Review: Restore Ollama Helm release to ops module (embedding pipeline broken)

Verdict: APPROVED

Previous Findings — Disposition

• [SCOPE] Cross-repo note — FIXED. Issue body now has a prominent "Note:" under the Repo section: "This issue is filed on pal-e-services for historical reasons (original triage). All file targets are in pal-e-platform. The agent must clone pal-e-platform, not pal-e-services."
• [BODY] outputs.tf promoted — FIXED. File Targets now explicitly lists terraform/modules/ops/outputs.tf as a definite restore target with exact output name (ollama_namespace). Verified: pre-removal file at 7872dac~1 contained exactly that output.
• [BODY] Backfill command specified — FIXED. AC 8 added: kubectl exec deployment/pal-e-docs-embedding-worker -n pal-e-docs -- python -m pal_e_docs.embedding_worker --backfill.
• [LABEL] Type corrected — FIXED. Issue ### Type is now "Feature". Board labels show type:feature.
• [LABEL] Titles aligned — FIXED. Board item and Forgejo issue both read "Restore Ollama Helm release to ops module (embedding pipeline broken)".
• [SCOPE] Missing arch-infra note — DEFERRED. No arch-infra note exists. Acknowledged as non-blocking by the caller. Does not prevent agent execution.

Template Completeness

• [x] Type — Feature
• [x] Lineage — standalone, references removal commit 7872dac
• [x] Repo — ldraney/pal-e-platform (terraform/modules/ops/main.tf), cross-repo note present
• [x] User Story — platform operator, Ollama restored, embedding worker functional
• [x] Context — thorough history of deploy, removal, and current state
• [x] File Targets — 2 modify targets (main.tf, outputs.tf) + don't-touch list
• [x] Feature Flag — none (correct for infra work)
• [x] Acceptance Criteria — 8 items, all verifiable
• [x] Test Expectations — 3 items with run command
• [x] Constraints — 6 items, specific and actionable
• [x] Checklist — present
• [x] Related — references project, commits, related issue

Traceability

• [x] story:superuser-query label — "I can query the knowledge base by meaning (semantic search)"
• [x] story note verified — found in project-pal-e-docs user-stories section, row 1
• [x] arch:infra label — infrastructure component
• [ ] arch note DEFERRED — no arch-infra note exists. Non-blocking per caller.
• [x] Forgejo issue — ldraney/pal-e-services#114, open
• [x] Cross-repo placement documented — Note in Repo section directs agent to clone pal-e-platform

File Targets

• [x] terraform/modules/ops/main.tf — verified: file exists (213 lines). Ollama resources (namespace + helm_release, ~70 lines) were removed between nvidia_device_plugin (line 26) and embedding_worker_metrics (line 28). Pre-removal config confirmed at 7872dac~1.
• [x] terraform/modules/ops/outputs.tf — verified: file exists, is empty. Pre-removal version at 7872dac~1 had output "ollama_namespace". Now listed as definite restore target in issue.

• [x] terraform/main.tf — module "ops" call at line 123, moved{} blocks present. No Ollama references remain. Correct to leave untouched.
• [x] terraform/modules/ops/variables.tf — no Ollama variables existed before removal (confirmed identical at 7872dac~1). Correctly excluded from file targets.
• [x] NVIDIA / embedding-worker-metrics resources — present in current main.tf, must not be modified.

Repo Placement

Dependencies

• #1434 (Scale embedding worker to 0) — alternative stopgap. Becomes unnecessary if #1432 succeeds. Not a blocker.
• #1433 (pg_stat_statements + Grafana) — independent, no dependency.
• NVIDIA device plugin — already deployed in ops module. Issue correctly notes depends_on relationship. Not a blocker.
• Embedding worker — downstream consumer. Config at src/pal_e_docs/config.py:11 hardcodes ollama.ollama.svc.cluster.local:11434. Will self-heal once Ollama is reachable.

Acceptance Criteria

• [x] AC 1-3: namespace + helm_release + output restored — verifiable via tofu plan
• [x] AC 4: add-only plan — verifiable via tofu plan
• [x] AC 5: tofu apply succeeds — verifiable (requires cluster access)
• [x] AC 6: pods running — verifiable via kubectl
• [x] AC 7: embedding worker connectivity — verifiable via logs
• [x] AC 8: embedding backfill — now specifies exact kubectl exec command

Blast Radius

• pal-e-docs embedding worker — direct consumer. Will start working once Ollama is reachable. No code changes needed.
• pal-e-docs semantic search — degraded to keyword-only. Will improve as backfill processes blocks.
• No other services reference Ollama — grep across pal-e-deployments and pal-e-services/terraform found zero Ollama references.
• GPU contention — Ollama will claim the GPU. Acknowledged in issue's Decision section.

Decomposition Assessment

• 2 file targets in 1 repo — under threshold
• 8 AC but work is essentially a single git-restore operation — under 5 minutes
• Single agent pass is appropriate

Recommendations