Jim Halfpenny created HDFS-17891:
------------------------------------

             Summary: HostSet.add() throws uncaught exception if DataNode 
hostname is not resolved
                 Key: HDFS-17891
                 URL: https://issues.apache.org/jira/browse/HDFS-17891
             Project: Hadoop HDFS
          Issue Type: Bug
          Components: namenode
    Affects Versions: 3.4.3
            Reporter: Jim Halfpenny


The HostSet.add() method that adds DataNode addresses to a HostSet checks to 
see if the hostname resolves. If this test fails then 
Preconditions.checkArgument() throws an uncaught IllegalArgumentException:

{{  void add(InetSocketAddress addr) {}}
{{    Preconditions.checkArgument(!addr.isUnresolved());}}
{{    addrs.put(addr.getAddress(), addr.getPort());}}

It would make more sense log the failure and return from the add() method 
rather than cause the creation of the HostSet to fail entirely. An example of 
when this issue manifests is when running `hdfs dfsadmin -report` on a cluster 
where one of the DataNode addresses cannot be resolved. The DataNode report 
fails with an IllegalArgumentException: 

{{stackable@simple-hdfs-namenode-default-0 /stackable/hadoop-3.4.0 $ hdfs 
dfsadmin -report}}
{{Configured Capacity: 2933688373248 (2.67 TB)}}
{{Present Capacity: 1299099013120 (1.18 TB)}}
{{DFS Remaining: 1299098914816 (1.18 TB)}}
{{DFS Used: 98304 (96 KB)}}
{{DFS Used%: 0.00%}}
{{Replicated Blocks:}}
{{ Under replicated blocks: 0}}
{{ Blocks with corrupt replicas: 0}}
{{ Missing blocks: 0}}
{{ Missing blocks (with replication factor 1): 0}}
{{ Low redundancy blocks with highest priority to recover: 0}}
{{ Pending deletion blocks: 0}}
{{Erasure Coded Block Groups:}}
{{ Low redundancy block groups: 0}}
{{ Block groups with corrupt internal blocks: 0}}
{{ Missing block groups: 0}}
{{ Low redundancy blocks with highest priority to recover: 0}}
{{ Pending deletion blocks: 0}}
{{-------------------------------------------------}}
{{report: java.lang.IllegalArgumentException}}

The chance of a DataNode address not resolving is much more likely when HDFS is 
being run in Kubernetes; if a DataNode pod is removed the corresponding DNS 
entry will also be removed.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to