[ https://issues.apache.org/jira/browse/HIVE-21329?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16778975#comment-16778975 ]
Gopal V commented on HIVE-21329: -------------------------------- LGTM - +1 {code} 2019-02-27 01:35:38,588 [INFO] [TezChild] |writers.UnorderedPartitionedKVWriter|: Reducer_2: numBuffers=2, sizePerBuffer=524288, skipBuffers=true, numPartitions=1, availableMemory=1048576, maxSingleBufferSizeBytes=2147483647, pipelinedShuffle=false, isFinalMergeEnabled=false, numPartitions=1, reportPartitionStats=ENABLED 2019-02-27 01:35:38,592 [INFO] [TezChild] |tez.MapRecordProcessor|: Main input name is Map 1 {code} The allocation does show up in the timings in quick tests. {code} 0: jdbc:hive2://localhost:10013/> set hive.tez.unordered.output.buffer.size.mb=1024; 0: jdbc:hive2://localhost:10013/> select count(1) from store_sales order by 1; INFO : Completed executing command(queryId=gopal_20190227015023_75eeb0ec-edd2-4248-8122-a3147c0bb502); Time taken: 5.436 seconds {code} with -1 {code} INFO : Completed executing command(queryId=gopal_20190227015049_2068d1e9-b9ac-463a-8bdb-738966a480a7); Time taken: 3.645 seconds {code} > Custom Tez runtime unordered output buffer size depending on operator pipeline > ------------------------------------------------------------------------------ > > Key: HIVE-21329 > URL: https://issues.apache.org/jira/browse/HIVE-21329 > Project: Hive > Issue Type: Improvement > Components: Tez > Reporter: Jesus Camacho Rodriguez > Assignee: Jesus Camacho Rodriguez > Priority: Major > Attachments: HIVE-21329.patch > > > For instance, if we have a reduce sink operator with no keys followed by a > Group By (merge partial), we can decrease the output buffer size since we > will only produce a single row. -- This message was sent by Atlassian JIRA (v7.6.3#76005)