If the world startup time is longer than 30 seconds, that message can be returned. That can happen for very large worlds or worlds that reference lots of compendiums, or worlds undergoing a migration on startup.
As mentioned here, we’re improving the feedback and extending the time waited for the server to startup and become ready, which should remove the disruption that the error causes.
You mention that the logs show that the startup happens twice before it begins listening for clients to connect, which isn’t intended behaviour to my understanding. Could I please ask you to provide logs detailing the timestamps and confirm whether you currently see the same happening on new start-ups?
Thank you!