DuckDB backend for benchmarks website (#7491)

## Summary

This PR started as a joke with @joseph-isaacs, but I actually think the
result is quite nice (even considering the diff size).

This PR changes the benchmarks website backend from having a specialized
ETL in JS, to processing the raw JSON into DuckDB tables, and using SQL
to serve all endpoints.

### Tooling

Moving to SQL also makes it easier to explore data locally, and this
change includes tools and documentation on how to get the data into a
local DuckDB instance, including generating all required SQL statements
and downloading the data.

### Docs

This PR includes new docs about the benchmarks website:
- README.md includes general setup, development and how to explore data
locally.
- SCHEMA.md describes the various tables and views, and how the API uses
them
- ETL.md attempts at describing the full lifecycle of the data, from
being generated by the post-merge CI run, how various fields actually
look and the processing steps it goes through before reaching the API.

---------

Signed-off-by: Adam Gutglick <adam@spiraldb.com>

Author: AdamGS

benchmarks-website/Dockerfile

@@ -1,16 +1,17 @@
-FROM node:24-alpine AS build
+FROM node:24-bookworm-slim AS build
 WORKDIR /app
 COPY package.json package-lock.json ./
 RUN npm ci
 COPY . .
 RUN npm run build
 
-FROM node:24-alpine
+FROM node:24-bookworm-slim
 WORKDIR /app
 COPY package.json package-lock.json ./
 RUN npm ci --omit=dev
 COPY --from=build /app/dist ./dist
 COPY server.js .
+COPY store ./store
 COPY src/config.js ./src/config.js
 EXPOSE 3000
 CMD ["node", "server.js"]

benchmarks-website/ETL.md

@@ -0,0 +1,206 @@
+# Benchmarks Website
+
+This directory contains the benchmark website frontend, the Node HTTP server, and the DuckDB-based
+refresh pipeline that turns the raw benchmark artifacts into chartable time series.
+
+For the data model and table relationships, start with [SCHEMA.md](./SCHEMA.md).
+For the upstream artifact generation and refresh/materialization flow, see [ETL.md](./ETL.md).
+
+## Prerequisites
+
+- Node.js `>=18`
+- npm
+- Optional: DuckDB CLI, if you want to query the cached artifacts directly
+
+Install dependencies:
+
+```bash
+cd benchmarks-website
+npm install
+```
+
+## Development Server
+
+Run the frontend and backend together:
+
+```bash
+cd benchmarks-website
+npm run dev
+```
+
+That starts:
+
+- Vite on `http://localhost:5173`
+- the API/static server on `http://localhost:3000`
+
+Useful endpoints:
+
+- `http://localhost:3000/api/metadata`
+- `http://localhost:3000/api/health`
+
+The backend refreshes from these artifact URLs by default:
+
+- `https://vortex-ci-benchmark-results.s3.amazonaws.com/data.json.gz`
+- `https://vortex-ci-benchmark-results.s3.amazonaws.com/commits.json`
+
+Relevant environment variables:
+
+```bash
+PORT=3000
+REFRESH_INTERVAL=300000
+DATA_URL=https://vortex-ci-benchmark-results.s3.amazonaws.com/data.json.gz
+COMMITS_URL=https://vortex-ci-benchmark-results.s3.amazonaws.com/commits.json
+CACHE_DIR=/path/to/local/cache
+```
+
+`CACHE_DIR` is the most useful one during development. If it is unset, the server uses a temp
+directory under `os.tmpdir()`.
+
+## Pull The Data Locally
+
+If you want a predictable local copy for exploration, populate a cache directory yourself and point
+the server at it.
+
+```bash
+cd benchmarks-website
+mkdir -p .cache/benchmarks
+
+curl -L \
+  https://vortex-ci-benchmark-results.s3.amazonaws.com/data.json.gz \
+  -o .cache/benchmarks/data.json.gz
+
+curl -L \
+  https://vortex-ci-benchmark-results.s3.amazonaws.com/commits.json \
+  -o .cache/benchmarks/commits.json
+```
+
+Then start the dev server against that cache:
+
+```bash
+cd benchmarks-website
+CACHE_DIR="$PWD/.cache/benchmarks" npm run dev
+```
+
+On first startup, the server will use the cached files immediately and then asynchronously
+revalidate them against S3.
+
+## Explore The Cached Data Directly
+
+Once `data.json.gz` and `commits.json` exist locally, you can query them with DuckDB without
+running the website.
+
+Example with the DuckDB CLI:
+
+```sql
+create view raw_commits as
+select *
+from read_json(
+  '.cache/benchmarks/commits.json',
+  format = 'newline_delimited',
+  compression = 'auto_detect',
+  columns = {
+    id: 'VARCHAR',
+    message: 'VARCHAR',
+    timestamp: 'VARCHAR',
+    author: 'JSON',
+    url: 'VARCHAR'
+  }
+);
+
+create view raw_benchmarks as
+select *
+from read_json(
+  '.cache/benchmarks/data.json.gz',
+  format = 'newline_delimited',
+  compression = 'auto_detect',
+  columns = {
+    name: 'VARCHAR',
+    unit: 'VARCHAR',
+    value: 'DOUBLE',
+    storage: 'VARCHAR',
+    dataset: 'JSON',
+    commit: 'JSON',
+    commit_id: 'VARCHAR'
+  }
+);
+```
+
+Useful starter queries:
+
+```sql
+select count(*) as commit_count from raw_commits;
+
+select count(*) as benchmark_count from raw_benchmarks;
+
+select split_part(name, '/', 1) as prefix, count(*) as rows
+from raw_benchmarks
+group by 1
+order by 2 desc
+limit 20;
+
+select
+  coalesce(json_extract_string(commit, '$.id'), commit_id) as resolved_commit_id,
+  count(*) as rows
+from raw_benchmarks
+group by 1
+order by 2 desc
+limit 20;
+```
+
+If you want the normalized relational model rather than the raw JSON views, follow the pipeline in
+[SCHEMA.md](./SCHEMA.md) and [`store/sql.js`](./store/sql.js).
+
+## Export The Full Bootstrap SQL
+
+If you want the exact SQL that the server uses to create all config tables, raw views, normalized
+tables, and derived projections, export it from the shared SQL builder:
+
+```bash
+cd benchmarks-website
+npm run export-sql -- \
+  --data-path "$PWD/.cache/benchmarks/data.json.gz" \
+  --commits-path "$PWD/.cache/benchmarks/commits.json" \
+  --output "$PWD/.cache/benchmarks/bootstrap.sql"
+```
+
+Then load it in DuckDB:
+
+```bash
+duckdb benchmark-explore.duckdb < .cache/benchmarks/bootstrap.sql
+```
+
+That creates the same tables and views the server uses, including:
+
+- `query_suites`
+- `valid_groups`
+- `engine_renames`
+- `raw_commits`
+- `raw_benchmarks`
+- `commit_dim`
+- `benchmarks_base`
+- `matched_suites`
+- `classified_benchmarks`
+- `benchmark_points`
+- `active_commits`
+- `benchmark_points_active`
+- `chart_defs`
+- `chart_latest_idx`
+- `chart_latest_values`
+- `chart_series_latest_values`
+
+If you want a portable template instead of path-specific SQL:
+
+```bash
+cd benchmarks-website
+npm run export-sql -- --placeholders --output bootstrap.template.sql
+```
+
+That emits a script using `__DATA_PATH__` and `__COMMITS_PATH__` placeholders.
+
+## Notes
+
+- The website only projects the subset of the raw benchmark JSON it needs for grouping, charting,
+  and summaries.
+- Benchmark names are part of the schema. Group, chart, and series identity are inferred from the
+  `name`, `storage`, and `dataset` fields during refresh.
+- The server returns `503` with `Retry-After` while the initial refresh is still loading.