Hi Tongqing,
No, we do not currently support per row TTL for any datasets. TTL can only be set at the dataset level.
In the meantime, you could implement your own version of KeyValueTable, which, in addition to storing the value in a column, would store an additional column (say "ttl"), with the TTL to apply for a given row ("key" + "value"). When reading data out, you could consult the stored TTL to see whether or not the given value should be returned. This would filter "expired" data on read, but you would still need to implement a background job to actually cleanup the TTL'd data. You could do this by setting up a CDAP workflow that runs a map reduce job periodically to scan the full table and delete any data that has exceeded its configured TTL.
Gary