[
https://issues.apache.org/jira/browse/HDFS-13749?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16605304#comment-16605304
]
Chao Sun commented on HDFS-13749:
---------------------------------
Thanks [~xkrogen]. Yes agree that it's better to put the proxy creation logic
in {{NameNodeProxiesClient}}. Will do that.
{quote}Does the proxy creation in {{NNProxyInfo}} actually set up the
connection to be able to throw an {{IOException}}? If so, I think we should
lazily instantiate later rather than in the constructor. As far as I could
tell, it doesn't actually create the connection yet, in which case the current
setup is fine. If you could help me confirm this that would be great.
{quote}
I think the connection is created only when {{Client.call}} is called (see
{{Connection#setupConnection}} and its call sites}}), so you're right, it
doesn't create connection upon creating the proxy. For now, I'll keep it in the
constructor since like you said, I don't see a big advantage to lazily
initialize it.
{quote}NNProxyInfo.serviceProto, I think this should be called serviceProxy ?
{quote}
Done.
{quote}In the previous patch, if we were unable to communicate with the NN, we
marked it as STANDBY. I think we should maintain this behavior - your patch
currently leaves the old cached state. I also think the message should be WARN,
not ERROR.
{quote}
Good catch. Done.
{quote}Rather than using a hard-coded 30 second timeout, can we reuse something
about the configured retry policy?
{quote}
In {{Connection#getConnectionId}}, if null {{RetryPolicy}} is provided, it will
use a default retry policy which will retry a number of times on a fixed
interval. For the purpose of checking service status, I'm not sure if we need
anything fancier. Thought?
I'm also not particularly happy with the hard-coded 30 sec timeout, As a
improvement, we can re-use {{ipc.client.rpc-timeout.ms}}. The same is also done
in {{NameNodeProxiesClient#createProxyWithAlignmentContext}}. What do you think?
> Use getServiceStatus to discover observer namenodes
> ---------------------------------------------------
>
> Key: HDFS-13749
> URL: https://issues.apache.org/jira/browse/HDFS-13749
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Chao Sun
> Assignee: Chao Sun
> Priority: Major
> Attachments: HDFS-13749-HDFS-12943.000.patch,
> HDFS-13749-HDFS-12943.001.patch
>
>
> In HDFS-12976 currently we discover NameNode state by calling
> {{reportBadBlocks}} as a temporary solution. Here, we'll properly implement
> this by using {{HAServiceProtocol#getServiceStatus}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
