[
https://issues.apache.org/jira/browse/YARN-1482?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13852274#comment-13852274
]
Xuan Gong commented on YARN-1482:
---------------------------------
The approach I am using in this patch is :
1. move the webApp from Active Service to Alway-on Service
2. the first and important part of webAppProxy is AppReportFetcher which is
using ApplicationClientProtocol to fetch the applications information. Since we
already do the automatically failover now, the AppReportFetcher is changed to
get ApplicationClientProtocol from ClientRMProxy which add the automatically
failover.
Did the test on single node cluster:
a. test webProxyServer separately:
1. Have two RMs (rm1, rm2), one NM. Change rm1 to active
2. start historyServer
3. start webProxyServer separately
4. run a mapreduce job. Paste the url into the web browser and we can find the
information
5. set rm1 into standby, and rm2 into active
6. paste the same url, and we can see the same web page, and from
webProxyServer screen, it did the failover
7. set rm2 into standby. Currently, all rms are in standby state
8. paste the same url, from the webProxyServer screen, it starts to do the
failover
9. set rm1 to active, then we can see the job information web page
b. test the embed webProxyServer
1. start two rms, one nm and historyserver
2. set rm1 into active, and run a mapreduce job
3. paste the url1(which starts with the web_address of rm1) into web browser,
and it shows the web pages which contains the job information
4. paste the url2 (which starts with web_address of rm2, other parts are the
same with url1) into the browser, and we will find the same web page, from the
screen of rm2, it shows that it did the failover.
5. set rm1 into standby, now all of rms are in standby state
6. paste url1 and url2 into browser, and from the screens of rm1 and rm2, they
show that both of them are doing the failover
7. set rm2 into active. Then we will find the same web pages
> WebApplicationProxy should be always-on w.r.t HA even if it is embedded in
> the RM
> ---------------------------------------------------------------------------------
>
> Key: YARN-1482
> URL: https://issues.apache.org/jira/browse/YARN-1482
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Vinod Kumar Vavilapalli
> Assignee: Xuan Gong
>
> This way, even if an RM goes to standby mode, we can affect a redirect to the
> active. And more importantly, users will not suddenly see all their links
> stop working.
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
