Currently, many of our background hooks make new connections to the cluster and, if this fails, do some checks to see if the error is transient. If the error does not pass the transient checks, then we re-throw the error. However, the logging from this is not particularly informative - it doesn't even include the error code. This can make it very difficult to understand what happened in the case of a failure because all you have is a generic "Connection refused" error message.

We should add better handling to the errors which we do not consider transient so that we at least know the error code that the connection failed with.