[ https://issues.apache.org/jira/browse/TEZ-1608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14146038#comment-14146038 ]
Krisztian Horvath commented on TEZ-1608: ---------------------------------------- I received few improvements which I'm going to apply: Also, for the topK, can the sum task maintain a local top K and output only that much and the writer can pick the global topK from the local topKs. Would reduce the data transfer quite a bit. Then we may be able to use an UnorderedKVEdge instead of an OrderedPartitionedKVEdge? That will avoid the need to sort at the output and merge sort at the input. > TopK example > ------------ > > Key: TEZ-1608 > URL: https://issues.apache.org/jira/browse/TEZ-1608 > Project: Apache Tez > Issue Type: Sub-task > Affects Versions: 0.5.0 > Reporter: Janos Matyas > Attachments: TEZ-1608-1.patch > > > The goal of this sample is to find the topK elements of a dataset, while > guiding through the basics of Tez (DAG creation, tokenizers, custom > comparators and parallelism). > An example use case for top K: > Given a large data set in CSV format of user comments on a site listed as: > userid,postid,commentid,comment,timestamp and we are looking for the top K > commenter or the posts with the most comment. -- This message was sent by Atlassian JIRA (v6.3.4#6332)