[
https://issues.apache.org/jira/browse/YARN-11871?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18022673#comment-18022673
]
ASF GitHub Bot commented on YARN-11871:
---------------------------------------
liubin101 opened a new pull request, #7997:
URL: https://github.com/apache/hadoop/pull/7997
### Description of PR
Under the YARN Federation architecture, when there's only one sub-cluster,
or when multiple sub-clusters exist but a queue is configured to submit
exclusively to a specific sub-cluster, the following problem occurs: If this
target sub-cluster failover, its state will remain "non-active" until the new
active RM completes registration and sends the first heartbeat (default 30
seconds later). During this window, any client attempting to submit
applications will fail with the error: "No active SubCluster available to
submit the request" or "No positive weight found on active subclusters".
In non-YARN Federation setups, clients automatically retry during RM
failover. We expect this retry behavior to be preserved when transitioning to
the YARN Federation architecture.
### How was this patch tested?
unit test
> Make client retry when no active sub-cluster available
> ------------------------------------------------------
>
> Key: YARN-11871
> URL: https://issues.apache.org/jira/browse/YARN-11871
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: federation
> Reporter: liu bin
> Priority: Major
>
> Under the YARN Federation architecture, when there's only one sub-cluster, or
> when multiple sub-clusters exist but a queue is configured to submit
> exclusively to a specific sub-cluster, the following problem occurs: If this
> target sub-cluster failover, its state will remain "non-active" until the new
> active RM completes registration and sends the first heartbeat (default 30
> seconds later). During this window, any client attempting to submit
> applications will fail with the error: "No active SubCluster available to
> submit the request" or "No positive weight found on active subclusters".
> In non-YARN Federation setups, clients automatically retry during RM
> failover. We expect this retry behavior to be preserved when transitioning to
> the YARN Federation architecture.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
