I have reviewd this patch.
I have checked the referances in Mapreduce & Hbase as well.
In Above comment, you already mentioned that MapReduce is using that DFSClient.stringifyToken.
LOG.debug("Submitting with " +
can we raise one Jira for that in MapReduce?
+1 Patch looks good to me.
We should co-ordinate to comitt both(mapred + hdfs issue) at same time to avoid the breakage in Mapreduce.
1. i did not see any big advantage by moving the random object, other than the object creations. If any big advantage then we will plan ,to move it to common utils. So, that we can change in mapred side also to use .
2. Now stringifyToken moved to DelegationTokenIdentifier, looks Mepred also has one DelegationTokenIdentifier please check. Can we think some common place for it?
We can check this nits from Mepreduce issue.