Critical Description
Large bulk imports that take more than 2 minutes to process are failing. The master ask a tablet sever to bulk import the files and times out after 2 minutes. It retries, but the retry will take over 2 minutes. Eventually it gives up and starts copying all of the data to the fail dir which takes forever. This copy is all through the master and for large data basically never completes. The bulk import threads on the tablet server probably complete eventually, so the data is still there.
The master should not time these request out.