fix(db): keep the free SQLite Cloud node from idling into a pause

.github/workflows/db-keepalive.yml

@@ -0,0 +1,51 @@
+name: db-keepalive
+
+# SQLite Cloud parks a free node after a stretch of inactivity, and a parked
+# node refuses new connections — so no amount of visitor traffic revives it,
+# only a restart from https://dashboard.sqlitecloud.io. Prevention is the only
+# automatable half: this keeps a query flowing so the node never goes idle
+# long enough to be parked.
+#
+# Two caveats worth knowing before trusting it as the sole guard:
+#   * GitHub delays scheduled runs under load and drops them entirely when the
+#     repo has had no activity for 60 days, so treat the cadence as best-effort.
+#   * It probes through the deployed site, so a Railway outage shows up here as
+#     a failure too. That is deliberate — both mean c0upons is down.
+on:
+  schedule:
+    - cron: '*/15 * * * *'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: db-keepalive
+  cancel-in-progress: true
+
+jobs:
+  ping:
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - name: Query the database
+        env:
+          BASE_URL: ${{ vars.KEEPALIVE_BASE_URL || 'https://c0upons.com' }}
+        run: |
+          set -uo pipefail
+          # Retry a couple of times so one dropped request doesn't page anyone;
+          # a genuinely paused node fails all three the same way.
+          for attempt in 1 2 3; do
+            status=$(curl -fsS -m 30 -o body.txt -w '%{http_code}' \
+              "$BASE_URL/api/health/db" || echo 000)
+            if [ "$status" = "200" ]; then
+              echo "database awake (attempt $attempt)"
+              exit 0
+            fi
+            echo "attempt $attempt: HTTP $status"
+            cat body.txt || true
+            sleep 10
+          done
+          echo "::error::Database unreachable after 3 attempts (last status $status)."
+          echo "::error::If the body says the node is paused, restart it at https://dashboard.sqlitecloud.io"
+          exit 1

apps/web/app/api/health/db/route.ts

@@ -0,0 +1,29 @@
+import { NextResponse } from 'next/server';
+import { getDb } from '@/lib/db';
+import { dbErrorResponse } from '@/lib/api-error';
+
+// Never cache: a cached 200 would keep reporting health after the node parks,
+// and the keep-alive query has to actually reach SQLite Cloud to count as
+// activity.
+export const dynamic = 'force-dynamic';
+
+/**
+ * Liveness probe for the database, and the query the keep-alive schedule runs.
+ *
+ * SQLite Cloud parks a free node after a stretch with no queries, and a parked
+ * node cannot be woken by traffic — only by a restart from the dashboard. So
+ * the cheapest cure is to never go idle: `.github/workflows/db-keepalive.yml`
+ * calls this on a schedule, and the `SELECT 1` is the activity that keeps the
+ * node awake. It doubles as monitoring — a paused node answers 503 here loudly
+ * instead of silently emptying the pages that swallow their own DB errors.
+ */
+export async function GET() {
+  try {
+    const db = getDb();
+    await db.sql`SELECT 1`;
+    return NextResponse.json({ ok: true }, { headers: { 'Cache-Control': 'no-store' } });
+  } catch (err) {
+    console.error(err);
+    return dbErrorResponse(err, 'Database health check failed');
+  }
+}