[
https://issues.apache.org/jira/browse/YARN-10775?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17550339#comment-17550339
]
fanshilun edited comment on YARN-10775 at 6/6/22 8:19 AM:
----------------------------------------------------------
[~zhengchenyu] Thank you for your answer, but I need help answering the
following questions
In Yarn-Federation
（[https://hadoop.apache.org/docs/stable/hadoop-yarn/hadoop-yarn-site/Federation.html]）
1.We can see that the client finds the RM through the router. I don't think it
needs to be adjusted in the RM. According to your design, if the number of
routers increases, all RMs need to be restarted to let the proxyserver know the
location of the router. seems unreasonable.
2.Why does the proxyserver need to know the list of routers? This seems to be
inconsistent with the design of yarn-federation？
3.For Answer 2, I mean, I think YARN-10786's solution is lightweight and
effective, your rebuttal is not very accurate.
4.For Answer 4, Why do you need a standalone proxyserver? If the user does not
enable federation, no proxyserver is needed. If the user enables federation,
can't you use the router directly? For security issues, I see that there are
already chapters in the yarn-federation design document. If AM really returns a
dangerous url, the risk should be borne by the router, not RM.And as far as the
current code implementation is concerned, there is no processing logic for
dangerous urls.
was (Author: slfan1989):
[~zhengchenyu] Thank you for your answer, but I need help answering the
following questions
In Yarn-Federation
（https://hadoop.apache.org/docs/stable/hadoop-yarn/hadoop-yarn-site/Federation.html）
1.We can see that the client finds the RM through the router. I don't think it
needs to be adjusted in the RM. According to your design, if the number of
routers increases, all RMs need to be restarted to let the proxyserver know the
location of the router. seems unreasonable.
2.Why does the proxyserver need to know the list of routers? This seems to be
inconsistent with the design of yarn-federation？
3.For Answer 2, I mean, I think YARN-10786's solution is lightweight and
effective, your rebuttal is not very accurate.
4.For Answer 4, Why do you need a standalone proxyserver? If the user does not
enable federation, no proxyserver is needed. If the user enables federation,
can't you use the router directly?
> Federation: YARN running app web can't be unable to connect, because
> AppMaster can't redirect to the right address.
> --------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-10775
> URL: https://issues.apache.org/jira/browse/YARN-10775
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: zhengchenyu
> Assignee: zhengchenyu
> Priority: Major
> Attachments: YARN-10775-design-doc.001.pdf
>
>
> I setup a yarn federation cluster, I can't connect the running app web, but
> the completed and accepted app's web works.
> So I think need two step:
> (a) make proxy server support federation.
> (b) make router support proxy server.
> Though it is a not difficult problem, but not easy to describe the problem.
> So I submit a document YARN-10775-design-doc.001.pdf to explain this.
>
> If standalone proxyserver is enable, after step (a), the problem is solved.
> If standalone proxyserver is disable, after step (a) and (b), we use router
> as web proxy server, so we hide the cluster info for client, I think it is
> reasonable.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
