razajafri commented on a change in pull request #31284:
URL: https://github.com/apache/spark/pull/31284#discussion_r563882773
##########
File path:
sql/core/src/main/java/org/apache/spark/sql/execution/datasources/parquet/VectorizedColumnReader.java
##########
@@ -122,7 +123,24 @@ public VectorizedColumnReader(
DictionaryPage dictionaryPage = pageReader.readDictionaryPage();
if (dictionaryPage != null) {
try {
- this.dictionary =
dictionaryPage.getEncoding().initDictionary(descriptor, dictionaryPage);
+ PrimitiveType primitiveType = descriptor.getPrimitiveType();
+ if (primitiveType.getOriginalType() == OriginalType.DECIMAL &&
+ primitiveType.getDecimalMetadata().getPrecision() <=
Decimal.MAX_INT_DIGITS() &&
+ primitiveType.getPrimitiveTypeName() ==
PrimitiveType.PrimitiveTypeName.INT64) {
+ // We need to make sure that we initialize the right type for the
dictionary otherwise
+ // Encoding#initDictionary will initialize it to
PrimitiveTypeName.INT64 and
+ // WritableColumnVector will throw an exception when trying to
decode to an Int when the
+ // dictionary is in fact initialized as Long
Review comment:
I don't know for sure but 3rd party only base the output on parquet
schema of the file.
The reason why we have to do this is because we are initializing the
`WritableColumnVector` based on the sparkSchema and thus initializes it based
on the precision and anything less than 10 precision is initialized to be
backed by an Int even if the parquet schema has it stored as a long.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
