There are some old Jiras suggesting that "testDecommissionStatus" is flaky.
HDFS-14854 fix accidentally commented out one of the checks in TestDecommissioningStatus.testDecommissionStatus()". This Jira will restore the commented out code and adds a blocking queue to make the test case deterministic.
My intuition is that monitor task launched by AdminManager may not have enough time to act before we start verifying the status. I suggest the force the main thread to block until the node is added to the blocked node.
- is caused by
HDFS-14854 Create improved decommission monitor implementation