Although the block-level logging in namenode is useful for debugging, it can add a significant overhead to busy hdfs clusters since they are done while the namespace write lock is held. One example is shown in
HDFS-4075. In this example, the write lock was held for 5 minutes while logging 11 million log messages for 5.5 million block invalidation events.
It will be useful if we have an option to disable these block-level log messages and keep other state change messages going. If others feel that they can turned into DEBUG (with addition of isDebugEnabled() checks), that may also work too, but there might be people depending on the messages.