haproxy reload command does not reload config

[skjbulcher]

Running an unaltered haproxy 1.6.9 image. The config is set up to proxy requests on port 80 to various other ports on the host (I use haproxy as an http proxy for docker-served webhosts).

docker-compose.yml

haproxy:
    image: haproxy:1.6.9
    ports:
      - "80:80"
    volumes:
      - ./config/haproxy.cfg:/usr/local/etc/haproxy/haproxy.cfg

According to the documentation, I should be able to reload changes to haproxy.cfg by running the command docker kill -s HUP <my-running-haproxy>.

Actions taken

1. Edit haproxy.cfg; add forwarding for a new host, e.g. newhost
2. docker kill -s HUP proxy_haproxy_1

Result

1. docker-compose logs displays:

   haproxy_1           | <5>haproxy-systemd-wrapper: re-executing on SIGHUP.
   haproxy_1           | <7>haproxy-systemd-wrapper: executing /usr/local/sbin/haproxy -p /run/haproxy.pid -f /usr/local/etc/haproxy/haproxy.cfg -Ds -sf 7
   

2. Visiting http://newhost in a browser results in a 503 Service Unavailable error

3. Issue docker-compose restart

4. Visiting http://newhost now works

Expected Result

1. Visiting http://newhost works immediately following the execution of docker kill -s HUP proxy_haproxy_1

[yosifkit]

The problem is not the reload. but the bind mount: ./config/haproxy.cfg:/usr/local/etc/haproxy/haproxy.cfg. When you bind mount a single file this is done by inode, so when you edit it with something like vim or sed, that creates a new inode and so your local file is not the file in the container anymore. You need to mount the whole directory if you want to edit the files on the host: ./config/:/usr/local/etc/haproxy/.

[skjbulcher]

Thanks yosifkit, your explanation makes a lot of sense (and solved my problem).

[mdbr000]

@yosifkit Thanks for this info. Just wanted to point out and check that the double colon mentioned in your answer (./config/::/usr/local/etc/haproxy/) caused problems for me, i.e. invalid mount config for type "volume": invalid mount path: '.' mount path must be absolute

Once I changed to using a single colon while maintaining the full directory mount as you indicated, things worked fine.