Allow a configurable keep alive timeout in next config

[jordie23]

Describe the feature you'd like to request

NodeJS defaults keepAliveTimeout to 5 seconds. When NextJS is behind an AWS ELB/ALB, it can cause occasional 502 errors as the ELB's timeout is longer than Next's. To prevent this it should be the other way around where Next has a longer timeout than the ELB. Also need to be able to configure headersTimeout for this circumstance.

More info can be found here: https://shuheikagawa.com/blog/2019/04/25/keep-alive-timeout/

Describe the solution you'd like

Ability to set in the NextJS config a way to set the values for keepAliveTimeout and headersTimeout.

Describe alternatives you've considered

The only solution I was able to find was to modify the IdleTimeout on the AWS ELB to be 4 seconds, i.e 1 second less than node's default 5 seconds. This is a problem for any longer running requests (think 99th percentile or some intensive /api request) which will result in a 504 (gateway timeout).

A custom server could also be set up however I don't think this is a unique use case (anyone hosting on AWS behind a load balancer will have this issue).

[narze]

@jordie23 I encounter this exact issue, did you solve it somehow?

[anuragkalia]

I will also want this feature, +1

[rawpixel-vincent]

another discussions about this #19618

[rawpixel-vincent]

More information, reducing the ALB timeout to 4 seconds has a performance hit, implementing a custom server is a solution but I believe exposing this configuration with some side notes about the issue with load balancing will have a better impact on the next.js community overall.

[tabuchid]

How does anyone deploy next for anything serious outside of vercel? Between this and no way of enabling logging or integrating APM its very hard to maintain.

[danroestorf]

This is a serious issue on our application and setting the ALB timeout to 4s is not a viable option. NextJS is turning out to be a real headache in a production environment.

[jhchill666]

This was moved to a discussion, but doesn't seem to be discussed.

This is a serious problem for running NextJS in production, outside of Vercel. While Next's commercial interests may lie with Vercel, this totally invalidates using NextJS in production on AWS, as a viable option.

Considering the fix is as simple as allowing some additional config on next's httpServer, can we not get this moving at all?

[tabuchid]

Another similar discussion here with similar response #28642

[jhchill666]

I'm not too familiar with the internals of next.config.js, to provide additional configuration options to the underlying httpServer, but am happy to take this on

[tabuchid]

@timneutkens ^