We have identified a stuck instance of our API servers and restarted it. Now the performance level should be back to normal.