sailfork / t3code-github / .repos/ alchemy-effect/ website/ src/ content/ docs/ cloudflare/ data/ branch-from-shared-database.mdx

import { Code, Tabs, TabItem } from "@astrojs/starlight/components";

In the Hyperdrive tutorial you provisioned a Neon.Project and a Neon.Branch, then Drizzle layered automatic migrations on top. Both work fine with a single stage — but the moment you start deploying PR previews (pr-147, pr-148, …) the default flow becomes painful:

  • Every PR stage provisions its own Neon.Project — a whole Postgres cluster, with cold starts, project-count limits, and a full migration replay against an empty branch.
  • Tearing the PR stage down deletes the project too, throwing away the data and any seeds you'd attached.

The fix is to share the project across stages — long-lived, owned by staging — and only fork a per-stage branch off it. PR stages get a fast, isolated, copy-on-write database without touching project creation.

Where we left off

Recap of src/Db.ts from the Drizzle tutorial — every stage unconditionally creates its own Neon.Project:

// src/Db.ts
import * as Cloudflare from "@alchemy.run/cloudflare";
import * as Drizzle from "@alchemy.run/drizzle";
import * as Neon from "@alchemy.run/neon";
import * as Effect from "effect/Effect";

export const NeonDb = Effect.gen(function* () {
  const schema = yield* Drizzle.Schema("app-schema", {
    schema: "./src/schema.ts",
    out: "./migrations",
  });

  const project = yield* Neon.Project("app-db", {
    region: "aws-us-east-1",
  });

  const branch = yield* Neon.Branch("app-branch", {
    project,
    migrations: schema,
  });

  return { project, branch, schema };
});

export const Hyperdrive = Effect.gen(function* () {
  const { branch } = yield* NeonDb;
  return yield* Cloudflare.Hyperdrive.Connection("app-hyperdrive", {
    origin: branch.origin,
  });
});

Today we'll change the project line — and only the project line — so PR stages reuse staging's.

Read the active stage

The branching decision needs to know which stage is being deployed. Pull it off the Stack service, which Alchemy provides to every stack effect:

+import * as Alchemy from "alchemy";
 import * as Cloudflare from "@alchemy.run/cloudflare";
 import * as Drizzle from "@alchemy.run/drizzle";
 import * as Neon from "@alchemy.run/neon";
 import * as Effect from "effect/Effect";

 export const NeonDb = Effect.gen(function* () {
+  const { stage } = yield* Alchemy.Stack;
+
   const schema = yield* Drizzle.Schema("app-schema", {
     schema: "./src/schema.ts",
     out: "./migrations",
   });

stage is whatever was passed to bun alchemy deploy --stage <name> — dev_sam, staging, prod, or pr-147. We'll branch on stage.startsWith("pr-") to decide whether to create a project or reference one.

Reference the shared project for PR stages

Replace the unconditional Neon.Project(...) with a conditional that uses Neon.Project.ref(...) when the stage is a PR preview:

-  const project = yield* Neon.Project("app-db", {
-    region: "aws-us-east-1",
-  });
+  const project = stage.startsWith("pr-")
+    ? yield* Neon.Project.ref("app-db", { stage: "staging" })
+    : yield* Neon.Project("app-db", {
+        region: "aws-us-east-1",
+      });

Resource.ref returns a typed reference to a resource that's already been deployed in another stage. The lookup is keyed by { stack, stage, id }:

  • The logical ID "app-db" matches the project the staging stage creates.
  • stage: "staging" tells the lookup which stage's state file to read from.
  • The current stack is implied — both stages share it.

project has type Neon.Project regardless of which branch ran. The downstream Neon.Branch doesn't know (or care) which path produced it.

For the bigger picture see Shared database across stages.

The branch is still per-stage

Nothing changes below the project line. Neon.Branch still creates a fresh branch per stage, off whichever project the conditional returned:

const branch = yield* Neon.Branch("app-branch", {
  project,
  migrations: schema,
});

Branches are copy-on-write and free, so each PR stage gets a fully isolated database — its own writes, its own migration history — without paying for a whole Postgres cluster.

Assembled file

// src/Db.ts
import * as Alchemy from "alchemy";
import * as Cloudflare from "@alchemy.run/cloudflare";
import * as Drizzle from "@alchemy.run/drizzle";
import * as Neon from "@alchemy.run/neon";
import * as Effect from "effect/Effect";

export const NeonDb = Effect.gen(function* () {
  const { stage } = yield* Alchemy.Stack;

  const schema = yield* Drizzle.Schema("app-schema", {
    schema: "./src/schema.ts",
    out: "./migrations",
  });

  const project = stage.startsWith("pr-")
    ? yield* Neon.Project.ref("app-db", { stage: "staging" })
    : yield* Neon.Project("app-db", {
        region: "aws-us-east-1",
      });

  const branch = yield* Neon.Branch("app-branch", {
    project,
    migrations: schema,
  });

  return { project, branch, schema };
});

export const Hyperdrive = Effect.gen(function* () {
  const { branch } = yield* NeonDb;
  return yield* Cloudflare.Hyperdrive.Connection("app-hyperdrive", {
    origin: branch.origin,
  });
});

Deploy staging to seed the shared project

PR stages can only reference what staging has already deployed, so staging has to go first:

bun alchemy deploy --stage staging

This runs the full path — creates app-db (the project), app-branch against it, and app-hyperdrive in front. The state for Neon.Project("app-db") lands in the staging state file, which is what PR stages will read against.

Deploy a PR stage

Now deploy a PR-preview stage. The same code, different stage flag:

bun alchemy deploy --stage pr-147

The plan is noticeably shorter:

  • Neon.Project("app-db") → resolved as a reference to staging's project. No API call.
  • Neon.Branch("app-branch") → freshly created against the referenced project.
  • Cloudflare.Hyperdrive.Connection("app-hyperdrive") → points at the new branch's origin.
  • Api → freshly created.

The PR stage has its own isolated branch, its own Hyperdrive, its own Worker — and it shares the project with staging.

:::tip If you alchemy deploy --stage pr-147 before staging exists, plan fails with InvalidReferenceError because there's nothing under { stack, stage: "staging", id: "app-db" } to read. Deploy staging first. :::

Tear it down

bun alchemy destroy --stage pr-147

Destroy removes the per-stage resources — the branch, the Hyperdrive, the Worker — but leaves the referenced project untouched. Alchemy doesn't own app-db from this stage's perspective, so it can't delete it. The shared project keeps serving every other PR stage and staging itself.

The same recipe on PlanetScale

Nothing here is Neon-specific. PlanetScale has the same shape — an expensive long-lived database, cheap per-stage branches — so the same conditional works, with Planetscale.PostgresDatabase.ref (or MySQLDatabase.ref) in place of Neon.Project.ref:

// src/Db.ts
export const PlanetscaleDb = Effect.gen(function* () {
  const { stage } = yield* Alchemy.Stack;

  const schema = yield* Drizzle.Schema("app-schema", {
    schema: "./src/schema.ts",
    out: "./migrations",
  });

  const database = stage.startsWith("pr-")
    ? yield* Planetscale.PostgresDatabase.ref("app-db", {
        stage: `staging-${stage}`,
      })
    : yield* Planetscale.PostgresDatabase("app-db", {
        region: { slug: "us-east" },
        clusterSize: "PS_10",
      });

  const branch = yield* Planetscale.PostgresBranch("app-branch", {
    database,
    migrations: schema,
  });

  const role = yield* Planetscale.PostgresRole("app-role", {
    database,
    branch,
    inheritedRoles: ["postgres"],
  });

  return { database, branch, role, schema };
});
// src/Db.ts
export const PlanetscaleDb = Effect.gen(function* () {
  const { stage } = yield* Alchemy.Stack;

  const database = stage.startsWith("pr-")
    ? yield* Planetscale.MySQLDatabase.ref("app-db", {
        stage: `staging-${stage}`,
      })
    : yield* Planetscale.MySQLDatabase("app-db", {
        region: { slug: "us-east" },
        clusterSize: "PS_10",
      });

  const branch = yield* Planetscale.MySQLBranch("app-branch", {
    database,
    isProduction: false,
    migrations: "./migrations",
  });

  const password = yield* Planetscale.MySQLPassword("app-password", {
    database,
    branch,
    role: "readwriter",
  });

  return { database, branch, password };
});

The database is the shared tier; the branch and its credential (PostgresRole / MySQLPassword) are per-stage, just like Neon.Branch was.

A staging tier per PR

Look closely at the ref: it targets staging-${stage}, not a single global staging. Deploying pr-42 reads the database from a parallel staging-pr-42 stage:

Planetscale.PostgresDatabase.ref("app-db", {
  stage: `staging-${stage}`, // pr-42 → staging-pr-42
});

Deriving the staging stage name from the PR stage gives every PR its own database owner — isolated data per PR — while the create-vs-reference split stays identical. Deploy staging-pr-42 first (it takes the create path), then pr-42 (it takes the ref path). Prefer one database shared by all PRs? Use stage: "staging" and you're back to exactly the Neon flow above.

Point Hyperdrive at the credential

PlanetScale's connection string comes from the per-stage credential rather than the branch, so the Hyperdrive origin changes accordingly:

export const Hyperdrive = Effect.gen(function* () {
  const { role } = yield* PlanetscaleDb; // { password } for MySQL
  return yield* Cloudflare.Hyperdrive.Connection("app-hyperdrive", {
    origin: role.origin, // password.origin for MySQL
    caching: { disabled: true },
  });
});

The full working projects live at cloudflare-planetscale-postgres-drizzle and cloudflare-planetscale-mysql-drizzle.

Where to from here

You now have:

  • A shared Neon project owned by staging, surviving every PR stage's lifecycle.
  • Per-stage Neon branches with their own migrations and data.
  • A deploy/destroy story that scales to dozens of preview environments without flooding Neon with one project per PR.

The same Resource.ref pattern works for any other expensive, shared resource you have — an S3 bucket of seed fixtures, a global rate-limiter Durable Object, a DNS zone. Anywhere a per-stage copy is wasteful, lift it to a long-lived stage and reference it from the rest.