[
https://issues.apache.org/jira/browse/HDFS-15032?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16991806#comment-16991806
]
Erik Krogen commented on HDFS-15032:
------------------------------------
Hey [~shv], that's a great question. In many cases, only a reference to the
proxy is kept, so it is the direct {{toString}} method of the proxy that you
see. For example, this is what a debugger stopped in
{{ObserverReadProxyProvider}} looks like without this change:
!debugger_without_tostring.png!
You can see that the proxy (which is really a combined proxy) is reporting that
it is a {{NameNodeProtocolTranslatorPB}}, because it is the {{toString()}}
method of the first proxy which is being used. This was misleading to me when I
was trying to investigate this issue, as it led me to believe a plain
{{NameNodeProtocol}} was showing up where I expected a {{BalancerProtocol}}.
However with the change, it is more obvious what is going on:
!debugger_with_tostring.png!
I see your concern about the performance, however. I've added a v003 patch
which replaces to string comparison with a call to {{Method.equals()}}, which I
confirmed internally only does a few reference equality checks:
{code}
public boolean equals(Object obj) {
if (obj != null && obj instanceof Method) {
Method other = (Method)obj;
if ((getDeclaringClass() == other.getDeclaringClass())
&& (getName() == other.getName())) {
if (!returnType.equals(other.getReturnType()))
return false;
return equalParamTypes(parameterTypes, other.parameterTypes);
}
}
return false;
}
{code}
Let me know if that addresses your concerns. If you think it's too risky for
performance, I'm fine with removing it.
> Balancer crashes when it fails to contact an unavailable NN via
> ObserverReadProxyProvider
> -----------------------------------------------------------------------------------------
>
> Key: HDFS-15032
> URL: https://issues.apache.org/jira/browse/HDFS-15032
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: balancer & mover
> Affects Versions: 2.10.0
> Reporter: Erik Krogen
> Assignee: Erik Krogen
> Priority: Major
> Attachments: HDFS-15032.000.patch, HDFS-15032.001.patch,
> HDFS-15032.002.patch, HDFS-15032.003.patch, debugger_with_tostring.png,
> debugger_without_tostring.png
>
>
> When trying to run the Balancer using ObserverReadProxyProvider (to allow it
> to read from the Observer Node as described in HDFS-14979), if one of the NNs
> isn't running, the Balancer will crash.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
