When a web dyno for my Sheild/Private Space app was restarting, H10 - App crashed errors were recorded or a client software received an empty response when trying to connect to the app, even other web dynos were running. Why does this sometimes happen?
The delay in deregistration of the web dyno from the load balancer caused some requests resulted in the errors.
When the platform starts to restart a web dyno for a Shield/Private Space app, the platform triggers to deregister the backend instance where the dyno was running from the load balancer. Normally, deregistration completes quickly. However, there is a slight chance of a delay of deregistration which results in the request being routed to the web dyno that is no longer accepting requests.
It is currently impossible to completely avoid this, unfortunately.