RanaPriyansh commented on code in PR #11138:
URL: https://github.com/apache/arrow-rs/pull/11138#discussion_r4057465358
##########
arrow-ipc/benches/ipc_reader.rs:
##########
@@ -174,16 +203,117 @@ fn criterion_benchmark(c: &mut Criterion) {
/// Return an IPC stream with 10 record batches
fn ipc_stream(options: IpcWriteOptions) -> Vec<u8> {
let batch = create_batch(8192, true);
+ ipc_stream_with_batch(&batch, options)
+}
+
+fn ipc_stream_with_batch(batch: &RecordBatch, options: IpcWriteOptions) ->
Vec<u8> {
let mut buffer = Vec::with_capacity(2 * 1024 * 1024);
let mut writer =
StreamWriter::try_new_with_options(&mut buffer,
batch.schema().as_ref(), options).unwrap();
for _ in 0..10 {
- writer.write(&batch).unwrap();
+ writer.write(batch).unwrap();
}
writer.finish().unwrap();
buffer
}
+fn lz4_options() -> IpcWriteOptions {
+ IpcWriteOptions::default()
+ .try_with_compression(Some(CompressionType::LZ4_FRAME))
+ .unwrap()
+}
+
+fn read_stream(buffer: &[u8]) {
+ let projection = None;
+ let mut reader = StreamReader::try_new(buffer, projection).unwrap();
+ for _ in 0..10 {
+ std::hint::black_box(reader.next().unwrap().unwrap());
+ }
+ assert!(reader.next().is_none());
+}
+
+fn validate_stream(buffer: &[u8], expected: &RecordBatch) {
+ let projection = None;
+ let mut reader = StreamReader::try_new(buffer, projection).unwrap();
+ for _ in 0..10 {
+ let actual = reader.next().unwrap().unwrap();
+ assert_eq!(&actual, expected);
+ }
+ assert!(reader.next().is_none());
+}
+
+fn assert_lz4_compressed_buffers(batch: &RecordBatch, require_all_positive:
bool) {
+ let mut dictionary_tracker = DictionaryTracker::new(false);
+ let mut write_context = IpcWriteContext::default();
+ let options = lz4_options();
+ let (_, encoded) = IpcDataGenerator::default()
+ .encode(batch, &mut dictionary_tracker, &options, &mut write_context)
+ .unwrap();
+ let message = root_as_message(&encoded.ipc_message).unwrap();
+ assert_eq!(message.header_type(), MessageHeader::RecordBatch);
+ let record_batch = message.header_as_record_batch().unwrap();
+ let buffers = record_batch.buffers().unwrap();
+ let mut compressed_buffers = 0;
+ for buffer in buffers {
+ let length = usize::try_from(buffer.length()).unwrap();
+ if length == 0 {
+ continue;
+ }
+ let offset = usize::try_from(buffer.offset()).unwrap();
+ let prefix_end = offset.checked_add(8).unwrap();
+ assert!(prefix_end <= encoded.arrow_data.len());
+ let prefix =
i64::from_le_bytes(encoded.arrow_data[offset..prefix_end].try_into().unwrap());
+ assert!(prefix == -1 || prefix > 0);
+ if prefix > 0 {
+ compressed_buffers += 1;
+ }
+ if require_all_positive {
+ assert!(prefix > 0);
+ }
+ }
+ assert!(compressed_buffers > 0);
+}
Review Comment:
Agreed. I removed the benchmark-only correctness and compression-layout
verification helpers. The functional benchmark runs now exercise the existing
read and write paths without extra setup assertions.
##########
arrow-ipc/benches/ipc_reader.rs:
##########
@@ -174,16 +203,117 @@ fn criterion_benchmark(c: &mut Criterion) {
/// Return an IPC stream with 10 record batches
fn ipc_stream(options: IpcWriteOptions) -> Vec<u8> {
let batch = create_batch(8192, true);
+ ipc_stream_with_batch(&batch, options)
+}
+
+fn ipc_stream_with_batch(batch: &RecordBatch, options: IpcWriteOptions) ->
Vec<u8> {
let mut buffer = Vec::with_capacity(2 * 1024 * 1024);
let mut writer =
StreamWriter::try_new_with_options(&mut buffer,
batch.schema().as_ref(), options).unwrap();
for _ in 0..10 {
- writer.write(&batch).unwrap();
+ writer.write(batch).unwrap();
}
writer.finish().unwrap();
buffer
}
+fn lz4_options() -> IpcWriteOptions {
+ IpcWriteOptions::default()
+ .try_with_compression(Some(CompressionType::LZ4_FRAME))
+ .unwrap()
+}
Review Comment:
Agreed. The LZ4 cases now follow the existing inline `IpcWriteOptions` and
setup pattern used by the ZSTD cases.
##########
arrow-ipc/benches/ipc_writer.rs:
##########
@@ -63,6 +66,74 @@ fn criterion_benchmark(c: &mut Criterion) {
})
});
+ group.bench_function("StreamWriter/write_10/lz4", |b| {
+ let batch = create_batch(8192, true);
+ let options = lz4_options();
+ let stream = write_stream(&batch, options.clone());
+ validate_stream(&stream, &batch);
+ assert_lz4_compressed_buffers(&batch, false);
+ let mut buffer = Vec::with_capacity(2 * 1024 * 1024);
+ b.iter(move || {
+ buffer.clear();
+ write_stream_into(&mut buffer, &batch, options.clone());
+ black_box(buffer.len());
+ })
+ });
+
+ group.bench_function("StreamWriter/write_10/fixed_size_binary_256", |b| {
+ let batch = create_fixed_size_binary_batch(1, 256);
+ let options = IpcWriteOptions::default();
+ let stream = write_stream(&batch, options.clone());
+ validate_stream(&stream, &batch);
+ let mut buffer = Vec::with_capacity(2 * 1024 * 1024);
+ b.iter(move || {
+ buffer.clear();
+ write_stream_into(&mut buffer, &batch, options.clone());
+ black_box(buffer.len());
+ })
+ });
+
+ group.bench_function("StreamWriter/write_10/fixed_size_binary_256/lz4",
|b| {
+ let batch = create_fixed_size_binary_batch(1, 256);
+ let options = lz4_options();
+ let stream = write_stream(&batch, options.clone());
+ validate_stream(&stream, &batch);
+ assert_lz4_compressed_buffers(&batch, true);
+ let mut buffer = Vec::with_capacity(2 * 1024 * 1024);
+ b.iter(move || {
+ buffer.clear();
+ write_stream_into(&mut buffer, &batch, options.clone());
+ black_box(buffer.len());
+ })
Review Comment:
Agreed. I removed the isolated fixed-size layouts and added one fixed-size
binary field to the shared `create_batch` fixture used by equivalent reader and
writer benchmarks.
--
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]
