You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@arrow.apache.org by "Wes McKinney (JIRA)" <ji...@apache.org> on 2017/04/13 02:53:41 UTC
[jira] [Comment Edited] (ARROW-785) possible issue on writing
parquet via pyarrow, subsequently read in Hive
[ https://issues.apache.org/jira/browse/ARROW-785?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15967036#comment-15967036 ]
Wes McKinney edited comment on ARROW-785 at 4/13/17 2:52 AM:
-------------------------------------------------------------
I tried to reproduce this issue with Impala on Arrow / Parquet master branches. I put the file in a temporary directory then ran
{code}
CREATE EXTERNAL TABLE __ibis_tmp.`__ibis_tmp_57ccb655a5b1425fbc99ea30054c6c60`
LIKE PARQUET '/tmp/test-parquet-binary/0.parq'
STORED AS PARQUET
LOCATION '/tmp/test-parquet-binary'
{code}
The resulting table, with schema inferred from the Parquet file, is:
{code}
describe __ibis_tmp.`__ibis_tmp_57ccb655a5b1425fbc99ea30054c6c60`
Out[30]:
[('year', 'bigint', 'Inferred from Parquet file.'),
('word', 'string', 'Inferred from Parquet file.')]
{code}
and
{code}
year word
0 2017 Word 1
1 2018 Word 2
{code}
string in Impala is a plain BYTE_ARRAY aka Binary. The Arrow table was
{code}
pyarrow.Table
YEAR: int64
WORD: binary
{code}
However, parquet-tool cat from parquet-mr 1.9.0 gives:
{code}
$ java -jar target/parquet-tools-1.9.0.jar cat test.parq
YEAR = 2017
WORD = V29yZCAx
YEAR = 2018
WORD = V29yZCAy
{code}
This suggests there's something wrong with the file metadata is Impala is able to read the file OK. I'm looking more closely into it
was (Author: wesmckinn):
I tried to reproduce this issue with Impala on Arrow / Parquet master branches. I put the file in a temporary directory then ran
{code}
CREATE EXTERNAL TABLE __ibis_tmp.`__ibis_tmp_57ccb655a5b1425fbc99ea30054c6c60`
LIKE PARQUET '/tmp/test-parquet-binary/0.parq'
STORED AS PARQUET
LOCATION '/tmp/test-parquet-binary'
{code}
The resulting table, with schema inferred from the Parquet file, is:
{code}
describe __ibis_tmp.`__ibis_tmp_57ccb655a5b1425fbc99ea30054c6c60`
Out[30]:
[('year', 'bigint', 'Inferred from Parquet file.'),
('word', 'string', 'Inferred from Parquet file.')]
{code}
string in Impala is a plain BYTE_ARRAY aka Binary. The Arrow table was
{code}
pyarrow.Table
YEAR: int64
WORD: binary
{code}
However, parquet-tool cat from parquet-mr 1.9.0 gives:
{code}
$ java -jar target/parquet-tools-1.9.0.jar cat test.parq
YEAR = 2017
WORD = V29yZCAx
YEAR = 2018
WORD = V29yZCAy
{code}
This suggests there's something wrong with the file metadata is Impala is able to read the file OK. I'm looking more closely into it
> possible issue on writing parquet via pyarrow, subsequently read in Hive
> ------------------------------------------------------------------------
>
> Key: ARROW-785
> URL: https://issues.apache.org/jira/browse/ARROW-785
> Project: Apache Arrow
> Issue Type: Bug
> Reporter: Jeff Reback
> Priority: Minor
> Fix For: 0.3.0
>
>
> details here: http://stackoverflow.com/questions/43268872/parquet-creation-conversion-from-pandas-dataframe-to-pyarrow-table-not-working-f
> This round trips in pandas->parquet->pandas just fine on released pandas (0.19.2) and pyarrow (0.2).
> OP stats that it is not readable in Hive however.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)