[ https://issues.apache.org/jira/browse/HIVE-26133?focusedWorklogId=755751&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-755751 ]
ASF GitHub Bot logged work on HIVE-26133: ----------------------------------------- Author: ASF GitHub Bot Created on: 12/Apr/22 13:34 Start Date: 12/Apr/22 13:34 Worklog Time Spent: 10m Work Description: marton-bod commented on code in PR #3202: URL: https://github.com/apache/hive/pull/3202#discussion_r848443468 ########## iceberg/iceberg-handler/src/main/java/org/apache/iceberg/mr/hive/HiveIcebergStorageHandler.java: ########## @@ -460,6 +461,13 @@ public void validateSinkDesc(FileSinkDesc sinkDesc) throws SemanticException { if (IcebergTableUtil.isBucketed(table)) { throw new SemanticException("Cannot perform insert overwrite query on bucket partitioned Iceberg table."); } + if (table.currentSnapshot() != null) { + if (table.currentSnapshot().allManifests().parallelStream().map(ManifestFile::partitionSpecId) + .filter(id -> id < table.spec().specId()).findAny().isPresent()) { + throw new SemanticException( + "Cannot perform insert overwrite query on Iceberg table where partition evolution happened."); Review Comment: Can you please add a note that the table must be rewritten according to the last spec in order to be able to start using IOW again? IOW can be an unsafe operation when there are different data files written out according to different partition specs Issue Time Tracking ------------------- Worklog Id: (was: 755751) Time Spent: 0.5h (was: 20m) > Insert overwrite on Iceberg tables can result in duplicate entries after > partition evolution > -------------------------------------------------------------------------------------------- > > Key: HIVE-26133 > URL: https://issues.apache.org/jira/browse/HIVE-26133 > Project: Hive > Issue Type: Improvement > Reporter: László Pintér > Assignee: László Pintér > Priority: Major > Labels: pull-request-available > Time Spent: 0.5h > Remaining Estimate: 0h > > Insert overwrite commands in Hive only rewrite partitions affected by the > query. > If we write out a record with specA (e.g. day(ts)), resulting in a datafile: > "/tableRoot/data/ts_day="2020-10-24"/ffffgggg.orc > If you then change to specB (e.g. day(ts), name), the same record would go to > a different partition: > "/tableRoot/data/ts_day="2020-10-24"/name="Mike"/ffffgggg.orc > If you then want to overwrite the table with itself, it will detect these two > records to belong to different partitions (as they do), and therefore does > not overwrite the original record with the new one, resulting in duplicate > entries. > {code:java} > create table testice1000 (a int, b string) stored by iceberg stored as orc > location 'file:/tmp/testice1000'; > insert into testice1000 values (11, 'ddd'), (22, 'ttt'); > alter table testice1000 set partition spec(truncate(2, b)); > insert into testice1000 values (33, 'rrfdfdf'); > insert overwrite table testice1000 select * from testice1000; > ------------------------------+ > testice1000.a testice1000.b > ------------------------------+ > 11 ddd > 11 ddd > 22 ttt > 22 ttt > 33 rrfdfdf > ------------------------------+ > {code} -- This message was sent by Atlassian Jira (v8.20.1#820001)