Merge pull request #4726 from dantengsky/fix-bump-parquet2

ISSUE-4668: Enable `Lz4Raw` & rm `parquet_format_async_temp`

Cargo.toml

@@ -66,4 +66,4 @@ object = { opt-level = 3 }
 rustc-demangle = { opt-level = 3 }
 
 [patch.crates-io]
-parquet2 = { version = "0.10", optional = true, git = "https://github.com/datafuse-extras/parquet2", rev = "b3efb4e" }
+parquet2 = { version = "0.10", optional = true, git = "https://github.com/datafuse-extras/parquet2", rev = "daae989" }

common/arrow/Cargo.toml

@@ -38,6 +38,4 @@ arrow-format = { version = "0.4.0", features = ["flight-data", "flight-service",
 futures = "0.3.21"
 parquet2 = { version = "0.10.3", default_features = false }
 
-parquet-format-async-temp = "=0.2.0"
-
 [dev-dependencies]

common/arrow/src/parquet_write.rs

@@ -22,7 +22,7 @@ use arrow::io::parquet::write::to_parquet_schema;
 use arrow::io::parquet::write::RowGroupIterator;
 use parquet2::write::FileWriter;
 use parquet2::write::WriteOptions;
-use parquet_format_async_temp::FileMetaData;
+use parquet2::FileMetaData;
 
 // a simple wrapper for code reuse
 pub fn write_parquet_file<W: Write, A, I>(

common/streams/tests/it/sources/source_parquet.rs

@@ -39,7 +39,7 @@ async fn test_source_parquet() -> Result<()> {
     use common_arrow::arrow::io::parquet::write::*;
     let options = WriteOptions {
         write_statistics: true,
-        compression: Compression::Lz4, // let's begin with lz4
+        compression: Compression::Lz4Raw,
         version: Version::V2,
     };
 

query/Cargo.toml

@@ -88,7 +88,6 @@ num_cpus = "1.13.1"
 octocrab = "0.15.4"
 once_cell = "1.10.0"
 opendal = "0.5.1"
-parquet-format-async-temp = "=0.2.0"
 paste = "1.0.7"
 petgraph = "0.6.0"
 poem = { version = "=1.3.16", features = ["rustls", "multipart", "compression"] }

query/src/interpreters/interpreter_user_stage_create.rs

@@ -55,7 +55,7 @@ impl Interpreter for CreateUserStageInterpreter {
         if user_stage.stage_type == StageType::Internal {
             let prefix = format!("stage/{}/", user_stage.stage_name);
             let op = self.ctx.get_storage_operator()?;
-            op.object(&prefix).create().await?;
+            op.object(&prefix).create().await?
         }
 
         let _create_stage = user_mgr

query/src/storages/fuse/fuse_part.rs

@@ -21,6 +21,8 @@ use common_exception::Result;
 use common_planners::PartInfo;
 use common_planners::PartInfoPtr;
 
+use crate::storages::fuse::meta::Compression;
+
 #[derive(serde::Serialize, serde::Deserialize, PartialEq)]
 pub struct ColumnMeta {
     pub offset: u64,
@@ -46,6 +48,7 @@ pub struct FusePartInfo {
     pub format_version: u64,
     pub nums_rows: usize,
     pub columns_meta: HashMap<usize, ColumnMeta>,
+    pub compression: Compression,
 }
 
 #[typetag::serde(name = "fuse")]
@@ -68,12 +71,14 @@ impl FusePartInfo {
         format_version: u64,
         rows_count: u64,
         columns_meta: HashMap<usize, ColumnMeta>,
+        compression: Compression,
     ) -> Arc<Box<dyn PartInfo>> {
         Arc::new(Box::new(FusePartInfo {
             location,
             format_version,
             columns_meta,
             nums_rows: rows_count as usize,
+            compression,
         }))
     }
 

query/src/storages/fuse/io/read/block_reader.rs

@@ -20,7 +20,7 @@ use common_arrow::arrow::io::parquet::read::column_iter_to_arrays;
 use common_arrow::arrow::io::parquet::read::ArrayIter;
 use common_arrow::arrow::io::parquet::read::RowGroupDeserializer;
 use common_arrow::arrow::io::parquet::write::to_parquet_schema;
-use common_arrow::parquet::compression::Compression;
+use common_arrow::parquet::compression::Compression as ParquetCompression;
 use common_arrow::parquet::metadata::ColumnDescriptor;
 use common_arrow::parquet::metadata::SchemaDescriptor;
 use common_arrow::parquet::read::BasicDecompressor;
@@ -41,6 +41,7 @@ use opendal::Operator;
 
 use crate::storages::fuse::fuse_part::ColumnMeta;
 use crate::storages::fuse::fuse_part::FusePartInfo;
+use crate::storages::fuse::meta::Compression;
 
 #[derive(Clone)]
 pub struct BlockReader {
@@ -76,11 +77,12 @@ impl BlockReader {
         rows: usize,
         descriptor: &ColumnDescriptor,
         field: Field,
+        compression: &Compression,
     ) -> Result<ArrayIter<'static>> {
         let pages = PageIterator::new(
             std::io::Cursor::new(chunk),
             meta.num_values as i64,
-            Compression::Lz4,
+            Self::to_parquet_compression(compression),
             descriptor.clone(),
             Arc::new(|_, _| true),
             vec![],
@@ -137,6 +139,7 @@ impl BlockReader {
                 rows,
                 column_descriptor,
                 field,
+                &part.compression,
             )?);
         }
 
@@ -165,6 +168,7 @@ impl BlockReader {
                 num_rows,
                 column_descriptor,
                 field,
+                &part.compression,
             )?);
         }
 
@@ -224,4 +228,11 @@ impl BlockReader {
             Some(Ok(chunk)) => DataBlock::from_chunk(&self.projected_schema, &chunk),
         }
     }
+
+    fn to_parquet_compression(meta_compression: &Compression) -> ParquetCompression {
+        match meta_compression {
+            Compression::Lz4 => ParquetCompression::Lz4,
+            Compression::Lz4Raw => ParquetCompression::Lz4Raw,
+        }
+    }
 }

query/src/storages/fuse/io/write/block_stream_writer.rs

@@ -15,6 +15,7 @@
 use std::collections::HashMap;
 use std::sync::Arc;
 
+use common_arrow::parquet::FileMetaData;
 use common_datablocks::DataBlock;
 use common_datavalues::DataSchema;
 use common_exception::ErrorCode;
@@ -25,7 +26,6 @@ use futures::stream::Stream;
 use futures::StreamExt;
 use futures::TryStreamExt;
 use opendal::Operator;
-use parquet_format_async_temp::FileMetaData;
 
 use super::block_writer;
 use crate::storages::fuse::io::TableMetaLocationGenerator;

query/src/storages/fuse/io/write/block_writer.rs

@@ -19,11 +19,11 @@ use common_arrow::arrow::datatypes::Schema as ArrowSchema;
 use common_arrow::arrow::io::parquet::write::WriteOptions;
 use common_arrow::arrow::io::parquet::write::*;
 use common_arrow::parquet::encoding::Encoding;
+use common_arrow::parquet::FileMetaData;
 use common_datablocks::DataBlock;
 use common_exception::ErrorCode;
 use common_exception::Result;
 use opendal::Operator;
-use parquet_format_async_temp::FileMetaData;
 
 pub async fn write_block(
     arrow_schema: &ArrowSchema,
@@ -33,7 +33,7 @@ pub async fn write_block(
 ) -> Result<(u64, FileMetaData)> {
     let options = WriteOptions {
         write_statistics: false,
-        compression: Compression::Lz4, // let's begin with lz4
+        compression: Compression::Lz4Raw,
         version: Version::V2,
     };
     let batch = Chunk::try_from(block)?;

query/src/storages/fuse/meta/common.rs

@@ -14,6 +14,8 @@
 
 use std::collections::HashMap;
 
+use serde::Deserialize;
+use serde::Serialize;
 use uuid::Uuid;
 
 use crate::storages::index::ColumnStatistics;
@@ -23,7 +25,7 @@ pub type FormatVersion = u64;
 pub type SnapshotId = Uuid;
 pub type Location = (String, FormatVersion);
 
-#[derive(serde::Serialize, serde::Deserialize, Clone, Debug, Default)]
+#[derive(Serialize, Deserialize, Clone, Debug, Default)]
 pub struct Statistics {
     pub row_count: u64,
     pub block_count: u64,
@@ -40,3 +42,15 @@ where Self: Sized
 {
     const VERSION: u64 = V;
 }
+
+#[derive(Serialize, Deserialize, PartialEq, Copy, Clone, Debug)]
+pub enum Compression {
+    Lz4,
+    Lz4Raw,
+}
+
+impl Compression {
+    pub fn legacy() -> Self {
+        Compression::Lz4
+    }
+}

query/src/storages/fuse/meta/mod.rs

@@ -22,6 +22,7 @@ mod v1;
 mod versions;
 
 pub use common::ColumnId;
+pub use common::Compression;
 pub use common::Location;
 pub use common::SnapshotId;
 pub use common::Statistics;

query/src/storages/fuse/meta/v1/segment.rs

@@ -16,8 +16,11 @@
 use std::collections::HashMap;
 
 use common_datablocks::DataBlock;
+use serde::Deserialize;
+use serde::Serialize;
 
 use crate::storages::fuse::meta::common::ColumnId;
+use crate::storages::fuse::meta::common::Compression;
 use crate::storages::fuse::meta::common::FormatVersion;
 use crate::storages::fuse::meta::common::Location;
 use crate::storages::fuse::meta::common::Statistics;
@@ -26,7 +29,7 @@ use crate::storages::fuse::meta::v0::ColumnMeta;
 use crate::storages::index::ColumnStatistics;
 
 /// A segment comprises one or more blocks
-#[derive(serde::Serialize, serde::Deserialize, Debug)]
+#[derive(Serialize, Deserialize, Debug)]
 pub struct SegmentInfo {
     /// format version
     format_version: FormatVersion,
@@ -38,14 +41,22 @@ pub struct SegmentInfo {
 
 /// Meta information of a block
 /// Part of and kept inside the [SegmentInfo]
-#[derive(serde::Serialize, serde::Deserialize, Clone, Debug)]
+#[derive(Serialize, Deserialize, Clone, Debug)]
 pub struct BlockMeta {
     pub row_count: u64,
     pub block_size: u64,
     pub file_size: u64,
     pub col_stats: HashMap<ColumnId, ColumnStatistics>,
     pub col_metas: HashMap<ColumnId, ColumnMeta>,
     pub location: Location,
+
+    /// Compression algo used to compress the columns of blocks
+    ///
+    /// If not specified, the legacy algo `Lz4` will be used.
+    /// `Lz4` is merely for backward compatibility, it will NO longer be
+    /// used in the write path.
+    #[serde(default = "Compression::legacy")]
+    pub compression: Compression,
 }
 
 impl SegmentInfo {
@@ -83,6 +94,7 @@ impl From<v0::BlockMeta> for BlockMeta {
             col_stats: s.col_stats,
             col_metas: s.col_metas,
             location: (s.location.path, DataBlock::VERSION),
+            compression: Compression::Lz4,
         }
     }
 }

query/src/storages/fuse/operations/mod.rs

@@ -16,11 +16,9 @@ mod append;
 mod commit;
 mod operation_log;
 mod optimize;
-mod part_info;
 mod read;
 mod read_partitions;
 mod truncate;
 
 pub use operation_log::AppendOperationLogEntry;
 pub use operation_log::TableOperationLog;
-pub use part_info::PartInfo;

query/src/storages/fuse/operations/optimize.rs

@@ -72,7 +72,6 @@ impl FuseTable {
         // segments which no longer need to be kept
         let seg_delta = prevs.difference(&current_segments).collect::<Vec<_>>();
 
-        // TODO rm those deference **
         // blocks to be removed
         let prev_blocks: HashSet<String> = self
             .blocks_of(seg_delta.iter().map(|i| **i), ctx.clone())