[
https://issues.apache.org/jira/browse/IMPALA-12918?focusedWorklogId=999776&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-999776
]
Kunal Siyag logged work on IMPALA-12918:
----------------------------------------
Author: Kunal Siyag
Created on: 13/Jan/26 04:02
Start Date: 12/Jan/26 04:02
Worklog Time Spent: 24h
Work Description: Pushed the patch to Gerrit for code review.
Issue Time Tracking
-------------------
Worklog Id: (was: 999776)
Time Spent: 360.5h (was: 336.5h)
> Do not allow non-numeric values in Hive table stats during an alter table
> -------------------------------------------------------------------------
>
> Key: IMPALA-12918
> URL: https://issues.apache.org/jira/browse/IMPALA-12918
> Project: IMPALA
> Issue Type: Bug
> Components: Catalog
> Affects Versions: Impala 4.0.0
> Reporter: Miklos Szurap
> Assignee: Kunal Siyag
> Priority: Major
> Labels: alter, alter-table, catalog-2024, newbie, ramp-up,
> stats, validation
> Time Spent: 360.5h
> Remaining Estimate: 335h 40m
>
> Hive table properties are string in their nature, however some of them have
> special meaning and should have numeric values, like the "totalSize",
> "numRows", "rawDataSize".
> Impala currently allows these to be set to non-numeric values (including
> empty string).
> From certain applications (like from Spark) we get quite obscure
> "NumberFormatException" errors while trying to access such broken tables.
> (see SPARK-47444)
> Impala should also validate "alter table" statements and not allow
> non-numeric values in the "totalSize", "numRows", "rawDataSize" table
> properties.
> For example a query which may break the table (after it can't be read from
> Spark):
> {code}
> [impalacoordinator:21000] default> alter table t1p set
> tblproperties('numRows'='', 'STATS_GENERATED_VIA_STATS_TASK'='true');
> {code}
> Note: beeline/Hive validates alter table statements with the "numRows" and
> "rawDataSize", the "totalSize" still needs validation there too.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
