hi,
we have (for long time) problem with TCP RST conection close between haproxy and backend (node.js)
its not about httpchk (we know its the default behaviour from 1.5 Repositories - haproxy.git/commit)
its about standard simple http GET
we have simple standard config like this
frontend example.com-in
bind 1.1.1.1:443 ssl crt example.pem
timeout client 2h
use_backend example_backend if { ssl_fc_sni example_backend }
backend example_backend
mode http
timeout connect 5m
timeout client 5m
server example.com example_server1:10002 check
our problem is
if i strike multiple requests like this
curl --location --request GET ‘https://example.com/health’
in single second
then i get the results BUT haproxy close connection with TCP RST (not TCP FIN)
there are no rate limits anywhere in path
if i strike 1 request once a time (i.e. 5 seconds)
then connection is closed with TCP FIN
it’s very similar problem reported here Why is HAProxy often sending TCP RESETs instead of FINs to close connections to its backend nodes? (#10589) · Issues · GitLab.com / GitLab Infrastructure Team / infrastructure · GitLab
sw versions
Debian9 + haproxy 1.7(distro)
Debian11 + haproxy 2.2.9(distro)
any ideas where to dig?
Marek
1 post - 1 participant
