We had too many batch writes which caused the reads to shut down due to healthy-read-lag