A recent change to delay the exit of the checkpoint thread until the exit of the flush thread created a situation where the checkpoint thread keeps running forever.