If a parquet file has more than 2^31 rows, the row count written into the file metadata is incorrect.
The cause of the problem is the use of an int instead of long data type for numRows in ParquetMetadataConverter, toParquetMetadata:
int numRows = 0;
for (BlockMetaData block : blocks) {
numRows += block.getRowCount();
addRowGroup(parquetMetadata, rowGroups, block);
}
Reporter: Thomas Friedrich / @tfriedr
Assignee: Thomas Friedrich / @tfriedr
PRs and other links:
Note: This issue was originally created as PARQUET-324. Please see the migration documentation for further details.