This is an automated email from the ASF dual-hosted git repository. github-merge-queue[bot] pushed a commit to branch gh-readonly-queue/main/pr-23708-774f0bc34ff3dcd70fff537f828557584b8603c6 in repository https://gitbox.apache.org/repos/asf/datafusion.git
commit 02a902c7ff061b574fbf64e504812235558ff161 Author: Phoenix <[email protected]> AuthorDate: Tue Jul 21 13:43:16 2026 +0800 refactor(proto): migrate FilterExec serde (#23708) ## Which issue does this PR close? <!-- We generally require a GitHub issue to be filed for all bug fixes and enhancements and this helps us generate change logs for our releases. You can link an issue to this PR using the GitHub syntax. For example `Closes #123` indicates that this PR will close issue #123. --> - Closes #23499. ## Rationale for this change <!-- Why are you proposing this change? If this is already explained clearly in the issue then this section is not needed. Explaining clearly why changes are proposed helps reviewers understand your changes and offer better suggestions for fixes. --> The centralized protobuf dispatcher couples each execution plan to datafusion-proto. Adding plan serialization therefore requires changes outside the plan implementation. Move FilterExec encoding and decoding behind its ExecutionPlan hooks. Preserve the existing wire format and retain the old helper methods as deprecated compatibility APIs. ## What changes are included in this PR? <!-- There is no need to duplicate the description in the issue here but it is sometimes worth providing a summary of the individual changes in this PR. --> ## Are these changes tested? <!-- We typically require tests for all PRs in order to: 1. Prevent the code from being accidentally broken by subsequent changes 2. Serve as another way to document the expected behavior of the code If tests are not included in your PR, please explain why (for example, are they covered by existing tests)? --> Yes ## Are there any user-facing changes? <!-- If there are user-facing changes then we may require documentation to be updated before approving the PR. --> <!-- If there are any breaking changes to public APIs, please add the `api change` label. --> --------- Signed-off-by: Jiawei Zhao <[email protected]> --- datafusion/physical-plan/src/filter.rs | 95 +++++++++++++++++++ datafusion/proto/src/physical_plan/mod.rs | 104 +++++---------------- .../proto/tests/cases/roundtrip_physical_plan.rs | 33 +++++++ 3 files changed, 152 insertions(+), 80 deletions(-) diff --git a/datafusion/physical-plan/src/filter.rs b/datafusion/physical-plan/src/filter.rs index ad40a3fb5f..d367be16eb 100644 --- a/datafusion/physical-plan/src/filter.rs +++ b/datafusion/physical-plan/src/filter.rs @@ -812,6 +812,101 @@ impl ExecutionPlan for FilterExec { .ok() }) } + + #[cfg(feature = "proto")] + fn try_to_proto( + &self, + ctx: &crate::proto::ExecutionPlanEncodeCtx<'_>, + ) -> Result<Option<datafusion_proto_models::protobuf::PhysicalPlanNode>> { + use datafusion_proto_models::protobuf; + let input = ctx.encode_child(self.input())?; + let expr = ctx.encode_expr(self.predicate())?; + // Preserve the exact wire format: `None` (full projection) is serialized + // as the identity projection `[0, 1, ..., num_fields - 1]` so that it is + // distinguishable from an explicit projection on decode. + let projection = if let Some(v) = self.projection() { + v.iter().map(|x| *x as u32).collect() + } else { + (0..self.input().schema().fields().len()) + .map(|i| i as u32) + .collect() + }; + Ok(Some(protobuf::PhysicalPlanNode { + physical_plan_type: Some( + protobuf::physical_plan_node::PhysicalPlanType::Filter(Box::new( + protobuf::FilterExecNode { + input: Some(Box::new(input)), + expr: Some(expr), + default_filter_selectivity: self.default_selectivity() as u32, + projection, + batch_size: self.batch_size() as u32, + fetch: self.fetch().map(|f| f as u32), + }, + )), + ), + })) + } +} + +#[cfg(feature = "proto")] +impl FilterExec { + /// Reconstruct a [`FilterExec`] from its protobuf representation. + /// + /// The exact inverse of [`ExecutionPlan::try_to_proto`]: it takes the whole + /// [`PhysicalPlanNode`] so every plan's `try_from_proto` shares one signature. + /// + /// [`PhysicalPlanNode`]: datafusion_proto_models::protobuf::PhysicalPlanNode + /// [`ExecutionPlan::try_to_proto`]: crate::ExecutionPlan::try_to_proto + pub fn try_from_proto( + node: &datafusion_proto_models::protobuf::PhysicalPlanNode, + ctx: &crate::proto::ExecutionPlanDecodeCtx<'_>, + ) -> Result<Arc<dyn ExecutionPlan>> { + use datafusion_proto_models::protobuf; + let filter = crate::expect_plan_variant!( + node, + protobuf::physical_plan_node::PhysicalPlanType::Filter, + "FilterExec", + ); + let input = + ctx.decode_required_child(filter.input.as_deref(), "FilterExec", "input")?; + let predicate = ctx.decode_required_expr( + filter.expr.as_ref(), + input.schema().as_ref(), + "FilterExec", + "expr", + )?; + let filter_selectivity = filter.default_filter_selectivity.try_into(); + + // `None` is encoded as the full identity projection. Reconstruct it only + // when all input columns are present in order, leaving an empty list as + // `Some(vec![])`. + let num_fields = input.schema().fields().len(); + let mut is_full_projection = filter.projection.len() == num_fields; + let mut projection_vec: Vec<usize> = Vec::with_capacity(filter.projection.len()); + for (i, idx) in filter.projection.iter().enumerate() { + let idx = *idx as usize; + is_full_projection &= idx == i; + projection_vec.push(idx); + } + let projection = if is_full_projection { + None + } else { + Some(projection_vec) + }; + let filter = FilterExecBuilder::new(predicate, input) + .apply_projection(projection)? + .with_batch_size(filter.batch_size as usize) + .with_fetch(filter.fetch.map(|f| f as usize)) + .build()?; + match filter_selectivity { + Ok(filter_selectivity) => Ok(Arc::new( + filter.with_default_selectivity(filter_selectivity)?, + )), + Err(_) => Err(datafusion_common::internal_datafusion_err!( + "filter_selectivity in PhysicalPlanNode is invalid" + )), + } + } } impl EmbeddedProjection for FilterExec { diff --git a/datafusion/proto/src/physical_plan/mod.rs b/datafusion/proto/src/physical_plan/mod.rs index 1ef8237595..8649630002 100644 --- a/datafusion/proto/src/physical_plan/mod.rs +++ b/datafusion/proto/src/physical_plan/mod.rs @@ -77,7 +77,7 @@ use datafusion_physical_plan::coop::CooperativeExec; use datafusion_physical_plan::empty::EmptyExec; use datafusion_physical_plan::explain::ExplainExec; use datafusion_physical_plan::expressions::PhysicalSortExpr; -use datafusion_physical_plan::filter::{FilterExec, FilterExecBuilder}; +use datafusion_physical_plan::filter::FilterExec; use datafusion_physical_plan::joins::utils::{ColumnIndex, JoinFilter}; use datafusion_physical_plan::joins::{ CrossJoinExec, HashJoinExec, NestedLoopJoinExec, PartitionMode, SortMergeJoinExec, @@ -756,8 +756,8 @@ pub trait PhysicalPlanNodeExt: Sized { PhysicalPlanType::Projection(_) => { ProjectionExec::try_from_proto(self.node(), &decode_ctx) } - PhysicalPlanType::Filter(filter) => { - self.try_into_filter_physical_plan(filter, ctx, proto_converter) + PhysicalPlanType::Filter(_) => { + FilterExec::try_from_proto(self.node(), &decode_ctx) } PhysicalPlanType::CsvScan(scan) => { self.try_into_csv_scan_physical_plan(scan, ctx, proto_converter) @@ -913,14 +913,6 @@ pub trait PhysicalPlanNodeExt: Sized { ); } - if let Some(exec) = plan.downcast_ref::<FilterExec>() { - return protobuf::PhysicalPlanNode::try_from_filter_exec( - exec, - codec, - proto_converter, - ); - } - if let Some(limit) = plan.downcast_ref::<GlobalLimitExec>() { return protobuf::PhysicalPlanNode::try_from_global_limit_exec( limit, @@ -1214,59 +1206,25 @@ pub trait PhysicalPlanNodeExt: Sized { Ok(Arc::new(ProjectionExec::try_new(proj_exprs, input)?)) } + #[deprecated( + since = "55.0.0", + note = "unused by DataFusion; `FilterExec` deserializes itself via `FilterExec::try_from_proto`" + )] fn try_into_filter_physical_plan( &self, filter: &protobuf::FilterExecNode, ctx: &PhysicalPlanDecodeContext<'_>, proto_converter: &dyn PhysicalProtoConverterExtension, ) -> Result<Arc<dyn ExecutionPlan>> { - let input: Arc<dyn ExecutionPlan> = - into_physical_plan(&filter.input, ctx, proto_converter)?; - - let predicate = filter - .expr - .as_ref() - .map(|expr| { - proto_converter.proto_to_physical_expr(expr, input.schema().as_ref(), ctx) - }) - .transpose()? - .ok_or_else(|| { - internal_datafusion_err!( - "filter (FilterExecNode) in PhysicalPlanNode is missing." - ) - })?; - - let filter_selectivity = filter.default_filter_selectivity.try_into(); - // Preserve the `None` state across proto boundaries. Proto cannot distinguish - // between `None` (full projection) and `Some(vec![])` (empty projection) since - // both serialize as an empty list. If all columns are included, we reconstruct - // `None` to avoid losing this semantic distinction on deserialization. - let num_fields = input.schema().fields().len(); - let mut is_full_projection = filter.projection.len() == num_fields; - let mut projection_vec: Vec<usize> = Vec::with_capacity(filter.projection.len()); - for (i, idx) in filter.projection.iter().enumerate() { - let idx = *idx as usize; - is_full_projection &= idx == i; - projection_vec.push(idx); - } - let projection = if is_full_projection { - None - } else { - Some(projection_vec) + let node = protobuf::PhysicalPlanNode { + physical_plan_type: Some(PhysicalPlanType::Filter(Box::new(filter.clone()))), }; - let filter = FilterExecBuilder::new(predicate, input) - .apply_projection(projection)? - .with_batch_size(filter.batch_size as usize) - .with_fetch(filter.fetch.map(|f| f as usize)) - .build()?; - match filter_selectivity { - Ok(filter_selectivity) => Ok(Arc::new( - filter.with_default_selectivity(filter_selectivity)?, - )), - Err(_) => Err(internal_datafusion_err!( - "filter_selectivity in PhysicalPlanNode is invalid " - )), - } + let decoder = ConverterPlanDecoder { + ctx, + proto_converter, + }; + let decode_ctx = ExecutionPlanDecodeCtx::new(&decoder); + FilterExec::try_from_proto(&node, &decode_ctx) } fn try_into_csv_scan_physical_plan( @@ -2963,36 +2921,22 @@ pub trait PhysicalPlanNodeExt: Sized { }) } + #[deprecated( + since = "55.0.0", + note = "unused by DataFusion; `FilterExec` serializes itself via `ExecutionPlan::try_to_proto`" + )] fn try_from_filter_exec( exec: &FilterExec, codec: &dyn PhysicalExtensionCodec, proto_converter: &dyn PhysicalProtoConverterExtension, ) -> Result<protobuf::PhysicalPlanNode> { - let input = protobuf::PhysicalPlanNode::try_from_physical_plan_with_converter( - exec.input().to_owned(), + let encoder = ConverterPlanEncoder { codec, proto_converter, - )?; - Ok(protobuf::PhysicalPlanNode { - physical_plan_type: Some(PhysicalPlanType::Filter(Box::new( - protobuf::FilterExecNode { - input: Some(Box::new(input)), - expr: Some( - proto_converter - .physical_expr_to_proto(exec.predicate(), codec)?, - ), - default_filter_selectivity: exec.default_selectivity() as u32, - projection: match exec.projection() { - None => (0..exec.input().schema().fields().len()) - .map(|i| i as u32) - .collect(), - Some(v) => v.iter().map(|x| *x as u32).collect(), - }, - batch_size: exec.batch_size() as u32, - fetch: exec.fetch().map(|f| f as u32), - }, - ))), - }) + }; + let encode_ctx = ExecutionPlanEncodeCtx::new(&encoder); + exec.try_to_proto(&encode_ctx)? + .ok_or_else(|| internal_datafusion_err!("FilterExec is not serializable")) } fn try_from_global_limit_exec( diff --git a/datafusion/proto/tests/cases/roundtrip_physical_plan.rs b/datafusion/proto/tests/cases/roundtrip_physical_plan.rs index 28e70f2ddf..8b7223ddd7 100644 --- a/datafusion/proto/tests/cases/roundtrip_physical_plan.rs +++ b/datafusion/proto/tests/cases/roundtrip_physical_plan.rs @@ -1028,6 +1028,39 @@ fn roundtrip_filter_with_fetch() -> Result<()> { roundtrip_test(Arc::new(filter)) } +#[test] +fn roundtrip_filter_projection_states() -> Result<()> { + let schema = Arc::new(Schema::new(vec![ + Field::new("a", DataType::Boolean, false), + Field::new("b", DataType::Int64, false), + ])); + let ctx = SessionContext::new(); + let codec = DefaultPhysicalExtensionCodec {}; + let proto_converter = DefaultPhysicalProtoConverter {}; + + for projection in [None, Some(vec![]), Some(vec![0])] { + let filter = FilterExecBuilder::new( + col("a", &schema)?, + Arc::new(EmptyExec::new(Arc::clone(&schema))), + ) + .apply_projection(projection.clone())? + .with_default_selectivity(37) + .with_batch_size(1024) + .with_fetch(Some(5)) + .build()?; + + let result = + roundtrip_test_and_return(Arc::new(filter), &ctx, &codec, &proto_converter)?; + let result = result.downcast_ref::<FilterExec>().unwrap(); + assert_eq!(result.projection().as_deref(), projection.as_deref()); + assert_eq!(result.default_selectivity(), 37); + assert_eq!(result.batch_size(), 1024); + assert_eq!(result.fetch(), Some(5)); + } + + Ok(()) +} + #[test] fn roundtrip_sort() -> Result<()> { let field_a = Field::new("a", DataType::Boolean, false); --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
