Cloudflare Hyperdrive

Status: Production Ready ✅ Last Updated: 2026-01-09 Dependencies: cloudflare-worker-base (recommended for Worker setup) Latest Versions: wrangler@4.58.0, pg@8.16.3+ (minimum), postgres@3.4.8, mysql2@3.16.0

Recent Updates (2025):

• July 2025: Configurable connection counts (min 5, max ~20 Free/~100 Paid)
• May 2025: 5x faster cache hits (regional prepared statement caching), FedRAMP Moderate authorization
• April 2025: Free plan availability (10 configs), MySQL GA support
• March 2025: 90% latency reduction (pools near database), IP access control (standard CF IP ranges)
• nodejs_compat_v2: pg driver no longer requires node_compat mode (auto-enabled with compatibility_date 2024-09-23+)
• Limits: 25 Hyperdrive configurations per account (Paid), 10 per account (Free)

---

Quick Start (5 Minutes)

1. Create Hyperdrive Configuration

# For PostgreSQL
npx wrangler hyperdrive create my-postgres-db \
  --connection-string="postgres://user:password@db-host.cloud:5432/database"

# For MySQL
npx wrangler hyperdrive create my-mysql-db \
  --connection-string="mysql://user:password@db-host.cloud:3306/database"

# Output:
# ✅ Successfully created Hyperdrive configuration
#
# [[hyperdrive]]
# binding = "HYPERDRIVE"
# id = "a76a99bc-7901-48c9-9c15-c4b11b559606"

Save the id value - you'll need it in the next step!

---

2. Configure Bindings in wrangler.jsonc

Add to your wrangler.jsonc:

{
  "name": "my-worker",
  "main": "src/index.ts",
  "compatibility_date": "2024-09-23",
  "compatibility_flags": ["nodejs_compat"],  // REQUIRED for database drivers
  "hyperdrive": [
    {
      "binding": "HYPERDRIVE",                     // Available as env.HYPERDRIVE
      "id": "a76a99bc-7901-48c9-9c15-c4b11b559606"  // From wrangler hyperdrive create
    }
  ]
}

CRITICAL:

• nodejs_compat flag is REQUIRED for all database drivers
• binding is how you access Hyperdrive in code (env.HYPERDRIVE)
• id is the Hyperdrive configuration ID (NOT your database ID)

---

3. Install Database Driver

# For PostgreSQL (choose one)
npm install pg           # node-postgres (most common)
npm install postgres     # postgres.js (modern, minimum v3.4.5)

# For MySQL
npm install mysql2       # mysql2 (minimum v3.13.0)

---

4. Query Your Database

PostgreSQL with node-postgres (pg):

import { Client } from "pg";

type Bindings = {
  HYPERDRIVE: Hyperdrive;
};

export default {
  async fetch(request: Request, env: Bindings, ctx: ExecutionContext) {
    const client = new Client({
      connectionString: env.HYPERDRIVE.connectionString
    });

    await client.connect();

    try {
      const result = await client.query('SELECT * FROM users LIMIT 10');
      return Response.json({ users: result.rows });
    } finally {
      // Clean up connection AFTER response is sent
      ctx.waitUntil(client.end());
    }
  }
};

MySQL with mysql2:

import { createConnection } from "mysql2/promise";

export default {
  async fetch(request: Request, env: Bindings, ctx: ExecutionContext) {
    const connection = await createConnection({
      host: env.HYPERDRIVE.host,
      user: env.HYPERDRIVE.user,
      password: env.HYPERDRIVE.password,
      database: env.HYPERDRIVE.database,
      port: env.HYPERDRIVE.port,
      disableEval: true  // REQUIRED for Workers (eval() not supported)
    });

    try {
      const [rows] = await connection.query('SELECT * FROM users LIMIT 10');
      return Response.json({ users: rows });
    } finally {
      ctx.waitUntil(connection.end());
    }
  }
};

---

5. Deploy

npx wrangler deploy

That's it! Your Worker now connects to your existing database via Hyperdrive with:

• ✅ Global connection pooling
• ✅ Automatic query caching
• ✅ Reduced latency (eliminates 7 round trips)

---

Known Issues Prevention

This skill prevents 11 documented issues with sources and solutions.

Issue #1: Windows/macOS Local Development - Hostname Resolution Failure

Error: Connection fails with hostname like xxx.hyperdrive.local Source: GitHub Issue #11556 Platforms: Windows, macOS 26 Tahoe, Ubuntu 24.04 LTS (wrangler@4.54.0+) Why It Happens: Hyperdrive local proxy hostname fails to resolve on certain platforms Prevention:

Use environment variable for local development:

export CLOUDFLARE_HYPERDRIVE_LOCAL_CONNECTION_STRING_HYPERDRIVE="postgres://user:password@localhost:5432/db"
npx wrangler dev

Or use wrangler dev --remote (caution: uses production database)

Status: Open issue, workaround available

---

Issue #2: postgres.js Hangs with IP Addresses

Error: Connection hangs indefinitely with no error message Source: GitHub Issue #6179 Why It Happens: Using IP address instead of hostname in connection string causes postgres.js to hang Prevention:

// ❌ WRONG - IP address causes indefinite hang
const connection = "postgres://user:password@192.168.1.100:5432/db"

// ✅ CORRECT - Use hostname
const connection = "postgres://user:password@db.example.com:5432/db"

Additional Gotcha: Miniflare (local dev) only supports A-z0-9 characters in passwords, despite Postgres allowing special characters. Use simple passwords for local development.

---

Issue #3: MySQL 8.0.43 Authentication Plugin Not Supported

Error: "unsupported authentication method" Source: GitHub Issue #10617 Why It Happens: MySQL 8.0.43+ introduces new authentication method not supported by Hyperdrive Prevention:

Use MySQL 8.0.40 or earlier, or configure user to use supported auth plugin:

ALTER USER 'username'@'%' IDENTIFIED WITH caching_sha2_password BY 'password';

Supported Auth Plugins: Only caching_sha2_password and mysql_native_password Status: Known issue tracked as CFSQL-1392

---

Issue #4: Local SSL/TLS Not Supported for Remote Databases

Error: SSL required but connection fails in local development Source: GitHub Issue #10124 Why It Happens: Hyperdrive local mode doesn't support SSL connections to remote databases (e.g., Neon, cloud providers) Prevention:

Use conditional connection in code:

const url = env.isLocal ? env.DB_URL : env.HYPERDRIVE.connectionString;
const client = postgres(url, {
  fetch_types: false,
  max: 2,
});

Alternative: Use wrangler dev --remote (⚠️ connects to production database) Timeline: SSL support planned for 2026 (requires workerd/Workers runtime changes, tracked as SQC-645)

---

Issue #5: Transaction Mode Resets SET Statements Between Queries

Error: SET statements don't persist across queries Source: Cloudflare Hyperdrive Docs - How Hyperdrive Works Why It Happens: Hyperdrive operates in transaction mode where connections are returned to pool after each transaction and RESET Prevention:

// ❌ WRONG - SET won't persist across queries
await client.query('SET search_path TO myschema');
await client.query('SELECT * FROM mytable'); // Uses default search_path!

// ✅ CORRECT - SET within transaction
await client.query('BEGIN');
await client.query('SET search_path TO myschema');
await client.query('SELECT * FROM mytable'); // Now uses myschema
await client.query('COMMIT');

⚠️ WARNING: Wrapping multiple operations in a single transaction to maintain SET state will affect Hyperdrive's performance and scaling.

---

Issue #6: Prisma Client Reuse Causes Hangs in Workers (Community-sourced)

Error: Worker hangs and times out after first request Source: GitHub Issue #28193 Verified: Multiple users confirmed Why It Happens: Prisma's connection pool attempts to reuse connections across request contexts, violating Workers' I/O isolation Prevention:

// ❌ WRONG - Global Prisma client reused across requests
const prisma = new PrismaClient({ adapter });

export default {
  async fetch(request: Request, env: Bindings) {
    // First request: works
    // Subsequent requests: hang indefinitely
    const users = await prisma.user.findMany();
    return Response.json({ users });
  }
};

// ✅ CORRECT - Create new client per request
export default {
  async fetch(request: Request, env: Bindings, ctx: ExecutionContext) {
    const pool = new Pool({
      connectionString: env.HYPERDRIVE.connectionString,
      max: 5
    });
    const adapter = new PrismaPg(pool);
    const prisma = new PrismaClient({ adapter });

    try {
      const users = await prisma.user.findMany();
      return Response.json({ users });
    } finally {
      ctx.waitUntil(pool.end());
    }
  }
};

---

Issue #7: Neon Serverless Driver Incompatible with Hyperdrive (Community-sourced)

Error: Hyperdrive provides no benefit with Neon serverless driver Source: Neon GitHub Repo, Cloudflare Docs Why It Happens: Neon's serverless driver uses WebSockets instead of TCP, bypassing Hyperdrive's connection pooling Prevention:

// ❌ WRONG - Neon serverless driver bypasses Hyperdrive
import { neon } from '@neondatabase/serverless';
const sql = neon(env.HYPERDRIVE.connectionString);
// This uses WebSockets, not TCP - Hyperdrive doesn't help

// ✅ CORRECT - Use traditional TCP driver with Hyperdrive
import postgres from 'postgres';
const sql = postgres(env.HYPERDRIVE.connectionString, {
  prepare: true,
  max: 5
});

Official Recommendation: Neon documentation states "On Cloudflare Workers, consider using Cloudflare Hyperdrive instead of this driver"

---

Issue #8: Supabase - Must Use Direct Connection String, Not Pooled (Community-sourced)

Error: Double-pooling causes connection issues Source: Cloudflare Docs - Supabase Why It Happens: Using Supabase pooled connection (Supavisor) creates double-pooling; Supavisor doesn't support prepared statements Prevention:

# ❌ WRONG - Using Supabase pooled connection (Supavisor)
npx wrangler hyperdrive create my-supabase \
  --connection-string="postgres://user:password@aws-0-us-west-1.pooler.supabase.com:6543/postgres"

# ✅ CORRECT - Use Supabase direct connection
npx wrangler hyperdrive create my-supabase \
  --connection-string="postgres://user:password@db.projectref.supabase.co:5432/postgres"

Reason: Hyperdrive provides its own pooling; double-pooling causes issues and breaks caching

---

Issue #9: Drizzle ORM with Nitro 3 - 95% Failure Rate with useDatabase (Community-sourced)

Error: 500 errors approximately 95% of the time Source: GitHub Issue #3893 Verified: Reproduced by multiple users Why It Happens: Nitro 3's built-in useDatabase (db0/integrations/drizzle) has I/O isolation issues Prevention:

// ❌ WRONG - Nitro's useDatabase fails ~95% of the time
import { useDatabase } from 'db0';
import { drizzle } from 'db0/integrations/drizzle';

export default eventHandler(async () => {
  const db = useDatabase();
  const users = await drizzle(db).select().from(usersTable);
  // Fails ~95% of the time with 500 error
});

// ✅ CORRECT - Create Drizzle client directly
import postgres from 'postgres';
import { drizzle } from 'drizzle-orm/postgres-js';

export default eventHandler(async (event) => {
  const sql = postgres(event.context.cloudflare.env.HYPERDRIVE.connectionString, {
    max: 5,
    prepare: true
  });
  const db = drizzle(sql);
  const users = await db.select().from(usersTable);
  event.context.cloudflare.ctx.waitUntil(sql.end());
  return { users };
});

Error Message: "Cannot perform I/O on behalf of a different request"

---

Issue #10: postgres.js Version Requirements for Caching (Community-sourced)

Error: Prepared statement caching doesn't work properly Source: Cloudflare Docs Why It Happens: postgres.js versions before 3.4.5 don't support Hyperdrive's prepared statement caching properly Prevention:

# Minimum version for Hyperdrive compatibility
npm install postgres@3.4.5

# Current recommended version
npm install postgres@3.4.8

Related: May 2025 prepared statement caching improvements require minimum version 3.4.5

---

Issue #11: WebSocket-Based Database Drivers Not Compatible

Error: Hyperdrive provides no benefit or causes connection issues Source: General pattern from Neon serverless driver issue Why It Happens: Hyperdrive requires TCP connections; WebSocket-based drivers bypass the TCP pooling layer Prevention:

Use traditional TCP-based drivers (pg, postgres.js, mysql2) instead of WebSocket-based drivers.

Affected Drivers: Any database driver using WebSockets instead of TCP

---

How Hyperdrive Works

Hyperdrive eliminates 7 connection round trips (TCP + TLS + auth) by:

• Edge connection setup near Worker (low latency)
• Connection pooling near database (March 2025: 90% latency reduction)
• Query caching at edge (May 2025: 5x faster cache hits)

Result: Single-region databases feel globally distributed.

---

Setup Steps

Prerequisites

• Cloudflare account with Workers access
• PostgreSQL (v9.0-17.x) or MySQL (v5.7-8.x) database
• Database accessible via public internet (TLS/SSL required) or private network (Cloudflare Tunnel)
• April 2025: Available on Free plan (10 configs) and Paid plan (25 configs)

Connection String Formats

# PostgreSQL
postgres://user:password@host:5432/database
postgres://user:password@host:5432/database?sslmode=require

# MySQL
mysql://user:password@host:3306/database

# URL-encode special chars: p@ssw$rd → p%40ssw%24rd

---

Connection Patterns

Single Connection (pg.Client)

const client = new Client({ connectionString: env.HYPERDRIVE.connectionString });
await client.connect();
const result = await client.query('SELECT ...');
ctx.waitUntil(client.end());  // CRITICAL: Non-blocking cleanup

Use for: Simple queries, single query per request

Connection Pool (pg.Pool)

const pool = new Pool({
  connectionString: env.HYPERDRIVE.connectionString,
  max: 5  // CRITICAL: Workers limit is 6 connections (July 2025: configurable ~20 Free, ~100 Paid)
});
const [result1, result2] = await Promise.all([
  pool.query('SELECT ...'),
  pool.query('SELECT ...')
]);
ctx.waitUntil(pool.end());

Use for: Parallel queries in single request

Connection Cleanup Rule

ALWAYS use ctx.waitUntil(client.end()) - non-blocking cleanup after response sent NEVER use await client.end() - blocks response, adds latency

---

ORM Integration

Drizzle ORM

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

const sql = postgres(env.HYPERDRIVE.connectionString, { max: 5 });
const db = drizzle(sql);
const allUsers = await db.select().from(users);
ctx.waitUntil(sql.end());

Prisma ORM

⚠️ CRITICAL: Do NOT reuse Prisma client across requests in Workers. Create new client per request.

import { PrismaPg } from "@prisma/adapter-pg";
import { PrismaClient } from "@prisma/client";
import { Pool } from "pg";

// ❌ WRONG - Global client causes hangs after first request
const prisma = new PrismaClient({ adapter });

export default {
  async fetch(request: Request, env: Bindings) {
    const users = await prisma.user.findMany(); // Hangs after first request
    return Response.json({ users });
  }
};

// ✅ CORRECT - Per-request client
export default {
  async fetch(request: Request, env: Bindings, ctx: ExecutionContext) {
    const pool = new Pool({ connectionString: env.HYPERDRIVE.connectionString, max: 5 });
    const adapter = new PrismaPg(pool);
    const prisma = new PrismaClient({ adapter });

    try {
      const users = await prisma.user.findMany();
      return Response.json({ users });
    } finally {
      ctx.waitUntil(pool.end());
    }
  }
};

Why: Prisma's connection pool attempts to reuse connections across request contexts, violating Workers' I/O isolation. Source: GitHub Issue #28193

Note: Prisma requires driver adapters (@prisma/adapter-pg).

---

Local Development

⚠️ SSL/TLS Limitations in Local Development

Important: Local Hyperdrive connections do NOT support SSL. This affects databases that require SSL (e.g., Neon, most cloud providers).

Workaround - Conditional Connection:

const url = env.isLocal ? env.DB_URL : env.HYPERDRIVE.connectionString;
const client = postgres(url, {
  fetch_types: false,
  max: 2,
});

Alternative: Use wrangler dev --remote (⚠️ connects to production database)

Timeline: SSL support planned for 2026 (requires workerd/Workers runtime changes) Source: GitHub Issue #10124, tracked as SQC-645

---

Local Connection Options

Option 1: Environment Variable (Recommended)

export CLOUDFLARE_HYPERDRIVE_LOCAL_CONNECTION_STRING_HYPERDRIVE="postgres://user:password@localhost:5432/local_db"
npx wrangler dev

Safe to commit config, no credentials in wrangler.jsonc.

Option 2: localConnectionString in wrangler.jsonc

{ "hyperdrive": [{ "binding": "HYPERDRIVE", "id": "prod-id", "localConnectionString": "postgres://..." }] }

⚠️ Don't commit credentials to version control.

Option 3: Remote Development

npx wrangler dev --remote  # ⚠️ Uses PRODUCTION database

---

Query Caching

Cached: SELECT (non-mutating queries) NOT Cached: INSERT, UPDATE, DELETE, volatile functions (LASTVAL, LAST_INSERT_ID)

May 2025: 5x faster cache hits via regional prepared statement caching.

Critical for postgres.js:

const sql = postgres(env.HYPERDRIVE.connectionString, {
  prepare: true  // REQUIRED for caching
});

Check cache status:

response.headers.get('cf-cache-status');  // HIT, MISS, BYPASS, EXPIRED

---

TLS/SSL Configuration

SSL Modes: require (default), verify-ca (verify CA), verify-full (verify CA + hostname)

Server Certificates (verify-ca/verify-full):

npx wrangler cert upload certificate-authority --ca-cert root-ca.pem --name my-ca-cert
npx wrangler hyperdrive create my-db --connection-string="postgres://..." --ca-certificate-id <ID> --sslmode verify-full

Client Certificates (mTLS):

npx wrangler cert upload mtls-certificate --cert client-cert.pem --key client-key.pem --name my-cert
npx wrangler hyperdrive create my-db --connection-string="postgres://..." --mtls-certificate-id <ID>

---

Private Database Access (Cloudflare Tunnel)

Connect to databases in private networks (VPCs, on-premises):

# 1. Install cloudflared (macOS: brew install cloudflare/cloudflare/cloudflared)
# 2. Create tunnel
cloudflared tunnel create my-db-tunnel

# 3. Configure config.yml
# tunnel: <TUNNEL_ID>
# ingress:
#   - hostname: db.example.com
#     service: tcp://localhost:5432

# 4. Run tunnel
cloudflared tunnel run my-db-tunnel

# 5. Create Hyperdrive
npx wrangler hyperdrive create my-private-db --connection-string="postgres://user:password@db.example.com:5432/database"

---

Critical Rules

Always Do

✅ Include nodejs_compat in compatibility_flags ✅ Use ctx.waitUntil(client.end()) for connection cleanup ✅ Set max: 5 for connection pools (Workers limit: 6) ✅ Enable TLS/SSL on your database (Hyperdrive requires it) ✅ Use prepared statements for caching (postgres.js: prepare: true) ✅ Set disableEval: true for mysql2 driver ✅ Handle errors gracefully with try/catch ✅ Use environment variables for local development connection strings ✅ Test locally with wrangler dev before deploying

Never Do

❌ Skip nodejs_compat flag (causes "No such module" errors) ❌ Use private IP addresses directly (use Cloudflare Tunnel instead) ❌ Use await client.end() (blocks response, use ctx.waitUntil()) ❌ Set connection pool max > 5 (exceeds Workers' 6 connection limit) ❌ Wrap all queries in transactions (limits connection multiplexing) ❌ Use SQL-level PREPARE/EXECUTE/DEALLOCATE (unsupported) ❌ Use advisory locks, LISTEN/NOTIFY (PostgreSQL unsupported features) ❌ Use multi-statement queries in MySQL (unsupported) ❌ Commit database credentials to version control ❌ Use IP addresses in connection strings instead of hostnames (causes postgres.js to hang) ❌ Use Neon serverless driver with Hyperdrive (uses WebSockets, bypasses pooling) ❌ Use Supabase pooled connection string (Supavisor) with Hyperdrive (double-pooling) ❌ Reuse Prisma client instances across requests in Workers (causes hangs and timeouts) ❌ Use Nitro 3's useDatabase with Drizzle and Hyperdrive (~95% failure rate) ❌ Expect SET statements to persist across queries (transaction mode resets connections) ❌ Use postgres.js versions before 3.4.5 (breaks prepared statement caching)

---

Wrangler Commands Reference

# Create Hyperdrive configuration
wrangler hyperdrive create <name> --connection-string="postgres://..."

# List all Hyperdrive configurations
wrangler hyperdrive list

# Get details of a configuration
wrangler hyperdrive get <hyperdrive-id>

# Update connection string
wrangler hyperdrive update <hyperdrive-id> --connection-string="postgres://..."

# Delete configuration
wrangler hyperdrive delete <hyperdrive-id>

# Upload CA certificate
wrangler cert upload certificate-authority --ca-cert <file>.pem --name <name>

# Upload client certificate pair
wrangler cert upload mtls-certificate --cert <cert>.pem --key <key>.pem --name <name>

---

Supported Databases

PostgreSQL (v9.0-17.x): AWS RDS/Aurora, Google Cloud SQL, Azure, Neon, Supabase, PlanetScale, Timescale, CockroachDB, Materialize, Fly.io, pgEdge, Prisma Postgres

MySQL (v5.7-8.x): AWS RDS/Aurora, Google Cloud SQL, Azure, PlanetScale, MariaDB (April 2025 GA)

NOT Supported: SQL Server, MongoDB, Oracle

---

Unsupported Features

PostgreSQL

• SQL-level prepared statements (PREPARE, EXECUTE, DEALLOCATE)
• Advisory locks
• LISTEN and NOTIFY
• Per-session state modifications

MySQL

• Non-UTF8 characters in queries
• USE statements
• Multi-statement queries
• Protocol-level prepared statements (COM_STMT_PREPARE)
• COM_INIT_DB messages
• Auth plugins other than caching_sha2_password or mysql_native_password

Workaround: For unsupported features, create a second direct client connection (without Hyperdrive).

---

Performance Best Practices

1. Avoid long-running transactions - Limits connection multiplexing
2. Use prepared statements - Enables query caching (postgres.js: prepare: true)
3. Set max: 5 for pools - Stays within Workers' 6 connection limit
4. Disable fetch_types if not needed - Reduces latency (postgres.js)
5. Use ctx.waitUntil() for cleanup - Non-blocking connection close
6. Cache-friendly queries - Prefer SELECT over complex joins
7. Index frequently queried columns - Improves query performance
8. Monitor with Hyperdrive analytics - Track cache hit ratios and latency
9. ⚠️ SET statement persistence - SET commands don't persist across queries due to transaction mode. Wrap SET + query in BEGIN/COMMIT if needed (impacts performance)

---

Troubleshooting

See references/troubleshooting.md for complete error reference with solutions.

Quick fixes:

Error	Solution
"No such module 'node:*'"	Add nodejs_compat to compatibility_flags
"TLS not supported by database"	Enable SSL/TLS on your database
"Connection refused"	Check firewall rules, allow public internet or use Tunnel
"Failed to acquire connection"	Use ctx.waitUntil() for cleanup, avoid long transactions
"Code generation from strings disallowed"	Set disableEval: true in mysql2 config
"Bad hostname"	Verify DNS resolves, check for typos
"Invalid database credentials"	Check username/password (case-sensitive)

---

Metrics and Analytics

Hyperdrive Dashboard → Select config → Metrics tab

Available: Query count, cache hit ratio, query latency (p50/p95/p99), connection latency, query/result bytes, error rate

---

Credential Rotation

# Option 1: Create new config (zero downtime)
wrangler hyperdrive create my-db-v2 --connection-string="postgres://new-creds..."
# Update wrangler.jsonc, deploy, delete old config

# Option 2: Update existing
wrangler hyperdrive update <id> --connection-string="postgres://new-creds..."

Best practice: Separate configs for staging/production.

---

References

• Official Documentation
• Get Started Guide
• How Hyperdrive Works
• Query Caching
• Local Development
• TLS/SSL Certificates
• Troubleshooting Guide
• Wrangler Commands
• Supported Databases

---

Last verified: 2026-01-21 | Skill version: 3.0.0 | Changes: Added 11 Known Issues Prevention entries (6 TIER 1, 5 TIER 2), expanded Prisma/Local Dev/Performance sections with critical warnings, updated Never Do rules

Package Versions: wrangler@4.58.0, pg@8.16.3+ (minimum), postgres@3.4.5+ (minimum for caching), postgres@3.4.8 (recommended), mysql2@3.16.0 Production Tested: Based on official Cloudflare documentation and community examples