[
https://issues.apache.org/jira/browse/NIFI-3026?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15675359#comment-15675359
]
ASF GitHub Bot commented on NIFI-3026:
--------------------------------------
Github user ijokarumawak commented on the issue:
https://github.com/apache/nifi/pull/1208
@markap14 I checked how Apache Flink uses SiteToSiteClient from
[NiFiSink](https://github.com/apache/flink/blob/d7b59d761601baba6765bb4fc407bcd9fd6a9387/flink-streaming-connectors/flink-connector-nifi/src/test/java/org/apache/flink/streaming/connectors/nifi/examples/NiFiSinkTopologyExample.java).
That is the same pattern Storm and Spark integration does.
I agree that third parties have been using SiteToSiteClient, but I don't
think many of them re-implement SiteToSiteClient actual implementation, rather
they just use the library. When they use it just to use SiteToSite protocol,
the URL is what they configured by themselves. So it should be fine. It doesn't
break anything.
For the case if a third party actually re-implementing a SiteToSiteClient
to customize it at lower level, excuse me, the comment you concerned about was
misleading. The interface hasn't been changed at all. What has changed by this
PR is the behavior of one implementation of SiteToSiteClient, that is the NiFi
default implementation, i.e. SocketClient and HttpClient. I didn't have to
change the comment in the first place. It doesn't break any other
SiteToSiteClient implementations written by third parties outside of NiFi
codebase.
So, to address your concern, I just removed the comment from
SiteToSiteClient.java. By doing so, the file becomes identical with the
previous version. So I amended the commit to keep it intact.
> S2S initial connection behavior enhancement
> -------------------------------------------
>
> Key: NIFI-3026
> URL: https://issues.apache.org/jira/browse/NIFI-3026
> Project: Apache NiFi
> Issue Type: Improvement
> Components: Core Framework, Core UI
> Reporter: Koji Kawamura
> Assignee: Koji Kawamura
>
> s2s client behavior and initial connection improvement is needed.
> Current experience is this: I, as a client (e.g. minifi), connect to a nifi
> cluster of e.g. 10 nodes. but i need to specify 1 node URL to establish this
> connection. this node may not be available 100% and go down, in which case my
> initial connection won't work.
> Once S2S makes the first connection, it then has a list of all nodes, and can
> check their status. But first connection failure would be a concern if the
> specified URL is somehow not working. Usually for these problems, the client
> should be able to specify multiple urls (according to multiple target cluster
> nodes), comma-separated.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
