[ https://issues.apache.org/jira/browse/HIVE-20932?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16699537#comment-16699537 ]
Gopal V commented on HIVE-20932: -------------------------------- [~bslim]: LGTM - +1 minor nit: there's a new array list allocation for each loop, which seems somewhat of a GC thrash for no good reason. Making a DruidSerdeRow class extending ArrayList<Object> would fix that & make it less functional, but more allocation friendly. > Vectorize Druid Storage Handler Reader > -------------------------------------- > > Key: HIVE-20932 > URL: https://issues.apache.org/jira/browse/HIVE-20932 > Project: Hive > Issue Type: Improvement > Components: Druid integration > Reporter: slim bouguerra > Assignee: slim bouguerra > Priority: Major > Labels: pull-request-available > Fix For: 4.0.0 > > Attachments: HIVE-20932.3.patch, HIVE-20932.4.patch, > HIVE-20932.5.patch, HIVE-20932.6.patch, HIVE-20932.7.patch, > HIVE-20932.8.patch, HIVE-20932.8.patch, HIVE-20932.8.patch, HIVE-20932.patch > > > This patch aims at adding support for vectorize read of data from Druid to > Hive. > [~t3rmin4t0r] suggested that this will improve the performance of the top > level operators that supports vectorization. > As a first cut am just adding a wrapper around the existing Record Reader to > read up to 1024 row at a time. > Future work will be to avoid going via old reader and convert straight the > Json (smile format) to Vector primitive types. -- This message was sent by Atlassian JIRA (v7.6.3#76005)