Fix container HEALTHCHECK when port 0 is used in confguration

[josecelano]

Context

We have a HEALTHCHECK instruction in the Containerfile:

HEALTHCHECK --interval=5s --timeout=5s --start-period=3s --retries=3 \  
  CMD /usr/bin/http_health_check http://localhost:${HEALTH_CHECK_API_PORT}/health_check \
    || exit 1

The implementation is here. It checks that all services are running:

• API
• 1 or more HTTP trackers
• 1 or more UDP trackers

For the API, for example, it tries to connect to the API healthcheck endpoint.

async fn api_health_check(config: &HttpApi) -> Option<Json<Report>> {
    // todo: when port 0 is specified in the configuration get the port from the
    // running service, after starting it as we do for testing with ephemeral
    // configurations.

    if config.enabled {
        let addr: SocketAddr = config.bind_address.parse().expect("invalid socket address for API");

        if addr.port() != UNKNOWN_PORT {
            let health_check_url = format!("http://{addr}/health_check");

            if !get_req_is_ok(&health_check_url).await {
                return Some(responses::error(format!(
                    "API is not healthy. Health check endpoint: {health_check_url}"
                )));
            }
        }
    }

    None
}

In the API configuration (Config.toml):

[http_api]
bind_address = "127.0.0.1:1212"
enabled = true
#...

you can specify the port = like this:

[http_api]
bind_address = "127.0.0.1:0"
enabled = true
#...

In that case, the application will assign a free port.

Problem

The healthcheck endpoint relies on static configuration. If you use port 0 the healthcheck skippies that service. We need to pass the list of actual ports to the "healthchecker" instead of the configuration.

Solution

When we run the services we have to send back the port used to the main app. And the main app has to inject that configuration (list of services and ports) to the "healthchecker".

Relates To

• Refactor: move service launchers to test code #274
• Application should abort on panic in any of the threads #213
• Binding to port 0 (uses a random unused port) works, but does not inform what port it bound to #186

[josecelano]

Hi @da2ce7 I think you started refactoring the healthcheck API by using closures instead of injecting the whole configuration in the handler. Could you maybe open a PR of that part or share it so I can continue working on this.

[josecelano]

Fixed via #623