Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
Currently, task metrics in Nemo are limited to only a few metrics, such as serialized/encoded read bytes and state transitions. This should be refactored and improved in the way so that the metrics are aggregated and managed in a single object, and we should be able to sample tasks for deriving future durations
There are a few references we can refer to:
monitoring.html
and
core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala in the Spark source code
Attachments
Issue Links
- Blocked
-
NEMO-441 Dynamic Task Sizing on Nemo
- Open
- links to