Skip to content

Conversation

@shangxinli
Copy link
Contributor

Implement direct Avro encoder to eliminate GenericDatum intermediate layer, matching the decoder approach for better performance.

Implementation:

  • Add avro_direct_encoder_internal.h with EncodeArrowToAvro API
  • Add avro_direct_encoder.cc implementing direct Arrow→Avro encoding
    • All primitive types: bool, int, long, float, double, string, binary
    • Temporal types: date, time, timestamp
    • Logical types: uuid, decimal
    • Nested types: struct, list, map (both string and non-string keys)
    • Union type handling for optional fields
  • Modify avro_writer.cc to use DataFileWriterBase with direct encoder
  • Add EncodeContext to reuse scratch buffers and avoid allocations

This matches Java Iceberg implementation using Encoder interface directly, avoiding intermediate object allocation overhead.

Implement direct Avro encoder to eliminate GenericDatum intermediate layer,
matching the decoder approach for better performance.

Implementation:
- Add avro_direct_encoder_internal.h with EncodeArrowToAvro API
- Add avro_direct_encoder.cc implementing direct Arrow→Avro encoding
  - All primitive types: bool, int, long, float, double, string, binary
  - Temporal types: date, time, timestamp
  - Logical types: uuid, decimal
  - Nested types: struct, list, map (both string and non-string keys)
  - Union type handling for optional fields
- Modify avro_writer.cc to use DataFileWriterBase with direct encoder
- Add EncodeContext to reuse scratch buffers and avoid allocations

This matches Java Iceberg implementation using Encoder interface directly,
avoiding intermediate object allocation overhead.
Copy link
Member

@wgtmac wgtmac left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I haven't finished my review yet, just post my findings so far.

Comment on lines +46 to +57
/// methods from Arrow arrays. Matches Java Iceberg's approach for better performance.
///
/// Features:
/// - All primitive, temporal, and logical types
/// - Nested types (struct, list, map)
/// - Union types for optional fields
///
/// Error Handling:
/// - Type mismatches → InvalidArgument
/// - Null in non-nullable field → InvalidArgument
/// - Invalid decimal precision → InvalidArgument
///
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
/// methods from Arrow arrays. Matches Java Iceberg's approach for better performance.
///
/// Features:
/// - All primitive, temporal, and logical types
/// - Nested types (struct, list, map)
/// - Union types for optional fields
///
/// Error Handling:
/// - Type mismatches → InvalidArgument
/// - Null in non-nullable field → InvalidArgument
/// - Invalid decimal precision → InvalidArgument
///
/// methods from Arrow arrays.
///

inline static Entry<int64_t> kAvroSyncInterval{"write.avro.sync-interval", 16 * 1024};

/// \brief Whether to skip GenericDatum and use direct encoder for Avro writing.
/// When true, uses direct encoder (faster). When false, uses GenericDatum (legacy).
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
/// When true, uses direct encoder (faster). When false, uses GenericDatum (legacy).
/// When true, uses direct encoder (faster). When false, uses GenericDatum.

Status EncodeArrowToAvro(const ::avro::NodePtr& avro_node, ::avro::Encoder& encoder,
const Type& type, const ::arrow::Array& array, int64_t row_index,
EncodeContext* ctx) {
if (!ctx) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I just noticed that we haven't checked null for DecodeContext* ctx in the decoder implementation. Can we use DecodeContext& and EncodeContext& ctx so we don't need to bother with null check?

const size_t num_fields = avro_node->leaves();

for (size_t i = 0; i < num_fields; ++i) {
const auto& field_node = avro_node->leafAt(i);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we need to check if num of fields mismatch between arrow array and avro node?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants