Rolling out updates in an A/B test slowly is the only way to reduce the occurrence of such issues _significantly_. There's no other way, literally, nothing.