[
https://issues.apache.org/jira/browse/HUDI-326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16971521#comment-16971521
]
Vinoth Chandar commented on HUDI-326:
-------------------------------------
If you only need just record_key, then you'd have to force the usage of
`GlobalBloomIndex` for both upserts and deletes.. @shivnarayan can add more
context..
Orthogonally, we should at some point do
https://issues.apache.org/jira/browse/HUDI-37, so this information is available
when trying to decide, whether global bloom could be used or not.. if you did
not use global bloom index while upserting, then there could be duplicates
across partitions.. and thus its not clear if the intent is to delete all the
keys across partitions.. https://issues.apache.org/jira/browse/HUDI-145 is also
an useful related ticket. Just sharing all the related tickets, in case you
were looking for something to get started; :)
True record level indexing is coming in couple releases.. but again, I expect
that to have both global and partition level flavors as well.
> Support deleting records with only record_key
> ---------------------------------------------
>
> Key: HUDI-326
> URL: https://issues.apache.org/jira/browse/HUDI-326
> Project: Apache Hudi (incubating)
> Issue Type: Improvement
> Reporter: Brandon Scheller
> Priority: Major
>
> Currently Hudi requires 3 things to issue a hard delete using
> EmptyHoodieRecordPayload. It requires (record_key, partition_key,
> precombine_key).
> This means that in many real use scenarios, you are required to issue a
> select query to find the partition_key and possibly precombine_key for a
> certain record before deleting it.
> We would like to avoid this extra step by being allowed to issue a delete
> based on only the record_key of a record.
> This means that it would blanket delete all records with that specific
> record_key across all partitions.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
