perf: reduce read amplification for partitioned JSON file scanning #19687

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open

Weijun-H wants to merge 7 commits into apache:main from Weijun-H:json-duplicated-read

+846 −16

Cargo.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

datafusion/datasource-json/Cargo.toml

-Original file line number
+Diff line change
@@ Expand Up / @@ -46,6 +46,9 @@ futures = { workspace = true } @@
     object_store = { workspace = true }
     tokio = { workspace = true }
+    [dev-dependencies]
+    criterion = { workspace = true }
     # Note: add additional linter rules in lib.rs.
     # Rust does not support workspace + new linter rules in subcrates yet
     # https://github.com/rust-lang/cargo/issues/13157
@@ Expand All / @@ -55,3 +58,7 @@ workspace = true @@
     [lib]
     name = "datafusion_datasource_json"
     path = "src/mod.rs"
+    [[bench]]
+    name = "json_boundary"
+    harness = false

datafusion/datasource-json/benches/json_boundary.rs

-Original file line number
+Diff line change
@@ -0,0 +1,302 @@
+    // Licensed to the Apache Software Foundation (ASF) under one
+    // or more contributor license agreements.  See the NOTICE file
+    // distributed with this work for additional information
+    // regarding copyright ownership.  The ASF licenses this file
+    // to you under the Apache License, Version 2.0 (the
+    // "License"); you may not use this file except in compliance
+    // with the License.  You may obtain a copy of the License at
+    //
+    //   http://www.apache.org/licenses/LICENSE-2.0
+    //
+    // Unless required by applicable law or agreed to in writing,
+    // software distributed under the License is distributed on an
+    // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+    // KIND, either express or implied.  See the License for the
+    // specific language governing permissions and limitations
+    // under the License.
+    use arrow::datatypes::{DataType, Field, Schema};
+    use async_trait::async_trait;
+    use bytes::Bytes;
+    use criterion::{Criterion, Throughput, criterion_group, criterion_main};
+    use datafusion_datasource::file::FileSource;
+    use datafusion_datasource::file_groups::FileGroup;
+    use datafusion_datasource::file_scan_config::FileScanConfigBuilder;
+    use datafusion_datasource::source::DataSourceExec;
+    use datafusion_datasource::{FileRange, PartitionedFile, TableSchema};
+    use datafusion_datasource_json::source::JsonSource;
+    use datafusion_execution::TaskContext;
+    use datafusion_execution::object_store::ObjectStoreUrl;
+    use datafusion_physical_plan::ExecutionPlan;
+    use futures::StreamExt;
+    use futures::stream::BoxStream;
+    use object_store::memory::InMemory;
+    use object_store::path::Path;
+    use object_store::{
+        GetOptions, GetRange, GetResult, ListResult, MultipartUpload, ObjectStore,
+        PutMultipartOptions, PutOptions, PutPayload, PutResult,
+    };
+    use std::fmt;
+    use std::sync::Arc;
+    use std::sync::atomic::{AtomicU64, Ordering};
+    use tokio::runtime::{Builder, Runtime};
+    // Add CPU cost per requested KB to make read amplification visible in timings.
+    const CPU_COST_PER_KB_ROUNDS: u32 = 64;
+    const BYTES_PER_KB: u64 = 1024;
+    #[derive(Debug)]
+    struct CountingObjectStore {
+        inner: Arc<dyn ObjectStore>,
+        requested_bytes: AtomicU64,
+        cpu_cost_per_kb_rounds: u32,
+    }
+    impl CountingObjectStore {
+        fn new_with_cpu_cost(
+            inner: Arc<dyn ObjectStore>,
+            cpu_cost_per_kb_rounds: u32,
+        ) -> Self {
+            Self {
+                inner,
+                requested_bytes: AtomicU64::new(0),
+                cpu_cost_per_kb_rounds,
+            }
+        }
+        fn reset(&self) {
+            self.requested_bytes.store(0, Ordering::Relaxed);
+        }
+        fn requested_bytes(&self) -> u64 {
+            self.requested_bytes.load(Ordering::Relaxed)
+        }
+    }
+    impl fmt::Display for CountingObjectStore {
+        fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+            write!(f, "CountingObjectStore({})", self.inner)
+        }
+    }
+    #[async_trait]
+    impl ObjectStore for CountingObjectStore {
+        async fn put_opts(
+            &self,
+            location: &Path,
+            payload: PutPayload,
+            opts: PutOptions,
+        ) -> object_store::Result<PutResult> {
+            self.inner.put_opts(location, payload, opts).await
+        }
+        async fn put_multipart_opts(
+            &self,
+            location: &Path,
+            opts: PutMultipartOptions,
+        ) -> object_store::Result<Box<dyn MultipartUpload>> {
+            self.inner.put_multipart_opts(location, opts).await
+        }
+        async fn get_opts(
+            &self,
+            location: &Path,
+            options: GetOptions,
+        ) -> object_store::Result<GetResult> {
+            let should_burn_cpu = self.cpu_cost_per_kb_rounds > 0;
+            let mut requested_len = 0u64;
+            if let Some(range) = options.range.as_ref() {
+                let requested = match range {
+                    GetRange::Bounded(r) => r.end.saturating_sub(r.start),
+                    GetRange::Offset(_) | GetRange::Suffix(_) => 0,
+                };
+                requested_len = requested;
+                self.requested_bytes.fetch_add(requested, Ordering::Relaxed);
+            }
+            let result = self.inner.get_opts(location, options).await;
+            if should_burn_cpu {
+                burn_cpu_kb(requested_len, self.cpu_cost_per_kb_rounds);
+            }
+            result
+        }
+        async fn delete(&self, location: &Path) -> object_store::Result<()> {
+            self.inner.delete(location).await
+        }
+        fn list(
+            &self,
+            prefix: Option<&Path>,
+        ) -> BoxStream<'static, object_store::Result<object_store::ObjectMeta>> {
+            self.inner.list(prefix)
+        }
+        async fn list_with_delimiter(
+            &self,
+            prefix: Option<&Path>,
+        ) -> object_store::Result<ListResult> {
+            self.inner.list_with_delimiter(prefix).await
+        }
+        async fn copy(&self, from: &Path, to: &Path) -> object_store::Result<()> {
+            self.inner.copy(from, to).await
+        }
+        async fn copy_if_not_exists(
+            &self,
+            from: &Path,
+            to: &Path,
+        ) -> object_store::Result<()> {
+            self.inner.copy_if_not_exists(from, to).await
+        }
+    }
+    fn build_fixed_json_lines(line_len: usize, lines: usize) -> Bytes {
+        let prefix = r#"{"value":""#;
+        let suffix = "\"}\n";
+        let min_len = prefix.len() + suffix.len() + 1;
+        assert!(line_len >= min_len, "line_len must be at least {min_len}");
+        let padding_len = line_len - prefix.len() - suffix.len();
+        let mut line = Vec::with_capacity(line_len);
+        line.extend_from_slice(prefix.as_bytes());
+        line.extend(std::iter::repeat_n(b'a', padding_len));
+        line.extend_from_slice(suffix.as_bytes());
+        let mut data = Vec::with_capacity(line_len * lines);
+        for _ in 0..lines {
+            data.extend_from_slice(&line);
+        }
+        Bytes::from(data)
+    }
+    fn burn_cpu_kb(bytes: u64, rounds: u32) {
+        if bytes == 0 || rounds == 0 {
+            return;
+        }
+        let kb = bytes.div_ceil(BYTES_PER_KB);
+        let mut checksum = 0u64;
+        let mut remaining = kb.saturating_mul(rounds as u64);
+        while remaining > 0 {
+            checksum = checksum.wrapping_add(remaining);
+            checksum = checksum.rotate_left(5) ^ 0x9e3779b97f4a7c15;
+            remaining -= 1;
+        }
+        std::hint::black_box(checksum);
+    }
+    struct Fixture {
+        store: Arc<CountingObjectStore>,
+        task_ctx: Arc<TaskContext>,
+        exec: Arc<dyn ExecutionPlan>,
+    }
+    fn build_fixture(rt: &Runtime) -> Fixture {
+        let inner: Arc<dyn ObjectStore> = Arc::new(InMemory::new());
+        let store = Arc::new(CountingObjectStore::new_with_cpu_cost(
+            Arc::clone(&inner),
+            CPU_COST_PER_KB_ROUNDS,
+        ));
+        let store_dyn: Arc<dyn ObjectStore> = store.clone();
+        let path = Path::from("bench.json");
+        let line_len = 128usize;
+        let lines = 65_536usize;
+        let data = build_fixed_json_lines(line_len, lines);
+        rt.block_on(inner.put(&path, data.into())).unwrap();
+        let object_meta = rt.block_on(inner.head(&path)).unwrap();
+        let start = 1_000_003usize;
+        let raw_end = start + 256_000;
+        let end = (raw_end / line_len).max(1) * line_len;
+        let task_ctx = Arc::new(TaskContext::default());
+        let runtime_env = task_ctx.runtime_env();
+        let object_store_url = ObjectStoreUrl::parse("test://bucket").unwrap();
+        // Register a CPU-costed store to approximate non-streaming remote reads.
+        runtime_env.register_object_store(object_store_url.as_ref(), Arc::clone(&store_dyn));
+        let schema = Arc::new(Schema::new(vec![Field::new(
+            "value",
+            DataType::Utf8,
+            false,
+        )]));
+        let table_schema = TableSchema::from_file_schema(schema);
+        let file_source: Arc<dyn FileSource> = Arc::new(JsonSource::new(table_schema));
+        let file = build_partitioned_file(object_meta.clone(), start, end);
+        let config = FileScanConfigBuilder::new(object_store_url, file_source)
+            .with_file_groups(vec![FileGroup::new(vec![file])])
+            .build();
+        let exec: Arc<dyn ExecutionPlan> = DataSourceExec::from_data_source(config);
+        Fixture {
+            store,
+            task_ctx,
+            exec,
+        }
+    }
+    fn measure_datasource_exec_bytes(rt: &Runtime, fixture: &Fixture) -> u64 {
+        fixture.store.reset();
+        let rows = rt.block_on(run_datasource_exec(
+            Arc::clone(&fixture.exec),
+            Arc::clone(&fixture.task_ctx),
+        ));
+        debug_assert!(rows > 0);
+        fixture.store.requested_bytes()
+    }
+    fn build_partitioned_file(
+        object_meta: object_store::ObjectMeta,
+        start: usize,
+        end: usize,
+    ) -> PartitionedFile {
+        PartitionedFile {
+            object_meta,
+            partition_values: vec![],
+            range: Some(FileRange {
+                start: start as i64,
+                end: end as i64,
+            }),
+            statistics: None,
+            ordering: None,
+            extensions: None,
+            metadata_size_hint: None,
+        }
+    }
+    async fn run_datasource_exec(
+        exec: Arc<dyn ExecutionPlan>,
+        task_ctx: Arc<TaskContext>,
+    ) -> usize {
+        let mut stream = exec.execute(0, task_ctx).unwrap();
+        let mut rows = 0;
+        while let Some(batch) = stream.next().await {
+            let batch = batch.unwrap();
+            rows += batch.num_rows();
+        }
+        rows
+    }
+    fn bench_json_boundary(c: &mut Criterion) {
+        let rt = Builder::new_current_thread().build().unwrap();
+        let fixture = build_fixture(&rt);
+        let exec_bytes = measure_datasource_exec_bytes(&rt, &fixture);
+        let mut exec_group = c.benchmark_group("json_boundary_datasource_exec");
+        exec_group.throughput(Throughput::Bytes(exec_bytes));
+        // Fixed benchmark id for baseline comparisons; read_bytes is reported as throughput.
+        exec_group.bench_function("execute", |b| {
+            b.iter(|| {
+                fixture.store.reset();
+                rt.block_on(run_datasource_exec(
+                    Arc::clone(&fixture.exec),
+                    Arc::clone(&fixture.task_ctx),
+                ));
+            });
+        });
+        exec_group.finish();
+    }
+    criterion_group!(benches, bench_json_boundary);
+    criterion_main!(benches);

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

perf: reduce read amplification for partitioned JSON file scanning #19687

Diff view

Diff view

There are no files selected for viewing

Uh oh!

Uh oh!

Uh oh!

perf: reduce read amplification for partitioned JSON file scanning #19687

Are you sure you want to change the base?

perf: reduce read amplification for partitioned JSON file scanning #19687

Uh oh!

Uh oh!

Diff view

Diff view

There are no files selected for viewing

Uh oh!

Uh oh!

Uh oh!