This sounds like a possible explanation for why I keep getting recommended different AWS client libraries from the boss to "solve instability" of a service that works 99.8% except for when the S3 endpoint (backblaze) has brief mystery issues.