Siddharth Seth created TEZ-1829:
-----------------------------------
Summary: Formalize per Input/Output counters
Key: TEZ-1829
URL: https://issues.apache.org/jira/browse/TEZ-1829
Project: Apache Tez
Issue Type: Improvement
Reporter: Siddharth Seth
Per Input / Output counters provide a much better view of how different Inputs
/ Outputs in a task executed. This isn't enabled by default since there's some
work left.
Putting down some points which need to be figured out to get these enabled by
default.
- How are the counters represented. What we have today is GROUP_NAMES changing
to reflect the Input/Output. The counter names under this remains unchanged.
- How will the counters be aggregated - Is there any point in an aggregate view
of BYTES_READ across all Inputs, instead of being aggregated only on the basis
of the specific Input.
- Access from the client. Do we want to expose an interface to access Input /
Output parameters. Currently this is accessed at a task level, and users would
have to know the Input/Output name to figure out which counter to fetch
programatically. We can leave this as is, and at a later point enhance the
getCounters API with information for the IOs.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
