[
https://issues.apache.org/jira/browse/YARN-3816?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14654298#comment-14654298
]
Junping Du commented on YARN-3816:
----------------------------------
Sorry for coming late on this. Thanks [~sjlee0], [~vrushalic] and [~gtCarrera9]
for review and comments!
bq. I propose to introduce the second dimension to the metrics explicitly. This
second dimension nearly maps to "toAggregate" (and/or the REP/SUM distinction)
in your patch. But I think it's probably better to introduce the metric types
explicitly as another enum or by subclassing TimelineMetric. Let me know what
you think.
Do you suggest to use gauge and counter type to replace "toAggregate"? But no
matter counter or gauge type of metrics, we may need to do aggregation. e.g.
CPU usage as guage, or map task number (launched, failed, etc.) as counter
(assume value is tack-on instead of accumulated). The idea to involve
"toAggregate" in metric is for client to indicate if this metric value should
be added/aggregated with other values or is a final value. If a client put a
metrics value that is already aggregated (like HDFS bytes written/read),
collector won't apply any aggregation logic on it.
bq. I'm still very confused by the usage of the word "aggregate". In this
patch, "aggregate" really means accumulating values of a metric along the time
dimension, which is completely different than the notion of aggregation we have
used all along. The aggregation has always been about rolling up values from
children to parents. Can we choose a different word to describe this aspect of
accumulating values along the time dimension, and avoid using "aggregation" for
this? "Accumulate"? "Cumulative"? Any suggestion?
Actually, v2 patch has both. In TimelineCollector, AggregatedMetrics mean
rolling up values from children to parents while AggregatedArea means
accumulating aggregated values of a metric along the time dimension. It may not
be necessary to separate calculating AggregatedArea out as a separated method.
Isn't it? It is a bit rush of naming for poc but we can have some better one
later.
bq. For example, consider HDFS bytes written. The time accumulation is already
built into it (see (1)). If you further accumulate this along the time
dimension, it becomes quadratic (doubly integrated) in time. I don't see how
that can be useful.
You are right. For some cases as you mentioned here, time accumulation is not
very useful. So beside "toAggregate", we may also need another flag (like:
"toAccumulate") to indicate if metric value need to be accumulated along the
time dimension? As we don't have assumption that all counters are already
accumulated over time or not, the client has flexibility to put accumulated or
tack-on values for counter. Thoughts?
bq. I think it would be OK to do this and not the average/max of the previous
discussion. I'd like to hear what others think about this.
Either way should work as we have area value at all timestamps, we can
recalculate average/max later if necessary. Would like to hear others' comments
too.
bq. Can we introduce a configuration that disables this time accumulation
feature? As we discussed, some may not want to have this feature enabled and
are perfectly happy with simple aggregation (from children to parents). It
would be good to isolate this part and be able to enable/disable it.
We surely can disable accumulation in system level. We also can disable
accumulation in metrics level (like proposed above) even accumulation is
enabled in system level.
bq. For timeseries, we need to decide what aggregation means. One option is
that we could normalize the values to a minute level granularity. For example,
add up values per min across each time. So anything that occurred within a
minute will be assigned to the top of that minute: eg if something happening at
2 min 10 seconds is considered to have occurred at 2 min. That way we can sum
up across flows/users/runs etc.
The other option is we only record /store accumulated values at different
timestamp and we do delta calculation later if necessary. This can address more
time granularity as query could apply on different granularity.
> [Aggregation] App-level Aggregation 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: Junping Du
> Attachments: Application Level Aggregation of Timeline Data.pdf,
> 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)
