[
https://issues.apache.org/jira/browse/SLIDER-721?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14244148#comment-14244148
]
Steve Loughran commented on SLIDER-721:
---------------------------------------
Commentary
This is potentially a race condition.
Line in question is
{{ServiceRecord self = yarnRegistry.getSelfRegistration();}}
therefore, yarnRegistry is null. This is passed in during registry setup, which
is taking place after registering with the AM (current code is depending on the
AM app attempt ID)
Tactics
# get that registry binding in earlier
# make the web page more robust at startup...there may be other issues there
too.
# delay web site startup until the rest of the app is ready.
This condition probably exists with the other parts of the web pages and the
REST APIs.
We are currently starting the web page early so that we know the port of the
webapp. We could scan for an empty port and register that with the AM, then
tell jetty the port to use, starting it after everything else.
There's one more option: have a filter which rejects HTTP operations until the
system is live, returning some error "503 not yet live" and add a retry delay
of 5+ seconds.
> NPE in slider web view
> ----------------------
>
> Key: SLIDER-721
> URL: https://issues.apache.org/jira/browse/SLIDER-721
> Project: Slider
> Issue Type: Bug
> Components: Web & REST
> Affects Versions: Slider 0.60
> Reporter: Steve Loughran
> Fix For: Slider 0.70
>
>
> Reported NPE in {{AbstractProviderService.buildEndpointDetails}} creating web
> view.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
