Le 20/07/2019 à 19:50, Sander Klein a écrit :

I just pathed up 2.0.2 and tested it. I still get 502's but a lot less.
I'm not sure if this is because I do less request/s or I hit something
else. The show errors show:

---
[20/Jul/2019:19:34:45.629] backend cluster1-xx (#11): invalid response
    frontend cluster1 (#3), server xxx (#1), event #0, src x.x.x.x:52007
    buffer starts at 0 (including 0 out), 10809 free,
    len 5575, wraps at 16336, error at position 0
    H1 connection flags 0x00000000, H1 stream flags 0x00004094
    H1 msg state MSG_RPVER(10), H1 msg flags 0x00001404
    H1 chunk len 0 bytes, H1 body len 0 bytes :
---

---
[20/Jul/2019:19:40:32.643] backend cluster1-xx (#11): invalid response
    frontend webservices (#18), server xxx (#2), event #13, src
x:x:x:x:x:x:x:x:59724
    buffer starts at 0 (including 0 out), 16377 free,
    len 7, wraps at 16384, error at position 0
    H1 connection flags 0x00000000, H1 stream flags 0x00004094
    H1 msg state MSG_RPBEFORE(8), H1 msg flags 0x00001404
    H1 chunk len 0 bytes, H1 body len 0 bytes :

    00000  :10}]}}
---

There is of course more with the first one, but I do not want to put
that on the mailinglist. It seems like a partial response body. I can
send it to you private if you want.

Sorry, I forgot to mention, I pushed another patch that may help you. In HAProxy 2.0, it is the commit 0bf28f856 ("BUG/MINOR: mux-h1: Close server connection if input data remains in h1_detach()").

I don't know if your HAProxy already includes it or not. If not, please give it a try. If your tests were made with this last commit, it means there is a bug somewhere else.

--
Christopher Faulet

Reply via email to