When building modern web applications, handling background jobs, managing state, and ensuring fast database access are critical. Recently, I migrated my Bun + Hono project to Cloudflare’s ecosystem, and the combination of Cloudflare Queues, Durable Objects, KV, R2, D1, and Hyperdrive with Planetscale turned out to be a game-changer—especially for managing repeatable jobs and dead-letter queues (DLQ).
The Stack at a Glance:
Bun + Hono: Lightweight, fast runtime and web framework running on Cloudflare Workers.
Cloudflare Queues: Reliable message passing for background tasks.
Durable Objects + D1: Stateful processing with persistent SQL storage (via D1) for job metadata and retry logic.
KV & R2: Ultra-low-latency key-value cache and object storage for files.
Planetscale + Hyperdrive: MySQL-compatible database with Hyperdrive’s connection pooling and query caching, accessed through Drizzle ORM.
How It All Flows:
Request Handling: An incoming HTTP request hits the Hono app. For simple reads/writes, the app uses Drizzle ORM to query Planetscale through Hyperdrive. Hyperdrive caches frequent queries, dramatically reducing latency and database load.
Job Orchestration: For tasks that need background processing (e.g., sending emails, image processing), the app enqueues a message into Cloudflare Queue. This guarantees delivery and decouples the request from heavy work.
Durable Object Processing: A Durable Object consumes messages from the queue. It maintains job state in its attached D1 database (for persistence across failures) and uses KV for ephemeral data like rate-limit counters. If the job involves files, it streams them to/from R2.
Database Interactions: The Durable Object also updates the Planetscale database via Drizzle + Hyperdrive. Since Hyperdrive pools connections and caches read results, repeated lookups (e.g., checking user quotas) are lightning fast.
Resilience & DLQ Handling: Failed jobs are retried according to logic inside the Durable Object. If all retries fail, the job can be moved to a dead-letter queue (simulated using D1 or KV), allowing manual inspection without losing data.
Why This Rocks:
Performance: Hyperdrive’s cache reduces database round trips; KV gives microsecond access to hot data; R2 handles large files without bloating the DB.
Reliability: Queues and Durable Objects ensure exactly-once processing semantics and state persistence—perfect for financial transactions or idempotent jobs.
Developer Experience: Bun’s speed, Hono’s simplicity, and Drizzle’s type-safe SQL make development a joy. Everything deploys seamlessly with Wrangler.
Top comments (2)
Also highly recommend looking at Rivet Actors for an open-source alternative to Durable Objects + Workflows. It can be self-hosted & supports lower latency to your PlanetScale database.
This workflow works really great when you don't want to leave the Cloudflare ecosystem and you want to manage your application serverless way.