HAProxy randomly forwards Connection to only one of two Backend Servers (Roundrobin / Healthcheck is normal). Only starts loadbalancing when "empty" Server is disabled and re-enabled


I currently have the problem that HAProxy randomly forwards connections to only one of two backend servers.
Both servers pass the healthcheck and are marked as online.
The stick-table is also empty when the error occurs.

When the error occurs, HAProxy forwards all connections to one server until you disable and re-enable the “empty” server in HAProxy. After that HAProxy starts to distribute connections normally.

I am currently using HAProxy version 2.2.9-2+deb11u5 2023/04/10 on debian 11.7.

Has anyone ever had this problem? Unfortunately I could not find any forum post / github issue, with a similar error.

Thanks in advance!

Here is my backend config:

backend bk_rdp_2022_store
  mode tcp
  balance roundrobin
  # Options
  timeout server 4h
  timeout connect 4s
  option redispatch
  option tcpka
  option tcplog
  stick-table type string len 32 size 20k expire 12h store conn_cur,conn_rate(5m) peers loadbalancer_replication
  stick on src
  server tsestore01 tsestore01:3389 weight 10 check inter 2s rise 2 fall 3
  server tsestore02 tsestore02:3389 weight 10 check inter 2s rise 2 fall 3