You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@arrow.apache.org by "Zosimova Zhanna (Jira)" <ji...@apache.org> on 2020/11/07 18:48:00 UTC
[jira] [Updated] (ARROW-10514) [C++][Parquet] Data inconsistency in
parquet-reader output modes
[ https://issues.apache.org/jira/browse/ARROW-10514?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Zosimova Zhanna updated ARROW-10514:
------------------------------------
Description:
I tried reading description for Parquet [file|https://github.com/ClickHouse/ClickHouse/blob/master/tests/queries/0_stateless/data_parquet/nested_maps.snappy.parquet] with nested maps using [parquet-reader tool|https://github.com/apache/arrow/blob/master/cpp/tools/parquet/parquet_reader.cc].
This file has the following structure:
{code:java}
required group field_id=0 spark_schema {
optional group field_id=1 a (Map) {
repeated group field_id=2 key_value {
required binary field_id=3 key (String);
optional group field_id=4 value (Map) {
repeated group field_id=5 key_value {
required int32 field_id=6 key;
required boolean field_id=7 value;
}
}
}
}
required int32 field_id=8 b;
required double field_id=9 c;
} {code}
When I print it using DebugPrint, I see:
{code:java}
$ ./parquet-reader nested_maps.snappy.parquet --only-metadata
<some text is omitted for the sake of readability>
Column 0: a.key_value.key (BYTE_ARRAY/UTF8)
Column 1: a.key_value.value.key_value.key (INT32)
Column 2: a.key_value.value.key_value.value (BOOLEAN)
Column 3: b (INT32)
Column 4: c (DOUBLE)
</some text is omitted for the sake of readability>{code}
When I pring it using JSONPrint, I see:
{code:java}
$ ./parquet-reader nested_maps.snappy.parquet --json
<some text is omitted for the sake of readability>
"Columns": [
{ "Id": "0", "Name": "key", "PhysicalType": "BYTE_ARRAY", "ConvertedType": "UTF8", "LogicalType": {"Type": "String"} },
{ "Id": "1", "Name": "key", "PhysicalType": "INT32", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
{ "Id": "2", "Name": "value", "PhysicalType": "BOOLEAN", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
{ "Id": "3", "Name": "b", "PhysicalType": "INT32", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
{ "Id": "4", "Name": "c", "PhysicalType": "DOUBLE", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} }
]
</some text is omitted for the sake of readability>{code}
Column 0 and Column 1 has the same Name in JSON output. That's very confusing. It would be more correct to output the full path of the column (key -> a.key_value.key).
This can be corrected by changing a single line: [https://github.com/apache/arrow/blob/master/cpp/src/parquet/printer.cc#L218]
The proposed patch in the attachment
was:
I tried reading description for Parquet [file|[https://github.com/ClickHouse/ClickHouse/blob/master/tests/queries/0_stateless/data_parquet/nested_maps.snappy.parquet]] with nested maps using [parquet-reader tool|[https://github.com/apache/arrow/blob/master/cpp/tools/parquet/parquet_reader.cc]].
This file has the following structure:
{code:java}
required group field_id=0 spark_schema {
optional group field_id=1 a (Map) {
repeated group field_id=2 key_value {
required binary field_id=3 key (String);
optional group field_id=4 value (Map) {
repeated group field_id=5 key_value {
required int32 field_id=6 key;
required boolean field_id=7 value;
}
}
}
}
required int32 field_id=8 b;
required double field_id=9 c;
} {code}
When I print it using DebugPrint, I see:
{code:java}
$ ./parquet-reader nested_maps.snappy.parquet --only-metadata
<some text is omitted for the sake of readability>
Column 0: a.key_value.key (BYTE_ARRAY/UTF8)
Column 1: a.key_value.value.key_value.key (INT32)
Column 2: a.key_value.value.key_value.value (BOOLEAN)
Column 3: b (INT32)
Column 4: c (DOUBLE)
</some text is omitted for the sake of readability>{code}
When I pring it using JSONPrint, I see:
{code:java}
$ ./parquet-reader nested_maps.snappy.parquet --json
<some text is omitted for the sake of readability>
"Columns": [
{ "Id": "0", "Name": "key", "PhysicalType": "BYTE_ARRAY", "ConvertedType": "UTF8", "LogicalType": {"Type": "String"} },
{ "Id": "1", "Name": "key", "PhysicalType": "INT32", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
{ "Id": "2", "Name": "value", "PhysicalType": "BOOLEAN", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
{ "Id": "3", "Name": "b", "PhysicalType": "INT32", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
{ "Id": "4", "Name": "c", "PhysicalType": "DOUBLE", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} }
]
</some text is omitted for the sake of readability>{code}
Column 0 and Column 1 has the same Name in JSON output. That's very confusing. It would be more correct to output the full path of the column (key -> a.key_value.key).
This can be corrected by changing a single line: [https://github.com/apache/arrow/blob/master/cpp/src/parquet/printer.cc#L218]
The proposed patch in the attachment
> [C++][Parquet] Data inconsistency in parquet-reader output modes
> ----------------------------------------------------------------
>
> Key: ARROW-10514
> URL: https://issues.apache.org/jira/browse/ARROW-10514
> Project: Apache Arrow
> Issue Type: Bug
> Components: C++
> Reporter: Zosimova Zhanna
> Priority: Minor
> Attachments: 0001-Make-the-column-name-the-same-for-both-output-format.patch
>
>
> I tried reading description for Parquet [file|https://github.com/ClickHouse/ClickHouse/blob/master/tests/queries/0_stateless/data_parquet/nested_maps.snappy.parquet] with nested maps using [parquet-reader tool|https://github.com/apache/arrow/blob/master/cpp/tools/parquet/parquet_reader.cc].
> This file has the following structure:
> {code:java}
> required group field_id=0 spark_schema {
> optional group field_id=1 a (Map) {
> repeated group field_id=2 key_value {
> required binary field_id=3 key (String);
> optional group field_id=4 value (Map) {
> repeated group field_id=5 key_value {
> required int32 field_id=6 key;
> required boolean field_id=7 value;
> }
> }
> }
> }
> required int32 field_id=8 b;
> required double field_id=9 c;
> } {code}
> When I print it using DebugPrint, I see:
> {code:java}
> $ ./parquet-reader nested_maps.snappy.parquet --only-metadata
> <some text is omitted for the sake of readability>
> Column 0: a.key_value.key (BYTE_ARRAY/UTF8)
> Column 1: a.key_value.value.key_value.key (INT32)
> Column 2: a.key_value.value.key_value.value (BOOLEAN)
> Column 3: b (INT32)
> Column 4: c (DOUBLE)
> </some text is omitted for the sake of readability>{code}
> When I pring it using JSONPrint, I see:
> {code:java}
> $ ./parquet-reader nested_maps.snappy.parquet --json
> <some text is omitted for the sake of readability>
> "Columns": [
> { "Id": "0", "Name": "key", "PhysicalType": "BYTE_ARRAY", "ConvertedType": "UTF8", "LogicalType": {"Type": "String"} },
> { "Id": "1", "Name": "key", "PhysicalType": "INT32", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
> { "Id": "2", "Name": "value", "PhysicalType": "BOOLEAN", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
> { "Id": "3", "Name": "b", "PhysicalType": "INT32", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} },
> { "Id": "4", "Name": "c", "PhysicalType": "DOUBLE", "ConvertedType": "NONE", "LogicalType": {"Type": "None"} }
> ]
> </some text is omitted for the sake of readability>{code}
> Column 0 and Column 1 has the same Name in JSON output. That's very confusing. It would be more correct to output the full path of the column (key -> a.key_value.key).
>
> This can be corrected by changing a single line: [https://github.com/apache/arrow/blob/master/cpp/src/parquet/printer.cc#L218]
>
> The proposed patch in the attachment
--
This message was sent by Atlassian Jira
(v8.3.4#803005)