[
https://issues.apache.org/jira/browse/YARN-5525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15613387#comment-15613387
]
Xuan Gong commented on YARN-5525:
---------------------------------
[~botong]
bq. What if the cluster provider doesn't want to use the existing
AppLogAggregator at all, then they need to ask all their user to set the
customized aggregator in LogAggregationContext for all jobs. Why don't we make
the default/fallback implementation also configurable?
Sorry, I am confused. Why do we need a new implementation for AppLogAggregator
? If different applications have different requirements for log aggregation, we
could modify the logAggregationContext. Please take a look at
ContainerLogAggregationPolicy in LogAggregationContext. We set different policy
for different applications, and based on the ContainerLogAggregationPolicy, we
could do the log aggregations differently.
> Make log aggregation service class configurable
> -----------------------------------------------
>
> Key: YARN-5525
> URL: https://issues.apache.org/jira/browse/YARN-5525
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: log-aggregation
> Reporter: Giovanni Matteo Fumarola
> Assignee: Botong Huang
> Priority: Minor
> Labels: oct16-medium
> Attachments: YARN-5525.v1.patch, YARN-5525.v2.patch,
> YARN-5525.v3.patch
>
>
> Make the log aggregation class configurable and extensible, so that
> alternative log aggregation behaviors like app specific log aggregation
> directory, log aggregation format can be implemented and plugged in.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
