Possible approach: when there is a new mutation and the queue is open (online) and empty, then forward that mutation. If mutation fails with network error (server unreachable), enqueue this mutation and try to forward it again. When mutation is done, try to forward next mutation in queue. Mutation is done when it succeeds or fails with graphql error. If it fails with network error, it is retried. When queue is closed, mutation that is currently being retried should be canceled.