[FEAT] connect: add parquet support

Cargo.toml

@@ -194,6 +194,7 @@ chrono-tz = "0.10.0"
 comfy-table = "7.1.1"
 common-daft-config = {path = "src/common/daft-config"}
 common-error = {path = "src/common/error", default-features = false}
+common-file-formats = {path = "src/common/file-formats"}
 common-runtime = {path = "src/common/runtime", default-features = false}
 daft-core = {path = "src/daft-core"}
 daft-dsl = {path = "src/daft-dsl"}

src/daft-connect/Cargo.toml

@@ -2,6 +2,7 @@
 arrow2 = {workspace = true}
 async-stream = "0.3.6"
 common-daft-config = {workspace = true}
+common-file-formats = {workspace = true}
 daft-core = {workspace = true}
 daft-dsl = {workspace = true}
 daft-local-execution = {workspace = true}

src/daft-connect/src/lib.rs

@@ -190,8 +190,9 @@ impl SparkConnectService for DaftSparkConnectService {
                     CommandType::RegisterFunction(_) => {
                         unimplemented_err!("RegisterFunction not implemented")
                     }
-                    CommandType::WriteOperation(_) => {
-                        unimplemented_err!("WriteOperation not implemented")
+                    CommandType::WriteOperation(op) => {
+                        let result = session.handle_write_command(op, operation).await?;
+                        return Ok(Response::new(result));
                     }
                     CommandType::CreateDataframeView(_) => {
                         unimplemented_err!("CreateDataframeView not implemented")

src/daft-connect/src/op/execute.rs

@@ -11,6 +11,7 @@ use uuid::Uuid;
 use crate::{DaftSparkConnectService, Session};
 
 mod root;
+mod write;
 
 pub type ExecuteStream = <DaftSparkConnectService as SparkConnectService>::ExecutePlanStream;
 

src/daft-connect/src/op/execute/write.rs

@@ -0,0 +1,144 @@
+use std::{collections::HashMap, future::ready};
+
+use common_daft_config::DaftExecutionConfig;
+use common_file_formats::FileFormat;
+use daft_local_execution::NativeExecutor;
+use eyre::{bail, WrapErr};
+use spark_connect::{
+    write_operation::{SaveMode, SaveType},
+    WriteOperation,
+};
+use tonic::Status;
+use tracing::warn;
+
+use crate::{
+    op::execute::{ExecuteStream, PlanIds},
+    session::Session,
+    translation,
+};
+
+impl Session {
+    pub async fn handle_write_command(
+        &self,
+        operation: WriteOperation,
+        operation_id: String,
+    ) -> Result<ExecuteStream, Status> {
+        use futures::StreamExt;
+
+        let context = PlanIds {
+            session: self.client_side_session_id().to_string(),
+            server_side_session: self.server_side_session_id().to_string(),
+            operation: operation_id,
+        };
+
+        let finished = context.finished();
+
+        let result = async move {
+            let WriteOperation {
+                input,
+                source,
+                mode,
+                sort_column_names,
+                partitioning_columns,
+                bucket_by,
+                options,
+                clustering_columns,
+                save_type,
+            } = operation;
+
+            let Some(input) = input else {
+                bail!("Input is required");
+            };
+
+            let Some(source) = source else {
+                bail!("Source is required");
+            };
+
+            if source != "parquet" {
+                bail!("Unsupported source: {source}; only parquet is supported");
+            }
+
+            let Ok(mode) = SaveMode::try_from(mode) else {
+                bail!("Invalid save mode: {mode}");
+            };
+
+            if !sort_column_names.is_empty() {
+                // todo(completeness): implement sort
+                warn!("Ignoring sort_column_names: {sort_column_names:?} (not yet implemented)");
+            }
+
+            if !partitioning_columns.is_empty() {
+                // todo(completeness): implement partitioning
+                warn!(
+                    "Ignoring partitioning_columns: {partitioning_columns:?} (not yet implemented)"
+                );
+            }
+
+            if let Some(bucket_by) = bucket_by {
+                // todo(completeness): implement bucketing
+                warn!("Ignoring bucket_by: {bucket_by:?} (not yet implemented)");
+            }
+
+            if !options.is_empty() {
+                // todo(completeness): implement options
+                warn!("Ignoring options: {options:?} (not yet implemented)");
+            }
+
+            if !clustering_columns.is_empty() {
+                // todo(completeness): implement clustering
+                warn!("Ignoring clustering_columns: {clustering_columns:?} (not yet implemented)");
+            }
+
+            match mode {
+                SaveMode::Unspecified => {}
+                SaveMode::Append => {}
+                SaveMode::Overwrite => {}
+                SaveMode::ErrorIfExists => {}
+                SaveMode::Ignore => {}
+            }
+
+            let Some(save_type) = save_type else {
+                bail!("Save type is required");
+            };
+
+            let path = match save_type {
+                SaveType::Path(path) => path,
+                SaveType::Table(table) => {
+                    let name = table.table_name;
+                    bail!("Tried to write to table {name} but it is not yet implemented. Try to write to a path instead.");
+                }
+            };
+
+            let plan = translation::to_logical_plan(input)?;
+
+            let plan = plan
+                .table_write(&path, FileFormat::Parquet, None, None, None)
+                .wrap_err("Failed to create table write plan")?;
+
+            let optimized_plan = plan.optimize()?;
+            let cfg = DaftExecutionConfig::default();
+            let native_executor = NativeExecutor::from_logical_plan_builder(&optimized_plan)?;
+            let mut result_stream = native_executor
+                .run(HashMap::new(), cfg.into(), None)?
+                .into_stream();
+
+            // this is so we make sure the operation is actually done
+            // before we return
+            //
+            // an example where this is important is if we write to a parquet file
+            // and then read immediately after, we need to wait for the write to finish
+            while let Some(_result) = result_stream.next().await {}
+
+            Ok(())
+        };
+
+        use futures::TryFutureExt;
+
+        let result = result.map_err(|e| Status::internal(format!("Error in Daft server: {e:?}")));
+
+        let future = result.and_then(|_| ready(Ok(finished)));
+        let stream = futures::stream::once(future);
+
+        Ok(Box::pin(stream))
+    }
+}

src/daft-connect/src/translation/logical_plan.rs

@@ -3,11 +3,14 @@ use eyre::{bail, Context};
 use spark_connect::{relation::RelType, Limit, Relation};
 use tracing::warn;
 
-use crate::translation::logical_plan::{aggregate::aggregate, project::project, range::range};
+use crate::translation::logical_plan::{
+    aggregate::aggregate, project::project, range::range, read::read,
+};
 
 mod aggregate;
 mod project;
 mod range;
+mod read;
 
 pub fn to_logical_plan(relation: Relation) -> eyre::Result<LogicalPlanBuilder> {
     if let Some(common) = relation.common {
@@ -25,6 +28,7 @@ pub fn to_logical_plan(relation: Relation) -> eyre::Result<LogicalPlanBuilder> {
         RelType::Aggregate(a) => {
             aggregate(*a).wrap_err("Failed to apply aggregate to logical plan")
         }
+        RelType::Read(r) => read(r).wrap_err("Failed to apply table read to logical plan"),
         plan => bail!("Unsupported relation type: {plan:?}"),
     }
 }

src/daft-connect/src/translation/logical_plan/read.rs

@@ -0,0 +1,29 @@
+use daft_logical_plan::LogicalPlanBuilder;
+use eyre::{bail, WrapErr};
+use spark_connect::read::ReadType;
+use tracing::warn;
+
+mod data_source;
+
+pub fn read(read: spark_connect::Read) -> eyre::Result<LogicalPlanBuilder> {
+    let spark_connect::Read {
+        is_streaming,
+        read_type,
+    } = read;
+
+    warn!("Ignoring is_streaming: {is_streaming}");
+
+    let Some(read_type) = read_type else {
+        bail!("Read type is required");
+    };
+
+    match read_type {
+        ReadType::NamedTable(table) => {
+            let name = table.unparsed_identifier;
+            bail!("Tried to read from table {name} but it is not yet implemented. Try to read from a path instead.");
+        }
+        ReadType::DataSource(source) => {
+            data_source::data_source(source).wrap_err("Failed to create data source")
+        }
+    }
+}

src/daft-connect/src/translation/logical_plan/read/data_source.rs

@@ -0,0 +1,44 @@
+use daft_logical_plan::LogicalPlanBuilder;
+use daft_scan::builder::ParquetScanBuilder;
+use eyre::{bail, ensure, WrapErr};
+use tracing::warn;
+
+pub fn data_source(
+    data_source: spark_connect::read::DataSource,
+) -> eyre::Result<LogicalPlanBuilder> {
+    let spark_connect::read::DataSource {
+        format,
+        schema,
+        options,
+        paths,
+        predicates,
+    } = data_source;
+
+    let Some(format) = format else {
+        bail!("Format is required");
+    };
+
+    if format != "parquet" {
+        bail!("Unsupported format: {format}; only parquet is supported");
+    }
+
+    ensure!(!paths.is_empty(), "Paths are required");
+
+    if let Some(schema) = schema {
+        warn!("Ignoring schema: {schema:?}; not yet implemented");
+    }
+
+    if !options.is_empty() {
+        warn!("Ignoring options: {options:?}; not yet implemented");
+    }
+
+    if !predicates.is_empty() {
+        warn!("Ignoring predicates: {predicates:?}; not yet implemented");
+    }
+
+    let builder = ParquetScanBuilder::new(paths)
+        .finish()
+        .wrap_err("Failed to create parquet scan builder")?;
+
+    Ok(builder)
+}

tests/connect/test_parquet.py

@@ -0,0 +1,36 @@
+from __future__ import annotations
+
+import os
+import shutil
+import tempfile
+
+
+def test_write_parquet(spark_session):
+    # Create a temporary directory
+    temp_dir = tempfile.mkdtemp()
+    try:
+        # Create DataFrame from range(10)
+        df = spark_session.range(10)
+
+        # Write DataFrame to parquet directory
+        parquet_dir = os.path.join(temp_dir, "test.parquet")
+        df.write.parquet(parquet_dir)
+
+        # List all files in the parquet directory
+        parquet_files = [f for f in os.listdir(parquet_dir) if f.endswith(".parquet")]
+        print(f"Parquet files in directory: {parquet_files}")
+
+        # Assert there is at least one parquet file
+        assert len(parquet_files) > 0, "Expected at least one parquet file to be written"
+
+        # Read back from the parquet directory (not specific file)
+        df_read = spark_session.read.parquet(parquet_dir)
+
+        # Verify the data is unchanged
+        df_pandas = df.toPandas()
+        df_read_pandas = df_read.toPandas()
+        assert df_pandas["id"].equals(df_read_pandas["id"]), "Data should be unchanged after write/read"
+
+    finally:
+        # Clean up temp directory
+        shutil.rmtree(temp_dir)