You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hudi.apache.org by "sivabalan narayanan (Jira)" <ji...@apache.org> on 2022/01/17 21:02:00 UTC
[jira] [Assigned] (HUDI-3194) Fix invisible writes(commits) during compaction (HoodieParquetRealtimeInputFormat)
[ https://issues.apache.org/jira/browse/HUDI-3194?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
sivabalan narayanan reassigned HUDI-3194:
-----------------------------------------
Assignee: sivabalan narayanan
> Fix invisible writes(commits) during compaction (HoodieParquetRealtimeInputFormat)
> ----------------------------------------------------------------------------------
>
> Key: HUDI-3194
> URL: https://issues.apache.org/jira/browse/HUDI-3194
> Project: Apache Hudi
> Issue Type: Bug
> Components: compaction, writer-core
> Reporter: Yuwei Xiao
> Assignee: sivabalan narayanan
> Priority: Critical
> Labels: pull-request-available
> Fix For: 0.11.0
>
>
> Suppose a compaction (with instant A) is going on, all writes related with the compaction (i.e., touch the file groups that are under compaction) will end up with timestamp A.
> For current `HoodieParquetRealtimeInputFormat` implementation, even the writes complete, the records are invisible until the compaction complete.
> The following pseudocode could reproduce the case
> ```
> write 200 records and complete
> scheduleCompaction
> write 200 records and complete
> read the table and only get 200 records
> ```
> Note, the Spark read path is correct and will cover the corner cases during compaction. But the hive path (also presto) is wrong.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)