If log saver has a non-transient error (such as hdfs being down) while saving attempting to save log files, it spams the master logs pretty darn hard.
In LogWriter.java, there's this message logged with a stack trace:
"Caught exception during save, will try again"
It spams about a thousand times per second, and it fills up the logs at about 100MB per minute.
At the minimum, we should sleep for some short duration before retrying. Or have a back-off retry mechanism, perhaps with an upper limit on the retry time.