[
https://issues.apache.org/jira/browse/SPARK-60442?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-60442:
-----------------------------------
Labels: pull-request-available (was: )
> Variant decoding rejects spec-valid decimals
> --------------------------------------------
>
> Key: SPARK-60442
> URL: https://issues.apache.org/jira/browse/SPARK-60442
> Project: Spark
> Issue Type: Bug
> Components: PySpark, SQL
> Affects Versions: 4.2.0, 4.3.0, 4.1.3, 4.0.4
> Reporter: Qiegang Long
> Priority: Major
> Labels: pull-request-available
>
> h3. Problem
> Spark rejects a variant decimal as {{MALFORMED_VARIANT}} when its scale is
> larger than the max precision of its physical type (9 for decimal4, 18 for
> decimal8):
> {code:java}
> // VariantUtil.checkDecimal
> if (d.precision() > maxPrecision || d.scale() > maxPrecision) throw
> malformedVariant(); {code}
> The [Variant
> spec|https://github.com/apache/parquet-format/blob/master/VariantEncoding.md]
> only ties precision to the physical type. All three decimal types store a "1
> byte scale in range [0, 38]". The physical type only sets the size of the
> unscaled integer; the scale just places the decimal point.
> h3. Examples
> The same value, 1.234E-10 (unscaled 1234, scale 13, precision 4), in
> different physical types:
> {noformat}
> Encoding Value bytes (hex) Value Spark today
> decimal8, scale 13 24 0D D2 04 00 00 00 00 00 00 1.234E-10
> 0.0000000001234
> decimal4, scale 13 20 0D D2 04 00 00 1.234E-10
> MALFORMED_VARIANT
> decimal8, scale 25 24 19 D2 04 00 00 00 00 00 00 1.234E-22
> MALFORMED_VARIANT
> {noformat}
> The first two rows are the same value, but only the decimal8 encoding is
> accepted.
> h3. Impact
> parquet-java 1.17.0+ ([PR
> #3335|https://github.com/apache/parquet-java/pull/3335]) picks the decimal
> type by precision only, so it writes 1.234E-10 as decimal4 with scale 13.
> Spark fails on any expression that decodes such a field ({{to_json}},
> {{variant_get}}, {{cast}}, {{schema_of_variant}}). Data written only by Spark
> is not affected, because Spark's writer picks the type by both scale and
> precision.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
