On-marked-down shutdown-sessions not working


I’m using HAproxy to balance connections to a PostgreSQL+repmgr master/slave configuration.
I use an external-check command to verify which postgres server is the master via an SQL query.
In front of HAproxy there’s a django-based application; all the services (django application, haproxy, postgres+repmgr) are in their own docker container.
I use the following HAproxy configuration:

    log global
    mode tcp
    retries 2
    timeout client 30m
    timeout connect 4s
    timeout server 30m
    timeout check 5s
listen stats
    mode            http  
    bind            *:18080
    maxconn         10   
    timeout server  100s
    timeout client  100s
    timeout connect 100s
    timeout queue   100s 
    stats enable
    stats hide-version
    stats refresh 30s
    stats show-node
    stats uri  /proxystats
frontend pg_frontend
    mode tcp
    bind *:5432
    default_backend pg_backend
backend pg_backend
    mode tcp
    option external-check
    external-check command /scripts/check_postgresql_master.sh
    default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
    server pg-0 pg-0:5432 check
    server pg-1 pg-1:5432 check

When a PostgreSQL server goes down, HAproxy detects the new master and the new connections are routed to the new master.
If I connect to HAproxy via psql client, I can connect to new master with no problem; If I had yet a connection to database via psql client before new master promotion, and I run a query I have the following message:

template1=# select client_addr,sync_state from pg_stat_replication;
server closed the connection unexpectedly
        This probably means the server terminated abnormally
        before or while processing the request.
The connection to the server was lost. Attempting reset: Succeeded.
psql (11.9 (Debian 11.9-0+deb10u1), server 11.10)

If I re-run the query, it works with no problems.

The django application, instead, doesn’t work until the old master becomes up and running; maybe the on-marked-down directive closes only the connection from backend to remote servers? Is there a way to force the frontend to close connections when the backend switch to another server?

Sorry for the text-wall;

Thank you,