[
https://issues.apache.org/jira/browse/HDFS-13687?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16516494#comment-16516494
]
Konstantin Shvachko commented on HDFS-13687:
--------------------------------------------
Was looking at your patch, [~csun]. The problem looks straightforward, but the
solution may be tricky here. Some thoughts.
# We cannot just call {{getServiceStatus()}} from
{{ConfiguredFailoverProxyProvider<T>}}
#* Because it requires superuser privileges on NN, which the client doesn't
generally have in case {{T = ClientProtocol}}
#* For some protocols like {{T = DatanodeProtocol}} we actually don't care if
the proxy connects to Active, because in the end DNs should report to all NNs
regardless of the state.
# If the proxy needs the HAState it should be doing something like
{{FailoverController}}. Particularly go through a factory to create a proxy,
rather than directly instantiating
{{HAServiceProtocolClientSideTranslatorPB()}}.
# Current implementation of {{ConfiguredFailoverProxyProvider}} doesn't have a
notion of dealing with NNs. It probably should, or the factory for it should.
So that we were able to add NameNode specific logic.
Anyways, this will need some thinking, I don't see a solution yet. In the mean
time, can we return back to HDFS-12976 and see what we can do there?
> ConfiguredFailoverProxyProvider could direct requests to SBN
> ------------------------------------------------------------
>
> Key: HDFS-13687
> URL: https://issues.apache.org/jira/browse/HDFS-13687
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Chao Sun
> Assignee: Chao Sun
> Priority: Minor
> Attachments: HDFS-13687.000.patch
>
>
> In case there are multiple SBNs, and {{dfs.ha.allow.stale.reads}} is set to
> true, failover could go to a SBN which then may serve read requests from
> client. This may not be the expected behavior. This issue arises when we are
> working on HDFS-12943 and HDFS-12976.
> A better approach for this could be to check {{HAServiceState}} and find out
> the active NN when performing failover. This also can reduce the # of
> failovers the client has to do in case of multiple SBNs.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
