[
https://issues.apache.org/jira/browse/FLINK-1015?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14057290#comment-14057290
]
Fabian Hueske commented on FLINK-1015:
--------------------------------------
The execution strategies for the minimum reducers look good to me.
Your program contains two reduce operators that call the
{{MinSimilarityReducer}}. The first reduce is not grouped, the second one is.
The first (non-grouped) reducer uses a "Reduce All" (ID 15) for its combiner
and a "Reduce All" (ID 14) for the reducer.
The second (grouped) reducer uses a "Sorted Combine/Reduce" (ID 10) for its
combiner and a "Sorted Reduce" (ID 9) for the reducer.
For easier plan debugging, you give names to your operators like this
{{data.reduce(new MyReducer()).name("My Red 1");}}.
This makes it easier to find the operators in the plan.
> Wrong execution strategy
> ------------------------
>
> Key: FLINK-1015
> URL: https://issues.apache.org/jira/browse/FLINK-1015
> Project: Flink
> Issue Type: Bug
> Reporter: Bastian Köcher
> Attachments: impro3-ss14-stratosphere.zip
>
>
> We have a min reducer and it's taking really long. It's look like the reducer
> is sorted in each iteration, but we don't need any sorting. We think that
> this is the false execution plan. Here is the printed execution plan:
> http://pastebin.com/cBXsBt6q
--
This message was sent by Atlassian JIRA
(v6.2#6252)
