[
https://issues.apache.org/jira/browse/HDFS-14563?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16864264#comment-16864264
]
Stephen O'Donnell commented on HDFS-14563:
------------------------------------------
Looking at the source, it seems it would be feasible to take the DNS lookup out
of the write lock.
Right now, it calls:
{code}
public void refreshNodes(final Configuration conf) throws IOException {
refreshHostsReader(conf);
namesystem.writeLock();
try {
refreshDatanodes();
countSoftwareVersions();
} finally {
namesystem.writeUnlock();
}
}
{code}
The line refreshHostsReader(conf); reads the new config file and does a DNS
lookup on each entry - the write lock is not held here. Then the main work is
done here:
{code}
private void refreshDatanodes() {
final Map<String, DatanodeDescriptor> copy;
synchronized (this) {
copy = new HashMap<>(datanodeMap);
}
for (DatanodeDescriptor node : copy.values()) {
// Check if not include.
if (!hostConfigManager.isIncluded(node)) {
node.setDisallowed(true);
} else {
long maintenanceExpireTimeInMS =
hostConfigManager.getMaintenanceExpirationTimeInMS(node);
if (node.maintenanceNotExpired(maintenanceExpireTimeInMS)) {
datanodeAdminManager.startMaintenance(
node, maintenanceExpireTimeInMS);
} else if (hostConfigManager.isExcluded(node)) {
datanodeAdminManager.startDecommission(node);
} else {
datanodeAdminManager.stopMaintenance(node);
datanodeAdminManager.stopDecommission(node);
}
}
node.setUpgradeDomain(hostConfigManager.getUpgradeDomain(node));
}
}
{code}
All the isIncluded(), isExcluded() methods call node.getResolvedAddress() which
does the DNS lookup. We could probably change things to perform all the DNS
lookups outside of the write lock, and then take the lock and process the
nodes. Also change or overload isIncluded() etc to take the inetAddress rather
than the datanode descriptor.
It would not shorten the time the operation takes to run overall, but it would
move the long duration out of the write lock and avoid blocking the namenode
for the entire time.
> Enhance interface about recommissioning/decommissioning
> -------------------------------------------------------
>
> Key: HDFS-14563
> URL: https://issues.apache.org/jira/browse/HDFS-14563
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: hdfs-client, namenode
> Reporter: He Xiaoqiao
> Assignee: He Xiaoqiao
> Priority: Major
>
> In current implementation, if we need to decommissioning or recommissioning
> one datanode, the only way is add the datanode to include or exclude file
> under namenode configuration path then execute command `bin/hadoop dfsadmin
> -refreshNodes` and trigger namenode to reload include/exclude and start to
> recommissioning or decommissioning datanode.
> The shortcomings of this approach is that:
> a. namenode reload include/exclude configuration file from devices, if I/O
> load is high, handler may be blocked.
> b. namenode has to process every datnodes in include and exclude
> configurations, if there are many datanodes (very common for large cluster)
> pending to process, namenode will be hung for hundred seconds to wait
> recommision/decommision finish at the worst since holding write lock.
> I think we should expose one lightweight interface to support recommissioning
> or decommissioning single datanode, thus we can operate datanode using
> dfsadmin more smooth.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
