Product analytics on React Router v8, Postgres (Neon) and Clerk

React Router v8 (framework mode): SSR, config/file routes under app/, loaders/actions, and resource routes for API endpoints.

Postgres on Neon via Drizzle ORM and the postgres-js driver.

Clerk: hosted identity (sign-in UI, sessions, user management) mounted via middleware + provider.

Product-analytics schema: tracked apps with unique ingest keys, an append-only event stream, session windows, and saved funnels, all hung off Better Auth's user table.

Database client

import { drizzle } from "drizzle-orm/postgres-js";
import postgres from "postgres";

// Neon pooled endpoint = PgBouncer transaction mode → prepared statements off.
// ponytail: single module-level client; the serverless runtime + PgBouncer do
// the pooling, so no custom pool/globalThis singleton dance needed.
const client = postgres(process.env.DATABASE_URL!, { prepare: false });

export const db = drizzle({ client });

// ponytail: Clerk is hosted — set the publishable + secret keys in the env (React Router v8
// reads VITE_CLERK_PUBLISHABLE_KEY client-side; CLERK_SECRET_KEY server-side). ClerkProvider +
// rootAuthLoader — wired in app/root.tsx by the shadcn UI shell — pick them up automatically.

owner-scoped app registrations, each with a unique api_key used to authenticate inbound event writes

the raw, immutable event stream keyed by app and timestamp, with freeform JSON properties and an optional anonymous visitor id

visitor session windows (start/end timestamps) correlated to events via anonymous_id, with nullable ended_at for open sessions

saved multi-step funnel definitions stored as JSON steps arrays, scoped per tracked app

Framework mode (not data/library mode): routes live under app/, declared in app/routes.ts. API endpoints are resource routes (a route module exporting loader/action but no default component).

Data flows through loaders (run on the server before render) and actions (mutations); components read it with useLoaderData / useActionData. There are no React Server Components — every server-rendered route is a loader plus a client component.

Auth gates in the loader, not in middleware: a protected route's loader calls requireAuth(request), which throws a redirect Response that React Router short-circuits on — so a logged-out user never reaches the protected data or renders the page.

The `@/` import alias maps to app/ (this framework's source root), so shared modules like @/lib/auth resolve under app/ — the one path prefix that differs from Next's src/, which is why the auth slice's mount code is framework-specific.

prepare: false is mandatory — Neon's pooled endpoint is PgBouncer in transaction mode, where server-side prepared statements break across the pool.

Drizzle is paired here (not Prisma): Prisma's prepared-statement reliance is incompatible with transaction-mode pooling.

Hosted: Clerk owns identity and does NOT create a local `user` table. Store `clerk_user_id` as text without a foreign key, or sync Clerk users into a local table via webhook before relying on FKs to `user`.

Route protection is clerkMiddleware() + auth.protect() in the Next proxy (getAuth() in a React Router loader, or event.context.auth() on Nuxt) — logged-out users bounce to Clerk's hosted sign-in, so there are no self-hosted auth pages to build or maintain.

Keeping the local mirror in sync is a webhook job: a svix-verified webhook route replays user.created / user.updated / user.deleted idempotently into the local `user` row, so app-type foreign keys to `user` resolve even though Clerk is the source of truth.

ClerkProvider (client) wraps the app so the hosted <SignIn/> / <UserButton/> components and hooks work; the publishable key is read client-side, while the secret key is only ever read server-side by clerkMiddleware.

api_key on tracked_apps carries a unique constraint: each app gets exactly one ingest key, and the key is the authentication surface for the event ingestion endpoint.

events is append-only (no update path, no status column) — the composite index on (app_id, occurred_at) is the only query handle; roll-up queries must scan that index, not mutate rows.

Clerk is a hosted identity provider and does not create a local `user` table. This schema's foreign keys to `user` assume a local identity table (as Better Auth provides). With Clerk, store `clerk_user_id` as a text column without a foreign key, or sync Clerk users into a local `users` table via webhook before relying on these FKs.

Product analytics

Ingest is the constraint everything else bends around. A write arrives carrying an api_key, and that column on tracked_apps is UNIQUE so the lookup is a single index probe returning an app_id — no join to identity on the write path, even though tracked_apps.owner_id foreign-keys Better Auth's user table for the dashboard side. Every other table hangs off that app id: events, sessions and funnels all reference tracked_apps.id with ON DELETE cascade, which makes an app registration the unit of scoping and the unit of deletion at once. events.occurred_at is notNull with no default, and that is the schema's most consequential line. The caller stamps the event; the database does not.

An SDK buffering through a dead network replays honestly because the backlog keeps its real instants — and there is no received-at column anywhere, so nothing records how late a batch showed up. Properties travel in a JSON column with no declared shape and no index, so a filter on a property value runs inside whatever slice idx_event_app_time (app_id, occurred_at) has already narrowed. sessions is not the parent of events. Nothing foreign-keys the two together; the correlation is anonymous_id, which is notNull on sessions, nullable on events, and indexed on neither, while idx_session_app covers app_id alone. Per-app, per-window questions are therefore cheap and stitching one visitor's path is not: bound by app and time first, then match ids inside that slice.

If visitor-level replay is the product rather than a report, add an index on (app_id, anonymous_id, occurred_at) before the event table gets large, because the schema will not do it for you. funnels stores definitions, not results — a name and a JSON steps array per app, with no index declared beyond the primary key. Evaluating a three-step funnel is three bounded scans of idx_event_app_time with a name filter applied inside each, and nothing caches the answer. That keeps funnel maintenance off the write path entirely and puts the whole cost on read, which is the right trade while an app's volume still fits a range scan and the wrong one once it does not.

React Router v8

React Router v8 in framework mode puts everything under app/, and `@/` maps to that root instead of Next's src/ — the one prefix that differs, which is why shared modules like @/lib/auth and @/db/schema stay byte-identical to their Next counterparts. initCode writes app/lib/db.ts, then the auth fragment adds app/lib/auth.ts, the resource route app/routes/api.auth.$.ts, and app/lib/require-auth.ts. The route table itself is app/routes.ts: routes are declared configuration, and a file becomes a URL because that table says so. There are no React Server Components here. Every server-rendered route is a loader plus an ordinary client component: the loader runs on the server before render, the component reads its result with useLoaderData, and mutations go through an action read back with useActionData.

An API endpoint is the same module minus the default export — a resource route, named with the flat dotted convention (app/routes/api.auth.$.ts for the auth splat, app/routes/webhooks.polar.ts for a webhook POST). Auth gates in the loader rather than in a middleware layer. A protected route awaits requireAuth(request) from app/lib/require-auth.ts, which calls auth.api.getSession({ headers: request.headers }) — a real server-side validation, not a cookie peek — and throws redirect("/sign-in") when there is no session. React Router treats a thrown Response as the route's outcome, so the loader short-circuits and neither the protected query nor the component ever runs. The trade that follows: there is no matcher array to widen and no edge tier to keep honest, but protection is per-route discipline.

A new route is protected because its loader calls requireAuth; forget the call and the page is public. In return, every gate sits one function call away from the data it guards, the session is already in hand when the loader queries db, and the same request-in / Response-out contract covers pages, API endpoints and the auth mount alike.

Postgres (Neon)

Postgres here is Neon reached through postgres-js, with Drizzle's pg-core dialect on top: drizzle({ client }) over a single module-level postgres(DATABASE_URL, { prepare: false }). That flag is not a preference. Neon's pooled (-pooler) endpoint is PgBouncer in transaction mode, where a backend is handed to a different session between statements, so server-side prepared statements break across the pool — and the same constraint is why this axis pairs with Drizzle rather than Prisma. One client per module is enough: PgBouncer and the runtime do the pooling, so there is no globalThis singleton dance. The schemas built on this dialect make three recurring type decisions. Primary keys are uuid(...).primaryKey().defaultRandom(), so ids come from the database. Timestamps are timestamp(..., { withTimezone: true }).defaultNow() — timestamptz, an absolute instant.

Closed value sets are text plus a CHECK constraint rather than pgEnum, so shipping a new role or subscription status is an ordinary constraint change instead of an ALTER TYPE migration. Counters are bigint({ mode: "number" }), and Better Auth's text user.id is referenced as text by the app tables rather than recast. Operationally, transaction-mode pooling forbids anything that spans statements on one backend: LISTEN/NOTIFY, session-scoped SET, advisory-lock sessions, WITH HOLD cursors. Those paths use Neon's direct endpoint instead. The connection client also changes with the deploy target — max: 1 per short-lived serverless instance, a real reused pool (max 10, idle_timeout 20) in a long-running Node process, and on Cloudflare Workers postgres-js is replaced outright by @neondatabase/serverless over HTTP, because Workers have no TCP sockets.

The capability that exists only on this side of the matrix is row-level security. Multi-tenant schemas ship ENABLE plus FORCE ROW LEVEL SECURITY with policies keyed on current_setting('app.current_org_id', true), which withTenant() sets per transaction — unset context yields no rows, so isolation fails closed inside the database rather than in application code. It requires a dedicated NOBYPASSRLS role: Neon's default neondb_owner carries BYPASSRLS, and connecting as it makes every policy silently inert.

Clerk

Clerk keeps identity on its own servers. The emitted app has no auth instance, no password column and no session table: the card at /sign-in is Clerk's <SignIn/> component rendered under a [[...sign-in]] catch-all so Clerk can mount its own verification and SSO-callback sub-routes there, and the endpoints behind it belong to Clerk. What does land in your database is a single mirror table. db/auth-schema.ts declares user with id set to the Clerk user id (text; varchar(255) on MySQL, so the app-type FK columns match exactly), plus email, first and last name, image URL, and an updated_at column used purely as a staleness key. It exists so app-type schemas can foreign-key user the way they would under a self-hosted auth.

It is not a source of truth, and application code should never write to it. Filling that mirror is a webhook job, and this fragment emits the whole path. lib/identity/record.ts holds recordClerkEvent; the mount is a Next route handler at src/app/api/webhooks/clerk/route.ts, an action in app/routes/webhooks.clerk.ts on React Router, or a Nitro .post.ts handler on Nuxt. Clerk delivers through svix, so the route verifies the raw body against CLERK_WEBHOOK_SECRET before anything reaches the database, and the record core is written for a delivery channel that retries and reorders: the staleness comparison lives inside the UPDATE's WHERE clause so an older event cannot clobber newer state, the insert path absorbs a concurrent duplicate (onConflictDoNothing on Postgres, an ER_DUP_ENTRY catch on MySQL), and user.deleted removes the row. Session checks never touch your Postgres.

Next's proxy.ts runs clerkMiddleware() and calls auth.protect() for anything matching createRouteMatcher(["/dashboard(.*)", "/settings(.*)"]); Nuxt reads event.context.auth() inside a Nitro middleware that the @clerk/nuxt module populates. Even the package name is framework-specific — @clerk/nextjs, @clerk/react-router, @clerk/nuxt — which upstreamPkgFor resolves per cell. The trade is concrete. You never build, style or maintain auth screens, and breaking changes arrive with Clerk's releases rather than your lockfile. In exchange, your user rows are eventually consistent with someone else's database, and a webhook you never configured is a table of missing foreign-key targets that only shows up when an app-type insert fails.