Start env

# docker-compose up -d

Start checking

# ./check.sh

Simulate failure

# docker-compose stop web1

Nginx will passively monitor backend health as requests are made. By default, proxy_connect_timeout is 60s and fail_timeout on the server is 10s. This means that a connection is stalled for 60s before nginx determines the backend is dead. After 10s, the backend is re-enabled and the next request will again take up to 60s before the backend is marked dead once more. And so on.

By keeping proxy_connect_timeout low, and fail_timeout high, this effect can be minimized. When set to respectively 2s and 60s, the backend will be declared dead after 2 seconds, and the will not be checked again for another 60s. At that point, the next request is again stalled 2s until timeout.

If you change the loadbalancer config, you can restart it with following command:

# docker-compose restart loadbalancer

Note that both web1 and web2 containers need to be running, or the loadbalancer will fail to start due to DNS resolution problems.