How to listen for the Hono shutdown event in middleware ?

[seepine]

I create a Redis middleware like this

export const cacheMiddleware = (opts?: RedisOptions) => {
  const cache = new Redis(opts)
  // How to close redis when Hono stop ?
  // await cache.close()

  return createMiddleware(async (c, next) => {
    c.set('cache', cache)
    await next()
  })
}

[ideacatlab]

There's no "Hono shutdown event" to hook, and that's by design — Hono is just a request handler, it doesn't own the server or the process lifecycle. Middleware runs per request, so it's the wrong place to do one-time teardown. The cleanup belongs where your resource is created (once, in the closure — which you're already doing correctly) and runs at the runtime level on shutdown.

Keep the client in the closure as you have it, but expose it so you can close it from a signal handler:

export function cacheMiddleware(opts?: RedisOptions) {
  const cache = new Redis(opts)           // created once, when middleware is registered
  const mw = createMiddleware(async (c, next) => {
    c.set('cache', cache)
    await next()
  })
  return { middleware: mw, close: () => cache.quit() }
}

Then wire the teardown to the runtime. On Node (@hono/node-server):

const { middleware, close } = cacheMiddleware()
app.use(middleware)

const server = serve({ fetch: app.fetch, port: 3000 })

function shutdown() {
  server.close(async () => {   // stop accepting connections first
    await close()              // then close Redis
    process.exit(0)
  })
}
process.on('SIGTERM', shutdown)
process.on('SIGINT', shutdown)

I tested this end to end on Node 26 with @hono/node-server: a request serves fine, and on SIGTERM the server drains, then the resource's close() runs before exit. Order matters — server.close() before closing Redis, so in-flight requests that still need the client aren't cut off.

It differs by runtime, since "shutdown" is a runtime concept, not a Hono one:

• Node / Bun: process.on('SIGTERM'|'SIGINT', ...) as above (Bun also supports process.on).
• Deno: Deno.addSignalListener("SIGTERM", ...).
• Cloudflare Workers / other serverless: there's no long-lived process to shut down — a Workers instance can be evicted at any time with no guaranteed teardown, so don't hold a Redis connection open across requests there; use a per-request client or a connection pool/HTTP-based client instead.

So: single client in the closure (✅ what you have), expose a close(), and call it from the runtime's signal handler — not from middleware.

[seepine]

Thanks, i do it now.