same flow as push (validate table name → kubeconfig/cluster discovery → pre-flight plan → act), with --dry-run, TTY confirmation (or --yes / refuse off-terminal), and exit codes aligned with push (2 bad name, 3 kube/refused, 4 missing release/PVC, 7 teardown failure).

Teardown is implemented in internal/push via PlanTeardown and Teardown: DROP TABLE by exec into a running mysql pod (using in-pod $MYSQL_ROOT_PASSWORD) and rm -rf on PVC paths via exec into jobs-manager. The central backend catalog entry is not removed (noted as #39 follow-up).

Unit tests cover PlanTeardown artifact paths and early exit 2 for invalid table names without touching the cluster.

^{Reviewed by Cursor Bugbot for commit 7809580. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

👋 Heads-up — Code review queue is at 14 / 8

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#102 — chore(kanban): declare staging as prod-truth branch via .kanban.yml · author: @LukasWodka · no reviewer assigned
• averaging-service#98 — ci: add fr-gate caller workflow · author: @LukasWodka · no reviewer assigned
• cli#38 — feat(feat: UX polish — guided launcher + --output json (stretch) #29): --output-json for dataset push (PR-b) · author: @aptracebloc · no reviewer assigned
• client#191 — fix(#190): fail image-refresh loudly when the ingestor (ghcr) digest can't resolve · author: @saadqbal · no reviewer assigned
• client#192 — Installer UX: drop PriorityClass, fix namespace, one-per-machine guard, surface version · author: @LukasWodka · no reviewer assigned
• client-runtime#63 — feat(docs: sync v0.3.0 README to main #59): single-writer PodTokenReconciler + RECONCILER_MODE wiring + deploy-window lock · author: @saadqbal · no reviewer assigned
• data-ingestors#146 — test(schema): lock category enum to the engine's TaskCategory (anti-drift) · author: @LukasWodka · reviewer: @saadqbal
• data-ingestors#148 — Develop to prod · author: @divyasinghds · reviewer: @saadqbal
• design-system#19 — fix: un-track coverage/ and node_modules/ from git · author: @LukasWodka · no reviewer assigned
• design-system#22 — ci: add Vitest test workflow · author: @LukasWodka · reviewer: @aptracebloc

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

[LukasWodka]

👋 Heads-up — Code review queue is at 15 / 8

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#102 — chore(kanban): declare staging as prod-truth branch via .kanban.yml · author: @LukasWodka · no reviewer assigned
• averaging-service#98 — ci: add fr-gate caller workflow · author: @LukasWodka · no reviewer assigned
• client#191 — fix(#190): fail image-refresh loudly when the ingestor (ghcr) digest can't resolve · author: @saadqbal · no reviewer assigned
• client#192 — Installer UX: drop PriorityClass, fix namespace, one-per-machine guard, surface version · author: @LukasWodka · no reviewer assigned
• client#194 — test(requests-proxy): add helm-unittest coverage for requests-proxy Deployment · author: @saadqbal · no reviewer assigned
• client-runtime#63 — feat(docs: sync v0.3.0 README to main #59): single-writer PodTokenReconciler + RECONCILER_MODE wiring + deploy-window lock · author: @saadqbal · no reviewer assigned
• data-ingestors#146 — test(schema): lock category enum to the engine's TaskCategory (anti-drift) · author: @LukasWodka · reviewer: @saadqbal
• data-ingestors#148 — Develop to prod · author: @divyasinghds · reviewer: @saadqbal
• design-system#19 — fix: un-track coverage/ and node_modules/ from git · author: @LukasWodka · no reviewer assigned
• design-system#22 — ci: add Vitest test workflow · author: @LukasWodka · reviewer: @aptracebloc

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