http socket might get reset instead of normal close while handling timeout

[mariusgrigaitis]

• Version: v10.15.1
• Platform: Linux 0fe3a1d57e94 4.9.125-linuxkit deps: update openssl to 1.0.1j #1 SMP Fri Sep 7 08:20:28 UTC 2018 x86_64 GNU/Linux
• Subsystem: http,net

We were experiencing unexpected dropped connections between AWS ALB and nodejs application resulting in 502 errors from load balancer.

Managed to replicate a condition when nodejs abnormally closes socket with RST packet instead of normal FIN packets.

Here's exact conditions:

server.timeout = 10 * 1000;
server.keepAliveTimeout = 0;

We kept doing requests to the same connection every 10 seconds to trigger the condition.

Note that data was received and acknowledged, but instead of handling data or closing socket in a normal way nodejs reset the tcp connection.

[thomasjungblut]

I'm currently debugging a similar issue with carbon and AWS classic ELB.
We do set the IDLE timeout on it to 60s and our keepAliveTimeout to 65s.

[thomasjungblut]

any idea? I realize that node.js uses libuv underneath, maybe it would be useful to involve them as well?

[mariusgrigaitis]

@thomasjungblut you might also want to check #26165

Our solution was to set the headersTimeout 0

[thomasjungblut]

@mariusgrigaitis so we tried that out on top of node:dubnium-jessie-slim (currently at v10.15.3)
furthermore we have a similar issue using ALBs and websockets which give us 502s in that scenario.