We see transient errors in production like "command update requires authentication: generic server error". These go away on retry. This is likely something wrong with our system.

While customers can choose the role a stream processor operators as on their database, these particular auth errors are not caused by a customer's auth configuration.

We should classify these errors as internal so they are always retried. We can specifically look for the string "requires authentication: generic server error" to distinguish this case from the the genuine "misconfigured auth role" user error.