ci: nightly pg_dump artifacts and restore runbook

.github/workflows/db-backup.yml

@@ -0,0 +1,62 @@
+name: Nightly DB backup
+
+# Logical dumps of the per-network Postgres (Neon/direct URL). Price history is
+# re-derivable from chain, but a nightly custom-format dump is the fast restore
+# path. See docs/backup-restore.md.
+on:
+  schedule:
+    # 03:00 UTC — one hour before the load test (04:00), so dump I/O and the
+    # 5k-RPS flood do not overlap.
+    - cron: '0 3 * * *'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: db-backup
+  cancel-in-progress: false
+
+jobs:
+  dump:
+    name: pg_dump (${{ matrix.network }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        network: [mainnet, testnet]
+    env:
+      DATABASE_URL_MAINNET: ${{ secrets.DATABASE_URL_MAINNET }}
+      DATABASE_URL_TESTNET: ${{ secrets.DATABASE_URL_TESTNET }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install postgresql-client
+        run: sudo apt-get update && sudo apt-get install -y postgresql-client
+
+      - name: Dump ${{ matrix.network }}
+        env:
+          NETWORK: ${{ matrix.network }}
+        run: |
+          set -euo pipefail
+          if [ "$NETWORK" = "mainnet" ]; then
+            DATABASE_URL="${DATABASE_URL_MAINNET:-}"
+          else
+            DATABASE_URL="${DATABASE_URL_TESTNET:-}"
+          fi
+          if [ -z "$DATABASE_URL" ]; then
+            echo "SKIP ${NETWORK}: secret not configured"
+            exit 0
+          fi
+          DATE_UTC=$(date -u +%Y%m%d)
+          OUT="lens-${NETWORK}-${DATE_UTC}.dump"
+          pg_dump --format=custom --no-owner --no-acl --dbname="$DATABASE_URL" --file="$OUT"
+          gzip -n "$OUT"
+
+      - uses: actions/upload-artifact@v4
+        if: success()
+        with:
+          name: lens-${{ matrix.network }}-${{ github.run_id }}
+          path: lens-${{ matrix.network }}-*.dump.gz
+          retention-days: 14
+          if-no-files-found: ignore

docs/DUAL_NETWORK.md

@@ -43,7 +43,7 @@ Dependencies: **#113 → #115 → #116 → #117**; #114 before the network selec
 | [#116](../../issues/116) | Per-network Soroswap/Reflector/Aquarius/token-list | #113 |
 | [#117](../../issues/117) | Launch ingesters per network | #113–#116 |
 | [#118](../../issues/118) | Network selector on routes + per-request x402 | #114 |
-| [#119](../../issues/119) | Nightly `pg_dump` backup + restore runbook | — |
+| [#119](../../issues/119) | Nightly `pg_dump` backup + [restore runbook](backup-restore.md) | — |
 | [#120](../../issues/120) | Mainnet deploy guide | the rest |
 
 ## Ops (Render + UptimeRobot + external Postgres)
@@ -53,5 +53,6 @@ Dependencies: **#113 → #115 → #116 → #117**; #114 before the network selec
   mainnet always-on, testnet on-demand / a second account / a paid instance.
 - **Database:** use Neon or another managed Postgres. **Do not** use Render's
   free Postgres — it is **deleted after 90 days**. One DB per network.
-- **Durability fallback:** nightly `pg_dump` (#119) for fast restore; Lens is an
-  aggregator, so the DB is also re-derivable by re-ingesting from chain.
+- **Durability fallback:** nightly `pg_dump` (#119) for fast restore; see
+  [backup-restore.md](backup-restore.md). Lens is an aggregator, so the DB is
+  also re-derivable by re-ingesting from chain.

docs/backup-restore.md

@@ -0,0 +1,74 @@
+# Backup and restore
+
+Nightly `pg_dump` of each network's Postgres. Lens stores price history in
+**one database per network** (no `network` column); restoring a mainnet dump
+into the testnet instance would mix pairs. See [DUAL_NETWORK.md](DUAL_NETWORK.md).
+
+The workflow is [`.github/workflows/db-backup.yml`](../.github/workflows/db-backup.yml):
+cron `0 3 * * *` UTC plus `workflow_dispatch`. Dumps are GitHub Actions
+artifacts, kept **14 days**.
+
+## Secrets
+
+Add these on the GitHub repo (Settings → Secrets and variables → Actions).
+Use the **direct** connection string, not the pooler:
+
+| Secret | Database |
+|--------|----------|
+| `DATABASE_URL_MAINNET` | mainnet Postgres |
+| `DATABASE_URL_TESTNET` | testnet Postgres |
+
+On Neon that means the host **without** `-pooler`, port **5432** (not 6543),
+and `sslmode=require`. Prisma's pooled `DATABASE_URL` is the runtime URL;
+`pg_dump` / `pg_restore` need the same class of URI as `DIRECT_DATABASE_URL`.
+
+A missing secret skips that network (`SKIP <network>: secret not configured`)
+so a deploy that only has mainnet still dumps. Configure both for full
+coverage.
+
+Do not put these URIs in the repo, `.env.example`, or workflow YAML.
+
+## Download a dump
+
+1. Actions → **Nightly DB backup** → the run you want.
+2. Artifacts: `lens-mainnet-<run_id>` / `lens-testnet-<run_id>`.
+3. File inside: `lens-<network>-YYYYMMDD.dump.gz`.
+
+`workflow_dispatch` is the way to take a dump on demand (and to verify the
+job once secrets are in place).
+
+## Restore into a fresh instance
+
+Do not `pg_restore` onto the live URL. Spin up an empty database, restore
+there, then swap the deployment env.
+
+1. Create a new Neon project (or local Postgres). Match the dump's major
+   version; if `pg_restore` complains, install a client ≥ the server
+   (`postgresql-client-16` / `17` as needed).
+2. Copy the **direct** URI for the new instance (`sslmode=require` on Neon).
+3. Restore **one network per instance**:
+
+```bash
+gzip -dc lens-mainnet-YYYYMMDD.dump.gz \
+  | pg_restore --no-owner --no-acl --dbname="$NEW_DIRECT_URL"
+```
+
+`pg_restore -l` on the gunzipped file lists the TOC if you want to inspect
+before loading.
+
+4. Point the matching Lens deployment at the new instance:
+   - `DATABASE_URL` — pooled URL is fine for the app
+   - `DIRECT_DATABASE_URL` — direct URL (Prisma migrations / `db push`)
+5. Check `/status`, then one `/prices/history` (or GraphQL equivalent) for a
+   watched pair.
+6. Rollback is swapping those two env vars back to the previous instance.
+
+Repeat independently for testnet with the testnet artifact and testnet
+deployment. Never restore a mainnet dump into testnet, or the reverse.
+
+## Client notes
+
+Compose uses Postgres 15. Managed Neon may be 16/17. `pg_dump` / `pg_restore`
+must be the same major as the server or newer. The workflow installs Ubuntu's
+`postgresql-client`; bump it in the workflow if dumps start failing with a
+version error.

src/__tests__/db-backup.workflow.test.ts

@@ -0,0 +1,72 @@
+import { describe, it, expect } from 'vitest'
+import { readFileSync } from 'node:fs'
+import { resolve } from 'node:path'
+import YAML from 'js-yaml'
+
+const root = resolve(__dirname, '../..')
+const workflowPath = resolve(root, '.github/workflows/db-backup.yml')
+const runbookPath = resolve(root, 'docs/backup-restore.md')
+
+function read(path: string): string {
+  return readFileSync(path, 'utf8')
+}
+
+describe('nightly pg_dump workflow', () => {
+  it('exists and is valid YAML', () => {
+    const raw = read(workflowPath)
+    expect(raw.length).toBeGreaterThan(0)
+    expect(() => YAML.load(raw)).not.toThrow()
+  })
+
+  it('runs on a nightly schedule and workflow_dispatch only', () => {
+    const raw = read(workflowPath)
+    const doc = YAML.load(raw) as {
+      on?: {
+        schedule?: { cron: string }[]
+        workflow_dispatch?: unknown
+        pull_request?: unknown
+        push?: unknown
+      }
+    }
+    const crons = (doc.on?.schedule ?? []).map((s) => s.cron)
+    expect(crons).toContain('0 3 * * *')
+    expect(doc.on?.workflow_dispatch).toBeDefined()
+    expect(doc.on?.pull_request).toBeUndefined()
+    expect(doc.on?.push).toBeUndefined()
+  })
+
+  it('covers both networks via secrets and keeps artifacts 14 days', () => {
+    const raw = read(workflowPath)
+    expect(raw).toMatch(/secrets\.DATABASE_URL_MAINNET/)
+    expect(raw).toMatch(/secrets\.DATABASE_URL_TESTNET/)
+    expect(raw).toMatch(/mainnet/)
+    expect(raw).toMatch(/testnet/)
+    expect(raw).toMatch(/retention-days:\s*14/)
+    expect(raw).not.toMatch(/postgres:\/\/[^\s]+:[^\s]+@/)
+  })
+
+  it('dumps custom format, gzips, and does not echo the URL', () => {
+    const raw = read(workflowPath)
+    expect(raw).toMatch(/pg_dump/)
+    expect(raw).toMatch(/--format=custom/)
+    expect(raw).toMatch(/\bgzip\b/)
+    expect(raw).toMatch(/--no-owner/)
+    expect(raw).toMatch(/--no-acl/)
+    expect(raw).not.toMatch(/echo\s+"?\$\{?DATABASE_URL/)
+    expect(raw).not.toMatch(/set\s+-x/)
+  })
+})
+
+describe('backup restore runbook', () => {
+  it('documents pg_restore into a fresh instance and URL swap for both networks', () => {
+    const raw = read(runbookPath)
+    expect(raw).toMatch(/pg_restore/)
+    expect(raw).toMatch(/Neon/i)
+    expect(raw).toMatch(/DATABASE_URL/)
+    expect(raw).toMatch(/DIRECT_DATABASE_URL/)
+    expect(raw).toMatch(/mainnet/i)
+    expect(raw).toMatch(/testnet/i)
+    expect(raw).toMatch(/pooler|direct/i)
+    expect(raw).toMatch(/fresh/i)
+  })
+})