[
https://issues.apache.org/jira/browse/YARN-10386?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17185790#comment-17185790
]
Gergely Pollak commented on YARN-10386:
---------------------------------------
I'm not sure I understand your defaultQueue comment properly, but on the
backend %default means root.default initially and if you user a
VariableUpdateAction, it will change it for default placement targets and
default fallbacks as well.
Why is the "defaultQueue" property a separate option not a policy? It is just a
similar action to anything, and it cannot be added as an extra to any rule. I
mean you cannot have a policy AND a change default queue, also default queue
change is for the whole application submission, not only for the given rule.
I think we are overusing the defaultQueue in the schema, we should use other
names like, changeDefaultQueue, placeToDefaultQueue, fallbackToDefaultQueue or
something along these lines to make sure it is clear what it will do.
I think we have too many options under the properties and it can easily lead to
incorrect configuration, I think we should move the nesterUserRule and the
(change)defaultQueue into the policies. Nested user rules are just complex
target paths, like root.%primary_group.%user, but nothing else, so I don't
think we should move them to a separate property (I don't even think we need to
do a separate policy for them, since we can just use regular queue placement).
The nested rules in FS always map to something like root.some.path.XXXX.%user,
and the rule decides the XXXX part, if create is not allowed and the queue does
not exist then it moves onto the next. (I know it also supports default, but
it's like place to %default) So all this can be easily mapped to regular queue
placement with rules like: root.%primary_group.%user on fallback skip.
I would also add a "custom" to the type, since matcher also can be custom, not
only rules.
> Create new JSON schema for Placement Rules
> ------------------------------------------
>
> Key: YARN-10386
> URL: https://issues.apache.org/jira/browse/YARN-10386
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: capacity scheduler, capacityscheduler
> Reporter: Peter Bacsko
> Assignee: Peter Bacsko
> Priority: Major
> Attachments: MappingRulesDescription_v1.json, YARN-10386-001.patch,
> YARN-10386-002.patch, YARN-10386-003.patch, YARN-10386-004.patch,
> YARN-10386-005.patch, YARN-10386-006.patch
>
>
> Tasks in this JIRA:
> # Create new JSON schema
> # Add Maven plugin which generates Java POJOs based on the schema
> # Add helper class which essentially does the same as #2 (for dev purposes)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
