Initial request take long time

[Kaniusiek]

Summary

Hey guys,

App deployed to self-hosted windows server

I have an application deployed to server using 'standalone' output. Application has like 4 pages, mostly done on client-side. I also have some Route Handlers/Api Routes which points our .NET api (so external api). Locally Next.js works very slow but after deployment I see that it works pretty good but! When I open page I have some requests done of course via SWR. 3 dictionaries (extra small) and one for search grid which is actually with pagination so also pretty small. Rest are regarding prefetching pages propably + current page (_rsc=xxx)

But when I start refreshing it starts working fine BUT only for like 5 seconds. After 5-10 seconds idle I'm getting again 1-2 seconds.

Problem is not with external api - when I try to use it for these endpoints, it's always between 50-250ms. I also tried running for example one of the dictionaries without UI so just /api/,,, endpoint and still the same. ON PROD. Question is what can be wrong? I search for like two days, reddit, discord channel and nothing.

Additional information

NextJs 15.5.4 - App Router

Example

No response

[marcusvramos]

Hi! This is a common issue with Next.js standalone mode. Here are a few things to check:

Possible Causes

1. Cold Start Issue

   • First request always takes longer due to server initialization
   • Next.js needs to compile and cache the initial pages

2. Windows Server Specifics

   • Make sure Node.js has proper permissions
   • Check if Windows Defender is scanning the files on first run

Solutions

Try adding these to your next.config.js:

module.exports = {
  experimental: {
    optimizeCss: true,
    optimizePackageImports: ['your-packages']
  },
  // Warm up the server
  onDemandEntries: {
    maxInactiveAge: 60 * 1000,
    pagesBufferLength: 5,
  }
}

Also, you can implement a health check endpoint that keeps the server warm.

Let me know if this helps!

[Kaniusiek]

Hey, sorry but it's not working. I did some more investigation on it. I've created simple API Route without any fetching from external, it's just returning message + status 200. I also used newest canary 15.6.0-canary.45 and nodejs 24.9.0 - locally and on windows vm.

Here are the results:

1. Local build + start

2. Prod server using localhost running url

3. Actual domain (I have reverse proxy in IIS of course from localhost to actual domain) - first requests are like spam Refresh in browser but still 5x or more slower and last requests are coming from my test-case, refresh then wait 5-10 seconds then refresh again

[icyJoseph]

And once you fall into the 1s range, it doesn't ever go back down? I think some data revalidation might be at play here...

Btw, do you fetch from your own Next.js endpoints into your server components? THat's an extra HTTP round-trip that's bound to add latency.

[icyJoseph]

Right, if the issue is Windows Server VM/Azure related,I don't know if I have the time or expertise to deep dive 🤔

I wonder if there's GC going on though, or in general, just memory related. I think you can setup your endpoint to return some data about memory usage, like https://nodejs.org/api/v8.html#v8getheapstatistics or https://nodejs.org/api/process.html#processmemoryusage

[Kaniusiek]

I hope you that someone will have some time :D If someone looking this discussion and have windows server VM please look :)

I never used these functions for statistics, can you tell me how to do it? I mean how my route should look like. Then I will re-deploy to VM and we will be able to see what's going on. Thanks

[icyJoseph]

Roughly speaking, since these are just objects:

// app/api/health/route.ts
import v8 from 'node:v8'

export function GET() {
    const heapStats = v8.getHeapStatistics()
    const memory = process.memoryUsage()
    return NextResponse.json({ heapStats, memory})
}

Something like that

[Kaniusiek]

Hey, just for descendants :) I resolved the issue with my teammates after days of research and trying different approaches. Final result - IIS URL Rewrite (proxy) on Windows Server. It seems like microsoft's module for proxy works in this way or nextjs cannot properly go through, hard to say. First I tried with setting keepAliveTimeout to some bigger value like 60_000. It started to work a little bit better but wasn't what I want. The solution was to change rewrite rules, not even the pattern but localhost - for example if you have localhost:3000 then it needs to be 127.0.0.1:3000 - seems like URL Rewrite module (proxy) has different steps when going through it till catch your Nextjs app port, hard to say :D So at the end we have different rewrite url + keepAliveTimeout as helper.

[Kaniusiek]

Btw, is it normal that when I try to run autocannon or artillery, I got slow results? I mean most of my route handlers (pretty simple, fetching from external api which also works great) are just slow and I'm getting low request/per second result and latency is very big. Maybe I'm checking the load in wrong way?