grafeo_core/execution/operators/
mod.rs

1//! Physical operators that actually execute queries.
2//!
3//! These are the building blocks of query execution. The optimizer picks which
4//! operators to use and how to wire them together.
5//!
6//! **Graph operators:**
7//! - [`ScanOperator`] - Read nodes/edges from storage
8//! - [`ExpandOperator`] - Traverse edges (the core of graph queries)
9//! - [`VariableLengthExpandOperator`] - Paths of variable length
10//! - [`ShortestPathOperator`] - Find shortest paths
11//!
12//! **Relational operators:**
13//! - [`FilterOperator`] - Apply predicates
14//! - [`ProjectOperator`] - Select/transform columns
15//! - [`HashJoinOperator`] - Efficient equi-joins
16//! - [`HashAggregateOperator`] - Group by with aggregation
17//! - [`SortOperator`] - Order results
18//! - [`LimitOperator`] - SKIP and LIMIT
19//!
20//! The [`push`] submodule has push-based variants for pipeline execution.
21
22pub mod accumulator;
23mod aggregate;
24mod apply;
25mod distinct;
26mod expand;
27mod factorized_aggregate;
28mod factorized_expand;
29mod factorized_filter;
30mod filter;
31mod horizontal_aggregate;
32mod join;
33mod leapfrog_join;
34mod limit;
35mod load_data;
36mod map_collect;
37mod merge;
38mod mutation;
39mod parameter_scan;
40mod project;
41pub mod push;
42mod scan;
43mod scan_vector;
44mod set_ops;
45mod shortest_path;
46pub mod single_row;
47mod sort;
48mod union;
49mod unwind;
50pub mod value_utils;
51mod variable_length_expand;
52mod vector_join;
53
54pub use accumulator::{AggregateExpr, AggregateFunction, HashableValue};
55pub use aggregate::{HashAggregateOperator, SimpleAggregateOperator};
56pub use apply::ApplyOperator;
57pub use distinct::DistinctOperator;
58pub use expand::ExpandOperator;
59pub use factorized_aggregate::{
60    FactorizedAggregate, FactorizedAggregateOperator, FactorizedOperator,
61};
62pub use factorized_expand::{
63    ExpandStep, FactorizedExpandChain, FactorizedExpandOperator, FactorizedResult,
64    LazyFactorizedChainOperator,
65};
66pub use factorized_filter::{
67    AndPredicate, ColumnPredicate, CompareOp as FactorizedCompareOp, FactorizedFilterOperator,
68    FactorizedPredicate, OrPredicate, PropertyPredicate,
69};
70pub use filter::{
71    BinaryFilterOp, ExpressionPredicate, FilterExpression, FilterOperator, LazyValue,
72    ListPredicateKind, Predicate, SessionContext, UnaryFilterOp,
73};
74pub use horizontal_aggregate::{EntityKind, HorizontalAggregateOperator};
75pub use join::{
76    EqualityCondition, HashJoinOperator, HashKey, JoinCondition, JoinType, NestedLoopJoinOperator,
77};
78pub use leapfrog_join::LeapfrogJoinOperator;
79pub use limit::{LimitOperator, LimitSkipOperator, SkipOperator};
80pub use load_data::{LoadDataFormat, LoadDataOperator};
81pub use map_collect::MapCollectOperator;
82pub use merge::{MergeConfig, MergeOperator, MergeRelationshipConfig, MergeRelationshipOperator};
83pub use mutation::{
84    AddLabelOperator, ConstraintValidator, CreateEdgeOperator, CreateNodeOperator,
85    DeleteEdgeOperator, DeleteNodeOperator, PropertySource, RemoveLabelOperator,
86    SetPropertyOperator,
87};
88pub use parameter_scan::{ParameterScanOperator, ParameterState};
89pub use project::{ProjectExpr, ProjectOperator};
90pub use push::{
91    AggregatePushOperator, DistinctMaterializingOperator, DistinctPushOperator, FilterPushOperator,
92    LimitPushOperator, ProjectPushOperator, SkipLimitPushOperator, SkipPushOperator,
93    SortPushOperator,
94};
95#[cfg(feature = "spill")]
96pub use push::{SpillableAggregatePushOperator, SpillableSortPushOperator};
97pub use scan::ScanOperator;
98pub use scan_vector::VectorScanOperator;
99pub use set_ops::{ExceptOperator, IntersectOperator, OtherwiseOperator};
100pub use shortest_path::ShortestPathOperator;
101pub use single_row::{EmptyOperator, NodeListOperator, SingleRowOperator};
102pub use sort::{NullOrder, SortDirection, SortKey, SortOperator};
103pub use union::UnionOperator;
104pub use unwind::UnwindOperator;
105pub use variable_length_expand::{PathMode as ExecutionPathMode, VariableLengthExpandOperator};
106pub use vector_join::VectorJoinOperator;
107
108use std::sync::Arc;
109
110use grafeo_common::types::{EdgeId, NodeId, TransactionId};
111use thiserror::Error;
112
113use super::DataChunk;
114use super::chunk_state::ChunkState;
115use super::factorized_chunk::FactorizedChunk;
116
117/// Trait for recording write operations during query execution.
118///
119/// This bridges `grafeo-core` mutation operators (which perform writes) with
120/// `grafeo-engine`'s `TransactionManager` (which tracks write sets for conflict
121/// detection). The trait lives in `grafeo-core` to avoid circular dependencies.
122pub trait WriteTracker: Send + Sync {
123    /// Records that a node was written (created, deleted, or modified).
124    fn record_node_write(&self, transaction_id: TransactionId, node_id: NodeId);
125
126    /// Records that an edge was written (created, deleted, or modified).
127    fn record_edge_write(&self, transaction_id: TransactionId, edge_id: EdgeId);
128}
129
130/// Type alias for a shared write tracker.
131pub type SharedWriteTracker = Arc<dyn WriteTracker>;
132
133/// Result of executing an operator.
134pub type OperatorResult = Result<Option<DataChunk>, OperatorError>;
135
136// ============================================================================
137// Factorized Data Traits
138// ============================================================================
139
140/// Trait for data that can be in factorized or flat form.
141///
142/// This provides a common interface for operators that need to handle both
143/// representations without caring which is used. Inspired by LadybugDB's
144/// unified data model.
145///
146/// # Example
147///
148/// ```rust
149/// use grafeo_core::execution::operators::FactorizedData;
150///
151/// fn process_data(data: &dyn FactorizedData) {
152///     if data.is_factorized() {
153///         // Handle factorized path
154///         let chunk = data.as_factorized().unwrap();
155///         // ... use factorized chunk directly
156///     } else {
157///         // Handle flat path
158///         let chunk = data.flatten();
159///         // ... process flat chunk
160///     }
161/// }
162/// ```
163pub trait FactorizedData: Send + Sync {
164    /// Returns the chunk state (factorization status, cached data).
165    fn chunk_state(&self) -> &ChunkState;
166
167    /// Returns the logical row count (considering selection).
168    fn logical_row_count(&self) -> usize;
169
170    /// Returns the physical size (actual stored values).
171    fn physical_size(&self) -> usize;
172
173    /// Returns true if this data is factorized (multi-level).
174    fn is_factorized(&self) -> bool;
175
176    /// Flattens to a DataChunk (materializes if factorized).
177    fn flatten(&self) -> DataChunk;
178
179    /// Returns as FactorizedChunk if factorized, None if flat.
180    fn as_factorized(&self) -> Option<&FactorizedChunk>;
181
182    /// Returns as DataChunk if flat, None if factorized.
183    fn as_flat(&self) -> Option<&DataChunk>;
184}
185
186/// Wrapper to treat a flat DataChunk as FactorizedData.
187///
188/// This enables uniform handling of flat and factorized data in operators.
189pub struct FlatDataWrapper {
190    chunk: DataChunk,
191    state: ChunkState,
192}
193
194impl FlatDataWrapper {
195    /// Creates a new wrapper around a flat DataChunk.
196    #[must_use]
197    pub fn new(chunk: DataChunk) -> Self {
198        let state = ChunkState::flat(chunk.row_count());
199        Self { chunk, state }
200    }
201
202    /// Returns the underlying DataChunk.
203    #[must_use]
204    pub fn into_inner(self) -> DataChunk {
205        self.chunk
206    }
207}
208
209impl FactorizedData for FlatDataWrapper {
210    fn chunk_state(&self) -> &ChunkState {
211        &self.state
212    }
213
214    fn logical_row_count(&self) -> usize {
215        self.chunk.row_count()
216    }
217
218    fn physical_size(&self) -> usize {
219        self.chunk.row_count() * self.chunk.column_count()
220    }
221
222    fn is_factorized(&self) -> bool {
223        false
224    }
225
226    fn flatten(&self) -> DataChunk {
227        self.chunk.clone()
228    }
229
230    fn as_factorized(&self) -> Option<&FactorizedChunk> {
231        None
232    }
233
234    fn as_flat(&self) -> Option<&DataChunk> {
235        Some(&self.chunk)
236    }
237}
238
239/// Error during operator execution.
240#[derive(Error, Debug, Clone)]
241pub enum OperatorError {
242    /// Type mismatch during execution.
243    #[error("type mismatch: expected {expected}, found {found}")]
244    TypeMismatch {
245        /// Expected type name.
246        expected: String,
247        /// Found type name.
248        found: String,
249    },
250    /// Column not found.
251    #[error("column not found: {0}")]
252    ColumnNotFound(String),
253    /// Execution error.
254    #[error("execution error: {0}")]
255    Execution(String),
256    /// Schema constraint violation during a write operation.
257    #[error("constraint violation: {0}")]
258    ConstraintViolation(String),
259}
260
261/// The core trait for pull-based operators.
262///
263/// Call [`next()`](Self::next) repeatedly until it returns `None`. Each call
264/// returns a batch of rows (a DataChunk) or an error.
265pub trait Operator: Send + Sync {
266    /// Pulls the next batch of data. Returns `None` when exhausted.
267    fn next(&mut self) -> OperatorResult;
268
269    /// Resets to initial state so you can iterate again.
270    fn reset(&mut self);
271
272    /// Returns a name for debugging/explain output.
273    fn name(&self) -> &'static str;
274}
275
276#[cfg(test)]
277mod tests {
278    use super::*;
279    use crate::execution::vector::ValueVector;
280    use grafeo_common::types::LogicalType;
281
282    fn create_test_chunk() -> DataChunk {
283        let mut col = ValueVector::with_type(LogicalType::Int64);
284        col.push_int64(1);
285        col.push_int64(2);
286        col.push_int64(3);
287        DataChunk::new(vec![col])
288    }
289
290    #[test]
291    fn test_flat_data_wrapper_new() {
292        let chunk = create_test_chunk();
293        let wrapper = FlatDataWrapper::new(chunk);
294
295        assert!(!wrapper.is_factorized());
296        assert_eq!(wrapper.logical_row_count(), 3);
297    }
298
299    #[test]
300    fn test_flat_data_wrapper_into_inner() {
301        let chunk = create_test_chunk();
302        let wrapper = FlatDataWrapper::new(chunk);
303
304        let inner = wrapper.into_inner();
305        assert_eq!(inner.row_count(), 3);
306    }
307
308    #[test]
309    fn test_flat_data_wrapper_chunk_state() {
310        let chunk = create_test_chunk();
311        let wrapper = FlatDataWrapper::new(chunk);
312
313        let state = wrapper.chunk_state();
314        assert!(state.is_flat());
315        assert_eq!(state.logical_row_count(), 3);
316    }
317
318    #[test]
319    fn test_flat_data_wrapper_physical_size() {
320        let mut col1 = ValueVector::with_type(LogicalType::Int64);
321        col1.push_int64(1);
322        col1.push_int64(2);
323
324        let mut col2 = ValueVector::with_type(LogicalType::String);
325        col2.push_string("a");
326        col2.push_string("b");
327
328        let chunk = DataChunk::new(vec![col1, col2]);
329        let wrapper = FlatDataWrapper::new(chunk);
330
331        // 2 rows * 2 columns = 4
332        assert_eq!(wrapper.physical_size(), 4);
333    }
334
335    #[test]
336    fn test_flat_data_wrapper_flatten() {
337        let chunk = create_test_chunk();
338        let wrapper = FlatDataWrapper::new(chunk);
339
340        let flattened = wrapper.flatten();
341        assert_eq!(flattened.row_count(), 3);
342        assert_eq!(flattened.column(0).unwrap().get_int64(0), Some(1));
343    }
344
345    #[test]
346    fn test_flat_data_wrapper_as_factorized() {
347        let chunk = create_test_chunk();
348        let wrapper = FlatDataWrapper::new(chunk);
349
350        assert!(wrapper.as_factorized().is_none());
351    }
352
353    #[test]
354    fn test_flat_data_wrapper_as_flat() {
355        let chunk = create_test_chunk();
356        let wrapper = FlatDataWrapper::new(chunk);
357
358        let flat = wrapper.as_flat();
359        assert!(flat.is_some());
360        assert_eq!(flat.unwrap().row_count(), 3);
361    }
362
363    #[test]
364    fn test_operator_error_type_mismatch() {
365        let err = OperatorError::TypeMismatch {
366            expected: "Int64".to_string(),
367            found: "String".to_string(),
368        };
369
370        let msg = format!("{err}");
371        assert!(msg.contains("type mismatch"));
372        assert!(msg.contains("Int64"));
373        assert!(msg.contains("String"));
374    }
375
376    #[test]
377    fn test_operator_error_column_not_found() {
378        let err = OperatorError::ColumnNotFound("missing_col".to_string());
379
380        let msg = format!("{err}");
381        assert!(msg.contains("column not found"));
382        assert!(msg.contains("missing_col"));
383    }
384
385    #[test]
386    fn test_operator_error_execution() {
387        let err = OperatorError::Execution("something went wrong".to_string());
388
389        let msg = format!("{err}");
390        assert!(msg.contains("execution error"));
391        assert!(msg.contains("something went wrong"));
392    }
393
394    #[test]
395    fn test_operator_error_debug() {
396        let err = OperatorError::TypeMismatch {
397            expected: "Int64".to_string(),
398            found: "String".to_string(),
399        };
400
401        let debug = format!("{err:?}");
402        assert!(debug.contains("TypeMismatch"));
403    }
404
405    #[test]
406    fn test_operator_error_clone() {
407        let err1 = OperatorError::ColumnNotFound("col".to_string());
408        let err2 = err1.clone();
409
410        assert_eq!(format!("{err1}"), format!("{err2}"));
411    }
412}
grafeo_core/execution/operators/mod.rs

grafeo_core/execution/operators/
mod.rs
