[
https://issues.apache.org/jira/browse/YARN-3033?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14293945#comment-14293945
]
Sangjin Lee commented on YARN-3033:
-----------------------------------
As mentioned in YARN-2928, we are now considering different options for
starting the timeline aggregator (formerly known as the ATS writer) so as to
avoid the implications of setting up special containers and handling the
associated resource allocation and failure scenarios.
First, we are going to start with a per-node aggregator as opposed to per-app.
Second, we're also considering making it embedded within the NM as an auxiliary
service as a starting point.
The bulk of that approach has actually been done as part of YARN-3030. In this
ticket we can work on the remaining tasks, for example:
- any needed change for configuration, esp. running it inside the NM (e.g. the
number of servlet threads)
- set up a start script that starts the per-node aggregator as a standalone
daemon
- for the standalone mode, implement a service that receives requests to set up
and tear down the app-level data
> implement NM starting the ATS writer companion
> ----------------------------------------------
>
> Key: YARN-3033
> URL: https://issues.apache.org/jira/browse/YARN-3033
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Sangjin Lee
> Assignee: Sangjin Lee
>
> Per design in YARN-2928, implement node managers starting the ATS writer
> companion.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
