[
https://issues.apache.org/jira/browse/HDFS-15016?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16982940#comment-16982940
]
Íñigo Goiri commented on HDFS-15016:
------------------------------------
Right now, we have DataNodes that change subclusters.
The Router is just returning a random one so it can be that is returning a dead
one.
It could just do something like:
{code}
for (Entry<FederationNamespaceInfo, DatanodeInfo[]> entry :
results.entrySet()) {
FederationNamespaceInfo ns = entry.getKey();
DatanodeInfo[] result = entry.getValue();
for (DatanodeInfo node : result) {
String nodeId = node.getXferAddr();
if (!datanodesMap.containsKey(nodeId) ||
node.getLastUpdate() > datanodesMap.get(nodeId).getLastUpdate()) {
// Add the subcluster as a suffix to the network location
node.setNetworkLocation(
NodeBase.PATH_SEPARATOR_STR + ns.getNameserviceId() +
node.getNetworkLocation());
datanodesMap.put(nodeId, node);
} else {
LOG.debug("{} is in multiple subclusters", nodeId);
}
}
}
{code}
> RBF: getDatanodeReport() should return the latest update
> --------------------------------------------------------
>
> Key: HDFS-15016
> URL: https://issues.apache.org/jira/browse/HDFS-15016
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Íñigo Goiri
> Priority: Major
>
> Currently, when the Router calls getDatanodeReport() (or
> getDatanodeStorageReport()) and the DN is in multiple clusters, it just takes
> the one that comes first. It should consider the latest update.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
