tcp connections taking too long to close

Bug Description

We recently migrated our http client from request-promise-native to undici on our GraphQL server. We have observed significant performance gains 🎉.

• Our GraphQL server connects through upstream micro-services through HTTP connection.
• After this migration we observed that connections between GraphQL and upstream services are not draining on regular interval.
• Upstream services are deployed using blue green deployments.
• Even after we deploy new stack of upstream service, there are connection open on older stack for more than 1.5 hrs.

Is there a config to specify to drain the TCP connections on certain intervals

Additional context

Sample Code-

import {IncomingHttpHeaders} from 'http'
import {request} from 'undici'

export const nativePostRequest = async (
  url: string,
  headers: IncomingHttpHeaders,
  body: any,
  extra: string
) => {
  const bodyString = JSON.stringify(body)
  return await request(url, {
    method: 'POST',
    headers: headers,
    body: bodyString,
    bodyTimeout: config.httpTimeout
  })
}

Start with the sample request(url, ...) entry point and investigate how Undici manages HTTP/TCP connection lifetimes after requests complete. Check whether the client exposes a documented drain or interval setting, and verify any finding against the blue-green deployment scenario described. Done means identifying a supported configuration or clearly documenting the limitation and expected connection behavior.