[HADOOP-10131] NetWorkTopology#countNumOfAvailableNodes() is returning wrong value if excluded nodes passed are not part of the cluster tree

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.5-alpha, 3.0.0-alpha1
• Fix Version/s: 2.6.0
• Hadoop Flags: Reviewed
• Assignee: Vinayakumar B
• Reporter: Vinayakumar B
• Created: 20/Aug/13 15:21
• Updated: 12/May/16 18:27
• Resolved: 22/Sep/14 06:02

Description

I got "File /hdfs_COPYING_ could only be replicated to 0 nodes instead of minReplication (=1). There are 1 datanode(s) running and 1 node(s) are excluded in this operation." in the following case

1. 2 DNs cluster,
2. One of the datanodes was not responding from last 10 min, but about to detect as dead at NN.
3. Tried to write one file, for the block NN allocated both DNs.
4. Client While creating the pipeline took some time to detect one node failure.
5. Before client detects pipeline failure, NN side dead node was removed from cluster map.
6. Now, client has abandoned previous block and asked for new block with dead node in excluded list and got above exception even though one more node was available live.

When I dig this more, found that,
NetWorkTopology#countNumOfAvailableNodes() is not giving correct count when the excludeNodes passed from client are not part of the cluster map.

Adding to this one more case where count is wrong.
1. If there is no node present for the normalized scope in cluster.

Issue Links

• is depended upon by: HDFS-6995 Block should be placed in the client's 'rack-local' node if 'client-local' node is not available