[
https://issues.apache.org/jira/browse/YARN-11766?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17925440#comment-17925440
]
Tao Yang commented on YARN-11766:
---------------------------------
[~cheersyang], [~prabhujoseph], could you please help to review this PR at your
convenience, Thanks a lot!
> Implement DELAYED_OR operator in the scheduling process for Rich Placement
> Constraint
> -------------------------------------------------------------------------------------
>
> Key: YARN-11766
> URL: https://issues.apache.org/jira/browse/YARN-11766
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: capacityscheduler
> Reporter: Tao Yang
> Assignee: Tao Yang
> Priority: Major
> Labels: pull-request-available
>
> DELAYED_OR operator has been defined in the [Rich Placement Constraints
> Design
> doc|https://issues.apache.org/jira/secure/attachment/12867869/YARN-6592-Rich-Placement-Constraints-Design-V1.pdf],
> and is already existing in the API. But haven't been implemented in the
> scheduling process yet.
> This ticket will implement DELAYED_OR operator in the scheduling process to
> support soft placement constraints requirements, according to the description
> in design doc:
> {code}
> DELAYED_OR: Similar to OR, but supports ordering and delays for constraints.
> Example : Try for 2 mins to place me to host n1; then try for 3 mins to place
> me on rack r1; if everything else fails, place me on any node. The following
> delay criteria can be specified to control the following:
> * Number of units of delay for each constraint in the expression.
> * Units of delay can be either missed opportunities or wall clock time.
> * Ability to reset units of delay after one of the allocations gets allocated
> (similar to today’s resetRackDelay).
> {code}
> The last delay criteria has been improved to make it usable: It's reasonable
> to reset units of delay after one of the allocations gets allocated for
> OPPORTUNITIES delay-unit, but for MILLISECONDS delay-unit, the update time
> should be reset only when the request is updated, otherwise the request with
> large size will be satisfied slowly: one by one satisfied with the specified
> delay interval. For example, if we specified the DELAY_OR constraint for a
> request with 100 containers and the delay interval is 1 minute, it is
> expected to be satisfied after 1 minute instead of 100 minutes for all the
> 100 containers.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
