feat: allow custom caching via logical node (#18688)

jizezhang · 2010YOUY01 · dependabot[bot] · web-flow · commit 6ea305ec4c15 · 2025-11-25T09:16:53.000Z
## Which issue does this PR close?  - Closes #17297. ## Rationale for this change  See #17297. ## What changes are included in this PR?  - Added methods in `SessionState` to enable registering a `CacheProducer` that creates a logical node for caching. - Added branching in `DataFrame::cache()` to apply logical node for caching on top of original dataframe plan if cache producer is supplied, otherwise use current implementation as is. ## Are these changes tested?  Yes ## Are there any user-facing changes?   Change in `SessionState` is user-facing. --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: Yongting You <2010youy01@gmail.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Daniël Heres <danielheres@gmail.com> Co-authored-by: Kumar Ujjawal <ujjawalpathak6@gmail.com> Co-authored-by: Aryan Bagade <73382554+AryanBagade@users.noreply.github.com> Co-authored-by: Oleks V <comphead@users.noreply.github.com> Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org> Co-authored-by: Jeffrey Vo <jeffrey.vo.australia@gmail.com> Co-authored-by: Vrishabh <psvrishabh@gmail.com> Co-authored-by: Dhanush <dhanushhs51@gmail.com>
diff --git a/datafusion/core/src/dataframe/mod.rs b/datafusion/core/src/dataframe/mod.rs
@@ -2330,6 +2330,10 @@ impl DataFrame {
 
     /// Cache DataFrame as a memory table.
     ///
+    /// Default behavior could be changed using
+    /// a [`crate::execution::session_state::CacheFactory`]
+    /// configured via [`SessionState`].
+    ///
     /// ```
     /// # use datafusion::prelude::*;
     /// # use datafusion::error::Result;
@@ -2344,14 +2348,20 @@ impl DataFrame {
     /// # }
     /// ```
     pub async fn cache(self) -> Result<DataFrame> {
-        let context = SessionContext::new_with_state((*self.session_state).clone());
-        // The schema is consistent with the output
-        let plan = self.clone().create_physical_plan().await?;
-        let schema = plan.schema();
-        let task_ctx = Arc::new(self.task_ctx());
-        let partitions = collect_partitioned(plan, task_ctx).await?;
-        let mem_table = MemTable::try_new(schema, partitions)?;
-        context.read_table(Arc::new(mem_table))
+        if let Some(cache_factory) = self.session_state.cache_factory() {
+            let new_plan =
+                cache_factory.create(self.plan, self.session_state.as_ref())?;
+            Ok(Self::new(*self.session_state, new_plan))
+        } else {
+            let context = SessionContext::new_with_state((*self.session_state).clone());
+            // The schema is consistent with the output
+            let plan = self.clone().create_physical_plan().await?;
+            let schema = plan.schema();
+            let task_ctx = Arc::new(self.task_ctx());
+            let partitions = collect_partitioned(plan, task_ctx).await?;
+            let mem_table = MemTable::try_new(schema, partitions)?;
+            context.read_table(Arc::new(mem_table))
+        }
     }
 
     /// Apply an alias to the DataFrame.
diff --git a/datafusion/core/src/execution/session_state.rs b/datafusion/core/src/execution/session_state.rs
@@ -185,6 +185,7 @@ pub struct SessionState {
     /// It will be invoked on `CREATE FUNCTION` statements.
     /// thus, changing dialect o PostgreSql is required
     function_factory: Option<Arc<dyn FunctionFactory>>,
+    cache_factory: Option<Arc<dyn CacheFactory>>,
     /// Cache logical plans of prepared statements for later execution.
     /// Key is the prepared statement name.
     prepared_plans: HashMap<String, Arc<PreparedPlan>>,
@@ -206,6 +207,7 @@ impl Debug for SessionState {
             .field("table_options", &self.table_options)
             .field("table_factories", &self.table_factories)
             .field("function_factory", &self.function_factory)
+            .field("cache_factory", &self.cache_factory)
             .field("expr_planners", &self.expr_planners);
 
         #[cfg(feature = "sql")]
@@ -355,6 +357,16 @@ impl SessionState {
         self.function_factory.as_ref()
     }
 
+    /// Register a [`CacheFactory`] for custom caching strategy
+    pub fn set_cache_factory(&mut self, cache_factory: Arc<dyn CacheFactory>) {
+        self.cache_factory = Some(cache_factory);
+    }
+
+    /// Get the cache factory
+    pub fn cache_factory(&self) -> Option<&Arc<dyn CacheFactory>> {
+        self.cache_factory.as_ref()
+    }
+
     /// Get the table factories
     pub fn table_factories(&self) -> &HashMap<String, Arc<dyn TableProviderFactory>> {
         &self.table_factories
@@ -941,6 +953,7 @@ pub struct SessionStateBuilder {
     table_factories: Option<HashMap<String, Arc<dyn TableProviderFactory>>>,
     runtime_env: Option<Arc<RuntimeEnv>>,
     function_factory: Option<Arc<dyn FunctionFactory>>,
+    cache_factory: Option<Arc<dyn CacheFactory>>,
     // fields to support convenience functions
     analyzer_rules: Option<Vec<Arc<dyn AnalyzerRule + Send + Sync>>>,
     optimizer_rules: Option<Vec<Arc<dyn OptimizerRule + Send + Sync>>>,
@@ -978,6 +991,7 @@ impl SessionStateBuilder {
             table_factories: None,
             runtime_env: None,
             function_factory: None,
+            cache_factory: None,
             // fields to support convenience functions
             analyzer_rules: None,
             optimizer_rules: None,
@@ -1030,7 +1044,7 @@ impl SessionStateBuilder {
             table_factories: Some(existing.table_factories),
             runtime_env: Some(existing.runtime_env),
             function_factory: existing.function_factory,
-
+            cache_factory: existing.cache_factory,
             // fields to support convenience functions
             analyzer_rules: None,
             optimizer_rules: None,
@@ -1319,6 +1333,15 @@ impl SessionStateBuilder {
         self
     }
 
+    /// Set a [`CacheFactory`] for custom caching strategy
+    pub fn with_cache_factory(
+        mut self,
+        cache_factory: Option<Arc<dyn CacheFactory>>,
+    ) -> Self {
+        self.cache_factory = cache_factory;
+        self
+    }
+
     /// Register an `ObjectStore` to the [`RuntimeEnv`]. See [`RuntimeEnv::register_object_store`]
     /// for more details.
     ///
@@ -1382,6 +1405,7 @@ impl SessionStateBuilder {
             table_factories,
             runtime_env,
             function_factory,
+            cache_factory,
             analyzer_rules,
             optimizer_rules,
             physical_optimizer_rules,
@@ -1418,6 +1442,7 @@ impl SessionStateBuilder {
             table_factories: table_factories.unwrap_or_default(),
             runtime_env,
             function_factory,
+            cache_factory,
             prepared_plans: HashMap::new(),
         };
 
@@ -1621,6 +1646,11 @@ impl SessionStateBuilder {
         &mut self.function_factory
     }
 
+    /// Returns the cache factory
+    pub fn cache_factory(&mut self) -> &mut Option<Arc<dyn CacheFactory>> {
+        &mut self.cache_factory
+    }
+
     /// Returns the current analyzer_rules value
     pub fn analyzer_rules(
         &mut self,
@@ -1659,6 +1689,7 @@ impl Debug for SessionStateBuilder {
             .field("table_options", &self.table_options)
             .field("table_factories", &self.table_factories)
             .field("function_factory", &self.function_factory)
+            .field("cache_factory", &self.cache_factory)
             .field("expr_planners", &self.expr_planners);
         #[cfg(feature = "sql")]
         let ret = ret.field("type_planner", &self.type_planner);
@@ -2047,6 +2078,19 @@ pub(crate) struct PreparedPlan {
     pub(crate) plan: Arc<LogicalPlan>,
 }
 
+/// A [`CacheFactory`] can be registered via [`SessionState`]
+/// to create a custom logical plan for [`crate::dataframe::DataFrame::cache`].
+/// Additionally, a custom [`crate::physical_planner::ExtensionPlanner`]/[`QueryPlanner`]
+/// may need to be implemented to handle such plans.
+pub trait CacheFactory: Debug + Send + Sync {
+    /// Create a logical plan for caching
+    fn create(
+        &self,
+        plan: LogicalPlan,
+        session_state: &SessionState,
+    ) -> datafusion_common::Result<LogicalPlan>;
+}
+
 #[cfg(test)]
 mod tests {
     use super::{SessionContextProvider, SessionStateBuilder};
diff --git a/datafusion/core/src/test_util/mod.rs b/datafusion/core/src/test_util/mod.rs
@@ -25,6 +25,7 @@ pub mod csv;
 use futures::Stream;
 use std::any::Any;
 use std::collections::HashMap;
+use std::fmt::Formatter;
 use std::fs::File;
 use std::io::Write;
 use std::path::Path;
@@ -36,16 +37,20 @@ use crate::dataframe::DataFrame;
 use crate::datasource::stream::{FileStreamProvider, StreamConfig, StreamTable};
 use crate::datasource::{empty::EmptyTable, provider_as_source};
 use crate::error::Result;
+use crate::execution::session_state::CacheFactory;
 use crate::logical_expr::{LogicalPlanBuilder, UNNAMED_TABLE};
 use crate::physical_plan::ExecutionPlan;
 use crate::prelude::{CsvReadOptions, SessionContext};
 
-use crate::execution::SendableRecordBatchStream;
+use crate::execution::{SendableRecordBatchStream, SessionState, SessionStateBuilder};
 use arrow::datatypes::{DataType, Field, Schema, SchemaRef};
 use arrow::record_batch::RecordBatch;
 use datafusion_catalog::Session;
-use datafusion_common::TableReference;
-use datafusion_expr::{CreateExternalTable, Expr, SortExpr, TableType};
+use datafusion_common::{DFSchemaRef, TableReference};
+use datafusion_expr::{
+    CreateExternalTable, Expr, LogicalPlan, SortExpr, TableType,
+    UserDefinedLogicalNodeCore,
+};
 use std::pin::Pin;
 
 use async_trait::async_trait;
@@ -282,3 +287,67 @@ impl RecordBatchStream for BoundedStream {
         self.record_batch.schema()
     }
 }
+
+#[derive(Hash, Eq, PartialEq, PartialOrd, Debug)]
+struct CacheNode {
+    input: LogicalPlan,
+}
+
+impl UserDefinedLogicalNodeCore for CacheNode {
+    fn name(&self) -> &str {
+        "CacheNode"
+    }
+
+    fn inputs(&self) -> Vec<&LogicalPlan> {
+        vec![&self.input]
+    }
+
+    fn schema(&self) -> &DFSchemaRef {
+        self.input.schema()
+    }
+
+    fn expressions(&self) -> Vec<Expr> {
+        vec![]
+    }
+
+    fn fmt_for_explain(&self, f: &mut Formatter) -> std::fmt::Result {
+        write!(f, "CacheNode")
+    }
+
+    fn with_exprs_and_inputs(
+        &self,
+        _exprs: Vec<Expr>,
+        inputs: Vec<LogicalPlan>,
+    ) -> Result<Self> {
+        assert_eq!(inputs.len(), 1, "input size inconsistent");
+        Ok(Self {
+            input: inputs[0].clone(),
+        })
+    }
+}
+
+#[derive(Debug)]
+struct TestCacheFactory {}
+
+impl CacheFactory for TestCacheFactory {
+    fn create(
+        &self,
+        plan: LogicalPlan,
+        _session_state: &SessionState,
+    ) -> Result<LogicalPlan> {
+        Ok(LogicalPlan::Extension(datafusion_expr::Extension {
+            node: Arc::new(CacheNode { input: plan }),
+        }))
+    }
+}
+
+/// Create a test table registered to a session context with an associated cache factory
+pub async fn test_table_with_cache_factory() -> Result<DataFrame> {
+    let session_state = SessionStateBuilder::new()
+        .with_cache_factory(Some(Arc::new(TestCacheFactory {})))
+        .build();
+    let ctx = SessionContext::new_with_state(session_state);
+    let name = "aggregate_test_100";
+    register_aggregate_csv(&ctx, name).await?;
+    ctx.table(name).await
+}
diff --git a/datafusion/core/tests/dataframe/mod.rs b/datafusion/core/tests/dataframe/mod.rs
@@ -61,7 +61,7 @@ use datafusion::prelude::{
 };
 use datafusion::test_util::{
     parquet_test_data, populate_csv_partitions, register_aggregate_csv, test_table,
-    test_table_with_name,
+    test_table_with_cache_factory, test_table_with_name,
 };
 use datafusion_catalog::TableProvider;
 use datafusion_common::test_util::{batches_to_sort_string, batches_to_string};
@@ -2335,6 +2335,29 @@ async fn cache_test() -> Result<()> {
     Ok(())
 }
 
+#[tokio::test]
+async fn cache_producer_test() -> Result<()> {
+    let df = test_table_with_cache_factory()
+        .await?
+        .select_columns(&["c2", "c3"])?
+        .limit(0, Some(1))?
+        .with_column("sum", cast(col("c2") + col("c3"), DataType::Int64))?;
+
+    let cached_df = df.clone().cache().await?;
+
+    assert_snapshot!(
+        cached_df.clone().into_optimized_plan().unwrap(),
+        @r###"
+    CacheNode
+      Projection: aggregate_test_100.c2, aggregate_test_100.c3, CAST(CAST(aggregate_test_100.c2 AS Int64) + CAST(aggregate_test_100.c3 AS Int64) AS Int64) AS sum
+        Projection: aggregate_test_100.c2, aggregate_test_100.c3
+          Limit: skip=0, fetch=1
+            TableScan: aggregate_test_100, fetch=1
+    "###
+    );
+    Ok(())
+}
+
 #[tokio::test]
 async fn partition_aware_union() -> Result<()> {
     let left = test_table().await?.select_columns(&["c1", "c2"])?;
