If there is an error inserting data points, like a write timeout, we simply propagate the error back to the client. We have no error handling or retry logic in place on the server. Clients can ignore errors which means some data points do not get stored. Or they can implement some sort of retry strategy. This places more complexity on clients.
We need some retry on the server with a back off strategy as well. We can probably make a lot of the writes eventually succeed, reducing the overall number of failures. We do not want to retry indefinitely though.