[
https://issues.apache.org/jira/browse/IMPALA-8409?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16832672#comment-16832672
]
ASF subversion and git services commented on IMPALA-8409:
---------------------------------------------------------
Commit c2516d220da8e532b6ebdb6f3a12e7ad97c4f597 in impala's branch
refs/heads/master from Csaba Ringhofer
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=c2516d2 ]
IMPALA-8409: Fix row-size for STRING columns with unknown stats
Explain returned row-size=11B for STRING columns without statistics.
The issue was caused by adding -1 (meaning unknown) to the 12 byte
slot size (sizeof(StringValue)). The code in TupleDescriptor.java
tried to handle this by checking if the size is -1, but it was
already 11 at this point.
There is more potential for cleanup, but I wanted to keep this
change minimal.
Testing:
- revived some tests in CatalogTest.java that were removed
in 2013 due to flakiness
- added an EE test that checks row size with and without stats
- fixed a similar test, test_explain_validate_cardinality_estimates
(the format of the line it looks for has changed, which lead to
skipping the actual verification and accepting everything)
- ran core FE and EE tests
Change-Id: I866acf10b2c011a735dee019f4bc29358f2ec4e5
Reviewed-on: http://gerrit.cloudera.org:8080/13190
Reviewed-by: Impala Public Jenkins <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> STRINGs without stats have too low row-size in explain plan
> -----------------------------------------------------------
>
> Key: IMPALA-8409
> URL: https://issues.apache.org/jira/browse/IMPALA-8409
> Project: IMPALA
> Issue Type: Bug
> Components: Frontend
> Affects Versions: Impala 3.2.0
> Reporter: Csaba Ringhofer
> Assignee: Csaba Ringhofer
> Priority: Minor
> Labels: explain, statistics
>
> STRING columns without avg_size statistic are calculated into the row-size as
> 11 bytes, while they take 12 bytes in the tuple (+ more somewhere in the
> memory if they are not empty). The issue is caused by adding -1 (meaning
> unknown) to the 12 byte slot size.
> I think that this doesn't cause problems, as the estimation is probably way
> off without statistics anyway, but row-size >= tuple size seems like a
> meaningful invariant that we shouldn't break.
> Reproduce:
> {code}
> create table test_row_size (s string);
> explain select * from test_row_size;
> Result:
> ...
> WARNING: The following tables are missing relevant table and/or column
> statistics.
> default.test_row_size
> ...
> 00:SCAN HDFS [default.test_row_size]
> partitions=1/1 files=0 size=0B
> row-size=11B cardinality=0
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
