I don’t think there exists well-researched Fly-specific guidance. I’d imagine, the usual AWS/Azure/GCP connectivity best practices apply just the same for Fly apps and machines, since Fly, out of necessity, mucks mercilessly with ingress (60s timeouts for TCP, reduced MTU for UDP), not egress (at least not mercilessly).
PS You may find useful pointers if you search for threads on AWS/RDS, like in this recent one: Fly.io + RDS & Elasticache