Any hint about how to debug this scenario?

[Pedro Narciso García Revington]

I have  a (connect) web application that, under certain strees, accumulates connections in the CLOSE_WAIT state. 

At some point, I need to reboot the application.

I have reproduced the behavior in my local machine by running stress tests with wrk (https://github.com/wg/wrk).

Everytime I run the test —no matter if the application is running in cluster mode or not— the first connection remains in CLOSE_WAIT state

-----

$ ss -e | grep 3000 | grep CLOSE

tcp    CLOSE-WAIT 0      0         ::ffff:127.0.0.1:3000       ::ffff:127.0.0.1:41818    uid:1000 ino:332823 sk:ffff8803265e5d00 -->

-----

I have tried to stress the application with ab or vegeta, but the issue only arises with wrk. As I said, I'm facing this issue in production with normal browsers when there is a peak.

I run wrk with -4 -c400 -d30s params.

Any hint or tip? I have no idea how to debug this issue.

Thanks, 

Pedro

[Pedro Narciso García Revington]

I've discovered that the issue is related to the proxywrap module https://github.com/findhit/proxywrap/issues/15