[
https://issues.apache.org/jira/browse/YARN-3816?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Li Lu updated YARN-3816:
------------------------
Attachment: YARN-3816-YARN-2928-v6.patch
OK v6 version of the patch. Addressed most of Sangjin's comments and removed
some unnecessary code. Specially, something I addressed in ways other than
Sangjin's suggestions:
- I did not move the aggregation logic to app-level collector completely.
Instead, I left the code infrastructure in TimelineCollector but moved the
logic to launch the aggregation into app-level collector. In this way, we keep
the aggregation infrastructure to be a fairly general one for future collectors
(like rack level collector proposed by Vinod a while ago) but can have specific
designs for app-level aggregations.
- With regard to the result of the aggregations, I store them in the
application entity with entity id equals to the application id. The id for each
of the aggregated metric is the original metric plus the aggregation group.
Note that I think we need to keep the "aggregation group" information in the
metric id because we may have multiple types of entities all posting the same
metric name (especially if there are user-defined metrics posted by the
application itself) and we may not want to aggregate them together.
- I refactored RealTimeAggregationOperation into TimelineMetricOperations. My
intuition here is we can provide a basic framework to define operations between
timeline metrics, no matter it's an aggregation operation or accumulation
operation. Right now the input of a timeline metric operation is the incoming
metric, the existing metric, the previous state. The output should be a new
timeline metric and the side effect can be reflected on the state. In this way
we can model aggregation operations like SUM, AVG (not supported yet) and
accumulation operations like REPLACE and MAX.
- I changed the code so that we're not storing the metric aggregation
operation. I'll rebuild them for offline aggregations through a config. Will
address that in YARN-3817. Right now, this patch lives well with the new filter
mechanism.
Please do let me know if there are other concerns, thanks!
> [Aggregation] App-level aggregation and accumulation for YARN system metrics
> ----------------------------------------------------------------------------
>
> Key: YARN-3816
> URL: https://issues.apache.org/jira/browse/YARN-3816
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Reporter: Junping Du
> Assignee: Li Lu
> Labels: yarn-2928-1st-milestone
> Attachments: Application Level Aggregation of Timeline Data.pdf,
> YARN-3816-YARN-2928-v1.patch, YARN-3816-YARN-2928-v2.1.patch,
> YARN-3816-YARN-2928-v2.2.patch, YARN-3816-YARN-2928-v2.3.patch,
> YARN-3816-YARN-2928-v2.patch, YARN-3816-YARN-2928-v3.1.patch,
> YARN-3816-YARN-2928-v3.patch, YARN-3816-YARN-2928-v4.patch,
> YARN-3816-YARN-2928-v5.patch, YARN-3816-YARN-2928-v6.patch,
> YARN-3816-feature-YARN-2928.v4.1.patch, YARN-3816-poc-v1.patch,
> YARN-3816-poc-v2.patch
>
>
> We need application level aggregation of Timeline data:
> - To present end user aggregated states for each application, include:
> resource (CPU, Memory) consumption across all containers, number of
> containers launched/completed/failed, etc. We need this for apps while they
> are running as well as when they are done.
> - Also, framework specific metrics, e.g. HDFS_BYTES_READ, should be
> aggregated to show details of states in framework level.
> - Other level (Flow/User/Queue) aggregation can be more efficient to be based
> on Application-level aggregations rather than raw entity-level data as much
> less raws need to scan (with filter out non-aggregated entities, like:
> events, configurations, etc.).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
