Did you try hitting each individual node in the cluster separately to see if perhaps one particular node is having that issue?