[ https://issues.apache.org/jira/browse/HIVE-26307?focusedWorklogId=780822&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-780822 ]
ASF GitHub Bot logged work on HIVE-26307: ----------------------------------------- Author: ASF GitHub Bot Created on: 13/Jun/22 13:37 Start Date: 13/Jun/22 13:37 Worklog Time Spent: 10m Work Description: szlta commented on code in PR #3354: URL: https://github.com/apache/hive/pull/3354#discussion_r895723865 ########## iceberg/iceberg-handler/src/main/java/org/apache/iceberg/mr/mapreduce/IcebergInputFormat.java: ########## @@ -381,100 +400,67 @@ private CloseableIterable<T> newAvroIterable( Avro.ReadBuilder avroReadBuilder = Avro.read(inputFile) .project(readSchema) .split(task.start(), task.length()); + Review Comment: nit: Should we keep the original exception ("Vectorized execution is not yet supported for Iceberg avro...") here in case the inMemoryDataModel==HIVE ? In theory HiveIcebergStorageHandler should prevent such combination, just wanted to know if this was a conscious decision. Issue Time Tracking ------------------- Worklog Id: (was: 780822) Time Spent: 20m (was: 10m) > Avoid FS init in FileIO::newInputFile in vectorized Iceberg reads > ----------------------------------------------------------------- > > Key: HIVE-26307 > URL: https://issues.apache.org/jira/browse/HIVE-26307 > Project: Hive > Issue Type: Improvement > Reporter: Peter Vary > Assignee: Peter Vary > Priority: Major > Labels: pull-request-available > Time Spent: 20m > Remaining Estimate: 0h > > With vectorized Iceberg reads we are creating {{HadoopInputFile}} objects > just to store the location of the files. If we can avoid this, then we can > improve the performance, since the {{path.getFileSystem(conf)}} calls can > become costly, especially for S3 -- This message was sent by Atlassian Jira (v8.20.7#820007)