[
https://issues.apache.org/jira/browse/HDFS-11036?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15879544#comment-15879544
]
Lei (Eddy) Xu commented on HDFS-11036:
--------------------------------------
Hi, [~vagarychen]
Thanks a lot for the efforts. Would you mind to address a few extra concerns:
* Could you provide tests for this patch to prove the correctness for the basic
cases?
* Would you consider to use {{ScheduledExecutorService#scheduleAtFixedRate()}}
to invoke {{ClientCleaner#run()}}?
* Can {{ClientCleaner}} be private or package-private?
* As [~anu] mentioned, it'd nice to not hold lock when iterate the map.
* {code}
} catch (InterruptedException e) {
}
{code}
Maybe {{LOG.warn()}} here?
* {code}
synchronized (openClient) {
info = openClient.get(containerName);
}
....
info = new XceiverClientWithAccessInfo(xceiverClient);
info.incrementReference();
synchronized (openClient) {
openClient.put(containerName, info);
}
{code}
I saw you are using container name as key here, would it be an issue if two
clients / pipelines access the same container at the same time?
* {code}
private static class XceiverClientWithAccessInfo {
private XceiverClient xceiverClient;
private AtomicInteger referenceCount;
{code}
Can these two fields be {{final}}?
Thanks.
> Ozone : reuse Xceiver connection
> --------------------------------
>
> Key: HDFS-11036
> URL: https://issues.apache.org/jira/browse/HDFS-11036
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Chen Liang
> Assignee: Chen Liang
> Attachments: HDFS-11036-HDFS-7240.001.patch,
> HDFS-11036-HDFS-7240.002.patch, HDFS-11036-HDFS-7240.003.patch
>
>
> Currently for every IO operation calling into XceiverClientManager will
> open/close a connection, this JIRA proposes to reuse connection to reduce
> connection setup/shutdown overhead.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
