feat(worker): install native Miller in hosted image

Author: chayprabs

README.md

@@ -67,6 +67,7 @@ Current qualification note: Lighthouse is green, browser p95 is `377.26 ms` for
 
 - Render Blueprint: `render.yaml`
 - Render worker deploy notes: `docs/deploy/RENDER_WORKER.md`
+- Worker container: `apps/worker/Dockerfile` now installs native Miller for hosted fallback paths
 
 ## License
 

apps/worker/Dockerfile

@@ -1,4 +1,8 @@
-FROM node:22-alpine
+FROM node:22-bookworm-slim
+
+RUN apt-get update \
+  && apt-get install -y --no-install-recommends miller ca-certificates \
+  && rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app
 
@@ -8,6 +12,12 @@ RUN npm install
 COPY apps/worker/src ./src
 COPY apps/worker/tsconfig.json ./tsconfig.json
 
+ENV NODE_ENV=production
+ENV PORT=8797
+ENV WORKER_ARTIFACT_TTL_SECONDS=900
+
 EXPOSE 8797
 
+HEALTHCHECK --interval=15s --timeout=5s --start-period=10s --retries=5 CMD node -e "fetch('http://127.0.0.1:' + (process.env.PORT || 8797) + '/health').then((r)=>process.exit(r.ok?0:1)).catch(()=>process.exit(1))"
+
 CMD ["npx", "tsx", "src/index.ts"]

docs/deploy/RENDER_WORKER.md

@@ -12,6 +12,7 @@ This repo includes a `render.yaml` Blueprint for the native worker.
 - Health check path: `/health`
 - Runtime port: `10000`
 - Worker artifact TTL: `900`
+- Native Miller: installed in the container via `apps/worker/Dockerfile`
 
 ## Deploy flow
 
@@ -26,4 +27,5 @@ This repo includes a `render.yaml` Blueprint for the native worker.
 
 - The worker is modeled as a web service because it exposes an HTTP API at `/v1/run`.
 - Local compose still uses port `8797`; Render uses `PORT=10000`.
+- The worker image now installs the `mlr` binary so hosted health checks can expose real native Miller availability.
 - Public hosted verification is not complete until the deployed Render URL is recorded and checked.

docs/qc/APPENDIX_B_REPORT.md

@@ -44,11 +44,13 @@ Acceptance evidence:
 Blocked:
 - 21.0 Hosted web URL now exists and returned `200` at `https://chayprabs.github.io/csv-jsonl-miller/`, and the latest `main` runs for `CI`, `Deploy Pages`, and `Publish Worker Image` all succeeded on commit `8936ed5`.
   The repo now also contains a Render Blueprint in `render.yaml` and deploy notes in `docs/deploy/RENDER_WORKER.md` for creating a public worker service.
+  The worker container now installs native Miller directly in `apps/worker/Dockerfile`.
   A hosted worker URL is still not provisioned, so this item remains blocked.
 - 21.1 Browser Miller-WASM integration is still blocked.
   Browser DuckDB-WASM is now exercised by `pnpm --filter @csvshape/web smoke:duckdb`, which runs supported DuckDB-WASM chains over the ecommerce CSV, access-log JSONL, and wide-sales CSV samples and records the results in `docs/qc/benchmarks/browser-duckdb-smoke.json`.
   Worker-native DuckDB is packaged via `@duckdb/node-api`, reports `duckdbNative: true` on `/health`, and executes inline CSV/JSONL SQL plus Parquet export in local tests.
   Worker-native Miller now has a repeatable smoke run via `pnpm --filter @csvshape/worker smoke:mlr`, which returns `engine=mlr-native`, `rowCount=3`, and CSV output for the paid-order ecommerce subset in `docs/qc/benchmarks/native-mlr-smoke.json`.
+  The worker container now installs the `mlr` binary directly, which makes the hosted native Miller path concrete at the image level even though the browser Miller-WASM path is still unresolved.
   `pnpm probe:miller-wasm` now records a direct upstream `GOOS=js GOARCH=wasm` probe in `docs/qc/benchmarks/browser-miller-wasm-probe.json`; at the current Miller version, Go fails the build in the generated parser with `function too big ... exceeds 65536 blocks`, so there is still no usable browser Miller artifact to verify.
 - 21.13 Privacy evidence is now local-only rather than hosted.
   Browser-first privacy evidence is now available both locally and on the hosted app.

docs/qc/SECTION_21_STATUS.md

@@ -34,10 +34,12 @@ It is not yet a qualified release record.
   The static app is now hosted at `https://chayprabs.github.io/csv-jsonl-miller/`, and a direct HTTP check returned `200`.
   GitHub Actions on `main` now show green `CI`, `Deploy Pages`, and `Publish Worker Image` runs for commit `8936ed5`.
   The repo now also includes a Render Blueprint in `render.yaml` plus deploy notes in `docs/deploy/RENDER_WORKER.md` for a public worker URL.
+  The worker container now installs native Miller in `apps/worker/Dockerfile`, so the hosted image path aligns with the advertised native fallback engine set.
   A hosted worker URL is still missing, so this gate is not fully closed yet.
 - [ ] 21.1 Real Miller-WASM, DuckDB-WASM, native Miller, and native DuckDB integration evidence.
   Browser DuckDB-WASM is exercised locally across the acceptance samples via `pnpm --filter @csvshape/web smoke:duckdb`, with artifact output in `docs/qc/benchmarks/browser-duckdb-smoke.json`.
   Worker-native DuckDB execution is covered by local tests, and worker-native Miller now has a repeatable local smoke run via `pnpm --filter @csvshape/worker smoke:mlr` with artifact output in `docs/qc/benchmarks/native-mlr-smoke.json`.
+  The worker container now explicitly installs the `mlr` binary in `apps/worker/Dockerfile`, strengthening the hosted native Miller path.
   `pnpm probe:miller-wasm` now records a repeatable upstream browser probe in `docs/qc/benchmarks/browser-miller-wasm-probe.json`; the current direct `GOOS=js GOARCH=wasm` build fails inside Miller's generated parser with `function too big ... exceeds 65536 blocks`.
   Browser Miller-WASM integration evidence is still missing because that upstream build path does not currently produce a usable browser artifact.
 - [ ] 21.12 Performance evidence for browser p95 and worker throughput.