Guosmilesmile commented on code in PR #15843:
URL: https://github.com/apache/iceberg/pull/15843#discussion_r3107125741
##########
data/src/test/java/org/apache/iceberg/data/BaseFormatModelTests.java:
##########
@@ -628,6 +628,401 @@ void
testReaderBuilderRecordsPerBatchNotSupported(FileFormat fileFormat) throws
.isInstanceOf(UnsupportedOperationException.class);
}
+ /**
+ * Schema evolution: Adding column (reading with wider schema). Write with
DefaultSchema, read
+ * with additional optional columns. The new columns should be filled with
null values.
+ */
+ @ParameterizedTest
+ @FieldSource("FILE_FORMATS")
+ void testSchemaEvolutionAddColumn(FileFormat fileFormat) throws IOException {
+ DataGenerator dataGenerator = new DataGenerators.DefaultSchema();
+ Schema writeSchema = dataGenerator.schema();
+
+ List<Record> genericRecords = dataGenerator.generateRecords();
+ writeGenericRecords(fileFormat, writeSchema, genericRecords);
+
+ Schema readSchema =
+ new Schema(
+ Types.NestedField.required(1, "col_a", Types.StringType.get()),
+ Types.NestedField.required(2, "col_b", Types.IntegerType.get()),
+ Types.NestedField.required(3, "col_c", Types.LongType.get()),
+ Types.NestedField.required(4, "col_d", Types.FloatType.get()),
+ Types.NestedField.required(5, "col_e", Types.DoubleType.get()),
+ Types.NestedField.optional(6, "new_string_col",
Types.StringType.get()),
+ Types.NestedField.optional(7, "new_int_col",
Types.IntegerType.get()));
+
+ InputFile inputFile = encryptedFile.encryptingOutputFile().toInputFile();
+ List<T> readRecords;
+ try (CloseableIterable<T> reader =
+ FormatModelRegistry.readBuilder(fileFormat, engineType(), inputFile)
+ .project(readSchema)
+ .engineProjection(engineSchema(readSchema))
+ .build()) {
+ readRecords = ImmutableList.copyOf(reader);
+ }
+
+ assertThat(readRecords).hasSize(genericRecords.size());
+
+ List<Record> expectedGenericRecords =
+ genericRecords.stream()
+ .map(
+ record -> {
+ Record expected = GenericRecord.create(readSchema);
+ for (Types.NestedField col : writeSchema.columns()) {
+ expected.setField(col.name(), record.getField(col.name()));
+ }
+
+ expected.setField("new_string_col", null);
+ expected.setField("new_int_col", null);
+ return expected;
+ })
+ .toList();
+
+ List<T> expectedEngineRecords =
convertToEngineRecords(expectedGenericRecords, readSchema);
+ assertEquals(readSchema, expectedEngineRecords, readRecords);
+ }
+
+ /**
+ * Schema evolution: Projection / Removing column (reading with narrower
schema). Write with
+ * DefaultSchema, read with only a subset of columns (skipping middle
columns).
+ */
+ @ParameterizedTest
+ @FieldSource("FILE_FORMATS")
+ void testSchemaEvolutionProjection(FileFormat fileFormat) throws IOException
{
+ DataGenerator dataGenerator = new DataGenerators.DefaultSchema();
+ Schema writeSchema = dataGenerator.schema();
+
+ List<Record> genericRecords = dataGenerator.generateRecords();
+ writeGenericRecords(fileFormat, writeSchema, genericRecords);
+
+ Schema projectedSchema =
+ new Schema(
+ Types.NestedField.required(1, "col_a", Types.StringType.get()),
+ Types.NestedField.required(5, "col_e", Types.DoubleType.get()));
+
+ InputFile inputFile = encryptedFile.encryptingOutputFile().toInputFile();
+ List<T> readRecords;
+ try (CloseableIterable<T> reader =
+ FormatModelRegistry.readBuilder(fileFormat, engineType(), inputFile)
+ .project(projectedSchema)
+ .engineProjection(engineSchema(projectedSchema))
+ .build()) {
+ readRecords = ImmutableList.copyOf(reader);
+ }
+
+ assertThat(readRecords).hasSize(genericRecords.size());
+
+ List<Record> expectedGenericRecords =
+ genericRecords.stream()
+ .map(
+ record -> {
+ Record expected = GenericRecord.create(projectedSchema);
+ expected.setField("col_a", record.getField("col_a"));
+ expected.setField("col_e", record.getField("col_e"));
+ return expected;
+ })
+ .toList();
+
+ assertEquals(
+ projectedSchema,
+ convertToEngineRecords(expectedGenericRecords, projectedSchema),
+ readRecords);
+ }
+
+ /**
+ * Schema evolution: Removing and adding a column with the same name (name
mapping). Write with
+ * DefaultSchema where col_b has field ID 2. Read with a schema where col_b
has field ID 6 (a new
+ * column). The NameMapping maps file column name "col_b" to old field ID 2.
Since the read schema
+ * expects field ID 6, the new col_b should be null.
+ */
+ @ParameterizedTest
+ @FieldSource("FILE_FORMATS")
+ void testSchemaEvolutionReNameMapping(FileFormat fileFormat) throws
IOException {
+
+ DataGenerator dataGenerator = new DataGenerators.DefaultSchema();
+ Schema writeSchema = dataGenerator.schema();
+
+ List<Record> genericRecords = dataGenerator.generateRecords();
+ writeGenericRecords(fileFormat, writeSchema, genericRecords);
+
+ // Remove col_b and add a new col_b with a different field ID
+ Schema readSchema =
+ new Schema(
+ Types.NestedField.required(1, "col_a", Types.StringType.get()),
+ Types.NestedField.optional(6, "col_b", Types.IntegerType.get()),
+ Types.NestedField.required(3, "col_c", Types.LongType.get()),
+ Types.NestedField.required(4, "col_d", Types.FloatType.get()),
+ Types.NestedField.required(5, "col_e", Types.DoubleType.get()));
+
+ InputFile inputFile = encryptedFile.encryptingOutputFile().toInputFile();
+ List<T> readRecords;
+ try (CloseableIterable<T> reader =
+ FormatModelRegistry.readBuilder(fileFormat, engineType(), inputFile)
+ .project(readSchema)
+ .engineProjection(engineSchema(readSchema))
+ .build()) {
+ readRecords = ImmutableList.copyOf(reader);
+ }
+
+ assertThat(readRecords).hasSize(genericRecords.size());
+
+ List<Record> expectedGenericRecords =
+ genericRecords.stream()
+ .map(
+ record -> {
+ Record expected = GenericRecord.create(readSchema);
+ expected.setField("col_a", record.getField("col_a"));
+ expected.setField("col_b", null);
+ expected.setField("col_c", record.getField("col_c"));
+ expected.setField("col_d", record.getField("col_d"));
+ expected.setField("col_e", record.getField("col_e"));
+ return expected;
+ })
+ .toList();
+
+ assertEquals(
+ readSchema, convertToEngineRecords(expectedGenericRecords,
readSchema), readRecords);
+ }
+
+ /**
+ * Schema evolution: Allowed type changes. Write with DefaultSchema
(col_b:int, col_d:float), read
+ * with promoted types (col_b:long, col_d:double).
+ */
+ @ParameterizedTest
+ @FieldSource("FILE_FORMATS")
+ void testSchemaEvolutionTypePromotion(FileFormat fileFormat) throws
IOException {
Review Comment:
> Date to Timestamp (v3+): date → timestamp or timestamp_ns
Added `testSchemaEvolutionTypePromotionDateToTimestamp` and
`testSchemaEvolutionTypePromotionDateToTimestampNano`. Currently, these two
test cases are failing, and the behavior of Flink and Spark are completely
different. Moreover, this type of conversion is not supported in
`TypeUtil.isPromotionAllowed`, but V3 types do support this conversion. So I'm
wondering and suspecting that this is implemented at the engine layer rather
than the file format layer.
Still investigating, feel free to add any additional information if you have
any.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
