When a service is updated on a route, the change is processed as two separate events: route removal and route addition (https://github.com/openshift/router/blob/7a688b0eab5a27fe13988a21022c774cdeb964b2/pkg/router/template/router.go#L1090-L1108).
In a DCM-enabled router, route removal is handled by disabling all associated servers while the route addition cannot be done dynamically and results in the router reload. If the idle-close-on-response option is enabled, this can lead to a 503 error when a client attempts to reuse a persistent connection.
The issue occurs because idle connections are not reset immediately but are deferred until the last request is handled (as per HAProxy's idle-close-on-response behavior). This causes the old HAProxy process to remain active with all servers in maintenance mode for the affected route. As a result, any subsequent request on the persistent connection to this route receives a 503 error, while all later requests correctly reach the new service endpoint(s).

4.18+ (when TechPreview featureset is used)  

Always

1. Make sure to use the router with Dynamic Configuration Manager enabled.
2. Make sure idle-close-on-response option is set in the router's configuration.
3. Make sure to use an http client which reuses connections.
4. Send an http request to a route.
5. Change the router's "to" service to another one with a ready endpoint.
6. Send another http request to the same route.
    

503 error for the first request. All subsequent requests succeed and come from the new endpoint.

The first request should go to the old endpoint. All subsequent requests succeed and come from the new endpoint.

Potential fix: fallback to the router reload for the route update event.