query/dist_plan/
planner.rs

1// Copyright 2023 Greptime Team
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15//! [ExtensionPlanner] implementation for distributed planner
16
17use std::sync::Arc;
18
19use ahash::HashMap;
20use arrow_schema::SortOptions;
21use async_trait::async_trait;
22use catalog::CatalogManagerRef;
23use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
24use common_telemetry::debug;
25use datafusion::catalog::Session;
26use datafusion::common::Result;
27use datafusion::datasource::DefaultTableSource;
28use datafusion::execution::context::SessionState;
29use datafusion::logical_expr::physical_planning_context::PhysicalPlanningContext;
30use datafusion::physical_plan::ExecutionPlan;
31use datafusion::physical_planner::{ExtensionPlanner, PhysicalPlanner};
32use datafusion_common::tree_node::{TreeNode, TreeNodeRecursion, TreeNodeVisitor};
33use datafusion_common::{DataFusionError, TableReference};
34use datafusion_expr::{LogicalPlan, UserDefinedLogicalNode};
35use datafusion_physical_expr::{LexOrdering, PhysicalSortExpr};
36use datatypes::prelude::ConcreteDataType;
37use partition::expr::PartitionExpr;
38use partition::manager::{PartitionRuleManagerRef, create_partitions_from_region_routes};
39use session::context::QueryContext;
40use snafu::{OptionExt, ResultExt};
41use store_api::storage::RegionId;
42use table::TableRef;
43use table::metadata::TableInfo;
44pub use table::metadata::TableType;
45use table::table::adapter::DfTableProviderAdapter;
46use table::table_name::TableName;
47
48use crate::dist_plan::PredicateExtractor;
49use crate::dist_plan::merge_scan::{MergeScanExec, MergeScanLogicalPlan};
50use crate::dist_plan::merge_sort::{MergeSortExec, MergeSortLogicalPlan};
51use crate::dist_plan::region_pruner::ConstraintPruner;
52use crate::error::{CatalogSnafu, PartitionRuleManagerSnafu, TableNotFoundSnafu};
53use crate::region_query::RegionQueryHandlerRef;
54
55/// Planner for converting merge sort logical plan to physical plan.
56///
57/// `MergeSortExec` always represents the distributed merge stage. It declares
58/// the required input ordering to DataFusion, so `EnforceSorting` inserts a
59/// `SortExec` below it when the input `MergeScanExec` cannot preserve per-region
60/// ordering, for example when one output partition may merge multiple region
61/// streams.
62pub struct MergeSortExtensionPlanner {}
63
64impl MergeSortExtensionPlanner {
65    fn ordering(
66        planner: &dyn PhysicalPlanner,
67        session: &dyn Session,
68        planning_ctx: &PhysicalPlanningContext,
69        merge_sort: &MergeSortLogicalPlan,
70    ) -> Result<LexOrdering> {
71        let ordering = merge_sort
72            .expr
73            .iter()
74            .map(|sort_expr| {
75                let physical_expr = planner.create_physical_expr(
76                    &sort_expr.expr,
77                    merge_sort.input.schema(),
78                    session,
79                    planning_ctx,
80                )?;
81                Ok(PhysicalSortExpr::new(
82                    physical_expr,
83                    SortOptions {
84                        descending: !sort_expr.asc,
85                        nulls_first: sort_expr.nulls_first,
86                    },
87                ))
88            })
89            .collect::<Result<Vec<_>>>()?;
90
91        LexOrdering::new(ordering).ok_or_else(|| {
92            DataFusionError::Internal(
93                "Expect MergeSort to have non-empty sort expressions".to_string(),
94            )
95        })
96    }
97}
98
99#[async_trait]
100impl ExtensionPlanner for MergeSortExtensionPlanner {
101    async fn plan_extension(
102        &self,
103        planner: &dyn PhysicalPlanner,
104        node: &dyn UserDefinedLogicalNode,
105        _logical_inputs: &[&LogicalPlan],
106        physical_inputs: &[Arc<dyn ExecutionPlan>],
107        session: &dyn Session,
108        planning_ctx: &PhysicalPlanningContext,
109    ) -> Result<Option<Arc<dyn ExecutionPlan>>> {
110        if let Some(merge_sort) = node.as_any().downcast_ref::<MergeSortLogicalPlan>() {
111            if let LogicalPlan::Extension(ext) = &merge_sort.input.as_ref()
112                && ext
113                    .node
114                    .as_any()
115                    .downcast_ref::<MergeScanLogicalPlan>()
116                    .is_some()
117            {
118                let input = physical_inputs.first().cloned().ok_or_else(|| {
119                    DataFusionError::Internal(
120                        "Expect MergeSort to have one physical input".to_string(),
121                    )
122                })?;
123                if input.downcast_ref::<MergeScanExec>().is_none() {
124                    return Err(DataFusionError::Internal(format!(
125                        "Expect MergeSort's input is a MergeScanExec, found {:?}",
126                        physical_inputs
127                    )));
128                }
129
130                let ordering = Self::ordering(planner, session, planning_ctx, merge_sort)?;
131                Ok(Some(Arc::new(MergeSortExec::new(
132                    ordering,
133                    input,
134                    merge_sort.fetch,
135                ))))
136            } else {
137                Ok(None)
138            }
139        } else {
140            Ok(None)
141        }
142    }
143}
144
145pub struct DistExtensionPlanner {
146    catalog_manager: CatalogManagerRef,
147    partition_rule_manager: PartitionRuleManagerRef,
148    region_query_handler: RegionQueryHandlerRef,
149    enable_per_region_metrics: bool,
150}
151
152impl DistExtensionPlanner {
153    pub fn new(
154        catalog_manager: CatalogManagerRef,
155        partition_rule_manager: PartitionRuleManagerRef,
156        region_query_handler: RegionQueryHandlerRef,
157        enable_per_region_metrics: bool,
158    ) -> Self {
159        Self {
160            catalog_manager,
161            partition_rule_manager,
162            region_query_handler,
163            enable_per_region_metrics,
164        }
165    }
166}
167
168#[async_trait]
169impl ExtensionPlanner for DistExtensionPlanner {
170    async fn plan_extension(
171        &self,
172        planner: &dyn PhysicalPlanner,
173        node: &dyn UserDefinedLogicalNode,
174        _logical_inputs: &[&LogicalPlan],
175        _physical_inputs: &[Arc<dyn ExecutionPlan>],
176        session: &dyn Session,
177        _planning_ctx: &PhysicalPlanningContext,
178    ) -> Result<Option<Arc<dyn ExecutionPlan>>> {
179        let Some(merge_scan) = node.as_any().downcast_ref::<MergeScanLogicalPlan>() else {
180            return Ok(None);
181        };
182
183        let input_plan = merge_scan.input();
184        let fallback = |logical_plan| async move {
185            let optimized_plan = self.optimize_input_logical_plan(session, logical_plan)?;
186            planner
187                .create_physical_plan(&optimized_plan, session)
188                .await
189                .map(Some)
190        };
191
192        if merge_scan.is_placeholder() {
193            // ignore placeholder
194            return fallback(input_plan).await;
195        }
196
197        let optimized_plan = input_plan;
198        let Some(table_name) = Self::extract_full_table_name(input_plan)? else {
199            // no relation found in input plan, going to execute them locally
200            return fallback(optimized_plan).await;
201        };
202
203        let Ok(regions) = self.get_regions(&table_name, input_plan).await else {
204            // no peers found, going to execute them locally
205            return fallback(optimized_plan).await;
206        };
207
208        // TODO(ruihang): generate different execution plans for different variant merge operation
209        let schema = merge_scan.schema().as_arrow();
210        let session_state = session
211            .as_any()
212            .downcast_ref::<SessionState>()
213            .ok_or_else(|| {
214                DataFusionError::Internal(
215                    "MergeScan requires a SessionState for physical planning".to_string(),
216                )
217            })?;
218        let query_ctx = session_state
219            .config()
220            .get_extension()
221            .unwrap_or_else(QueryContext::arc);
222        let merge_scan_plan = MergeScanExec::new(
223            session_state,
224            table_name,
225            regions,
226            input_plan.clone(),
227            schema,
228            self.region_query_handler.clone(),
229            query_ctx,
230            session.config().target_partitions(),
231            merge_scan.partition_cols().clone(),
232            merge_scan.remote_dyn_filter_producer_id(),
233            self.enable_per_region_metrics,
234        )?;
235        Ok(Some(Arc::new(merge_scan_plan) as _))
236    }
237}
238
239impl DistExtensionPlanner {
240    /// Extract fully resolved table name from logical plan
241    fn extract_full_table_name(plan: &LogicalPlan) -> Result<Option<TableName>> {
242        let mut extractor = TableScanExtractor::default();
243        let _ = plan.visit(&mut extractor)?;
244        Ok(extractor.table_name)
245    }
246
247    async fn get_regions(
248        &self,
249        table_name: &TableName,
250        logical_plan: &LogicalPlan,
251    ) -> Result<Vec<RegionId>> {
252        let mut extractor = TableScanExtractor::default();
253        let _ = logical_plan.visit(&mut extractor)?;
254        // Resolving by name again could bind an authorized scan to a replacement table.
255        let table = match extractor.captured_table {
256            Some(table) => table,
257            None => self
258                .catalog_manager
259                .table(
260                    &table_name.catalog_name,
261                    &table_name.schema_name,
262                    &table_name.table_name,
263                    None,
264                )
265                .await
266                .context(CatalogSnafu)?
267                .with_context(|| TableNotFoundSnafu {
268                    table: table_name.to_string(),
269                })?,
270        };
271
272        let table_info = table.table_info();
273        let (physical_table_id, physical_table_route) = self
274            .partition_rule_manager
275            .find_physical_table_route_with_id(table_info.table_id())
276            .await
277            .context(PartitionRuleManagerSnafu)?;
278        let all_regions = physical_table_route
279            .region_routes
280            .iter()
281            .map(|r| RegionId::new(table_info.table_id(), r.region.id.region_number()))
282            .collect::<Vec<_>>();
283        let logical_partition_columns = partition_column_types(&table_info);
284        let partition_columns = logical_partition_columns
285            .iter()
286            .map(|(name, _)| name.clone())
287            .collect::<Vec<_>>();
288        debug!(
289            "DistExtensionPlanner: loaded table partition metadata, table: {}, table_id: {}, partition_key_indices: {:?}, partition_columns: {:?}, all_regions: {:?}",
290            table_name,
291            table_info.table_id(),
292            table_info.meta.partition_key_indices,
293            partition_columns,
294            all_regions,
295        );
296        if partition_columns.is_empty() {
297            return Ok(all_regions);
298        }
299        // Extract predicates from logical plan
300        let partition_expressions = match PredicateExtractor::extract_partition_expressions(
301            logical_plan,
302            &partition_columns,
303        ) {
304            Ok(expressions) => expressions,
305            Err(err) => {
306                common_telemetry::debug!(
307                    "Failed to extract partition expressions for table {} (id: {}), using all regions: {:?}",
308                    table_name,
309                    table.table_info().table_id(),
310                    err
311                );
312                return Ok(all_regions);
313            }
314        };
315
316        if partition_expressions.is_empty() {
317            return Ok(all_regions);
318        }
319
320        let Some(partition_column_types) = self
321            .partition_column_types_for_pruning(
322                table_name,
323                table_info.as_ref(),
324                physical_table_id,
325                &partition_expressions,
326                &all_regions,
327            )
328            .await
329        else {
330            return Ok(all_regions);
331        };
332
333        // Get partition information for the table if partition rule manager is available
334        let partitions = match create_partitions_from_region_routes(
335            table_info.table_id(),
336            &physical_table_route.region_routes,
337        ) {
338            Ok(partitions) => partitions,
339            Err(err) => {
340                common_telemetry::debug!(
341                    "Failed to get partition information for table {}, using all regions: {:?}",
342                    table_name,
343                    err
344                );
345                return Ok(all_regions);
346            }
347        };
348        if partitions.is_empty() {
349            return Ok(all_regions);
350        }
351        // Apply region pruning based on partition rules
352        let pruned_regions = match ConstraintPruner::prune_regions(
353            &partition_expressions,
354            &partitions,
355            partition_column_types,
356        ) {
357            Ok(regions) => regions,
358            Err(err) => {
359                common_telemetry::debug!(
360                    "Failed to prune regions for table {}, using all regions: {:?}",
361                    table_name,
362                    err
363                );
364                return Ok(all_regions);
365            }
366        };
367
368        common_telemetry::debug!(
369            "Region pruning for table {}: {} partition expressions applied, pruned from {} to {} regions",
370            table_name,
371            partition_expressions.len(),
372            all_regions.len(),
373            pruned_regions.len()
374        );
375
376        Ok(pruned_regions)
377    }
378
379    /// Resolves the partition-column types that are safe to use for region pruning.
380    ///
381    /// A logical metric table may not contain every physical partition column, either for backward
382    /// compatibility or because its physical table was repartitioned after the logical table was
383    /// created. Predicate extraction must remain bounded by the logical schema, while pruning
384    /// needs the physical datatypes to evaluate route expressions. Any lookup failure or
385    /// logical/physical datatype mismatch returns `None`, causing the caller to scan all regions.
386    async fn partition_column_types_for_pruning(
387        &self,
388        table_name: &TableName,
389        logical_table_info: &TableInfo,
390        physical_table_id: u32,
391        partition_expressions: &[PartitionExpr],
392        all_regions: &[RegionId],
393    ) -> Option<HashMap<String, ConcreteDataType>> {
394        let physical_partition_columns = if physical_table_id == logical_table_info.table_id() {
395            partition_column_types(logical_table_info)
396        } else {
397            match self
398                .catalog_manager
399                .table_info_by_id(physical_table_id)
400                .await
401            {
402                Ok(Some(physical_table_info)) => {
403                    partition_column_types(physical_table_info.as_ref())
404                }
405                Ok(None) => {
406                    debug!(
407                        "DistExtensionPlanner: physical table info not found for table {} (id: {}), using all regions: {:?}",
408                        table_name, physical_table_id, all_regions
409                    );
410                    return None;
411                }
412                Err(err) => {
413                    debug!(
414                        "DistExtensionPlanner: failed to load physical table info for table {} (id: {}): {}, using all regions: {:?}",
415                        table_name, physical_table_id, err, all_regions
416                    );
417                    return None;
418                }
419            }
420        };
421        let physical_column_types = physical_partition_columns
422            .into_iter()
423            .collect::<HashMap<_, _>>();
424        let logical_column_types = partition_column_types(logical_table_info)
425            .into_iter()
426            .collect::<HashMap<_, _>>();
427        let mut predicate_column_names = std::collections::HashSet::new();
428        for expression in partition_expressions {
429            expression.collect_column_names(&mut predicate_column_names);
430        }
431        if predicate_column_names
432            .iter()
433            .any(|name| logical_column_types.get(name) != physical_column_types.get(name))
434        {
435            debug!(
436                "DistExtensionPlanner: logical and physical partition metadata mismatch for table {} (physical id: {}), using all regions: {:?}",
437                table_name, physical_table_id, all_regions
438            );
439            return None;
440        }
441
442        Some(physical_column_types)
443    }
444
445    /// Input logical plan is analyzed. Thus only call logical optimizer to optimize it.
446    fn optimize_input_logical_plan(
447        &self,
448        session: &dyn Session,
449        plan: &LogicalPlan,
450    ) -> Result<LogicalPlan> {
451        let session_state = session
452            .as_any()
453            .downcast_ref::<SessionState>()
454            .ok_or_else(|| {
455                DataFusionError::Internal(
456                    "MergeScan requires a SessionState for logical optimization".to_string(),
457                )
458            })?;
459
460        session_state
461            .optimizer()
462            .optimize(plan.clone(), session_state, |_, _| {})
463    }
464}
465
466fn partition_column_types(table_info: &TableInfo) -> Vec<(String, ConcreteDataType)> {
467    table_info
468        .meta
469        .partition_columns()
470        .map(|column| (column.name.clone(), column.data_type.clone()))
471        .collect()
472}
473
474/// Extract the scan name and captured table identity from a logical plan.
475#[derive(Default)]
476struct TableScanExtractor {
477    pub table_name: Option<TableName>,
478    captured_table: Option<TableRef>,
479}
480
481impl TreeNodeVisitor<'_> for TableScanExtractor {
482    type Node = LogicalPlan;
483
484    fn f_down(&mut self, node: &Self::Node) -> Result<TreeNodeRecursion> {
485        match node {
486            LogicalPlan::TableScan(scan) => {
487                if let Some(source) = scan.source.downcast_ref::<DefaultTableSource>()
488                    && let Some(provider) = source
489                        .table_provider
490                        .downcast_ref::<DfTableProviderAdapter>()
491                {
492                    if provider.table().table_type() == TableType::Base {
493                        self.captured_table = Some(provider.table());
494                        let info = provider.table().table_info();
495                        self.table_name = Some(TableName::new(
496                            info.catalog_name.clone(),
497                            info.schema_name.clone(),
498                            info.name.clone(),
499                        ));
500                    }
501                    return Ok(TreeNodeRecursion::Stop);
502                }
503                match &scan.table_name {
504                    TableReference::Full {
505                        catalog,
506                        schema,
507                        table,
508                    } => {
509                        self.table_name = Some(TableName::new(
510                            catalog.to_string(),
511                            schema.to_string(),
512                            table.to_string(),
513                        ));
514                        Ok(TreeNodeRecursion::Stop)
515                    }
516                    // TODO(ruihang): Maybe the following two cases should not be valid
517                    TableReference::Partial { schema, table } => {
518                        self.table_name = Some(TableName::new(
519                            DEFAULT_CATALOG_NAME.to_string(),
520                            schema.to_string(),
521                            table.to_string(),
522                        ));
523                        Ok(TreeNodeRecursion::Stop)
524                    }
525                    TableReference::Bare { table } => {
526                        self.table_name = Some(TableName::new(
527                            DEFAULT_CATALOG_NAME.to_string(),
528                            DEFAULT_SCHEMA_NAME.to_string(),
529                            table.to_string(),
530                        ));
531                        Ok(TreeNodeRecursion::Stop)
532                    }
533                }
534            }
535            _ => Ok(TreeNodeRecursion::Continue),
536        }
537    }
538}
539
540#[cfg(test)]
541mod tests {
542    use std::collections::HashMap;
543    use std::sync::Arc;
544
545    use api::v1::region::{RemoteDynFilterUnregister, RemoteDynFilterUpdate};
546    use async_trait::async_trait;
547    use catalog::memory::MemoryCatalogManager;
548    use catalog::{CatalogManagerRef, RegisterTableRequest};
549    use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
550    use common_meta::cache::new_table_route_cache;
551    use common_meta::key::TableMetadataManager;
552    use common_meta::key::table_route::TableRouteValue;
553    use common_meta::kv_backend::memory::MemoryKvBackend;
554    use common_meta::rpc::router::{Region, RegionRoute};
555    use common_query::request::QueryRequest;
556    use common_recordbatch::SendableRecordBatchStream;
557    use datafusion::datasource::DefaultTableSource;
558    use datafusion_expr::{LogicalPlan, LogicalPlanBuilder, col as df_col, lit};
559    use datatypes::prelude::ConcreteDataType;
560    use datatypes::schema::{ColumnSchema, Schema};
561    use datatypes::value::Value;
562    use moka::future::CacheBuilder;
563    use partition::cache::new_partition_info_cache;
564    use partition::expr::{PartitionExpr, col as partition_col};
565    use partition::manager::PartitionRuleManager;
566    use session::ReadPreference;
567    use store_api::storage::RegionId;
568    use table::metadata::{TableInfo, TableInfoBuilder, TableMeta, TableType};
569    use table::table::adapter::DfTableProviderAdapter;
570    use table::table_name::TableName;
571    use table::test_util::EmptyTable;
572
573    use super::DistExtensionPlanner;
574    use crate::region_query::{RegionQueryHandler, RegionQueryTarget};
575
576    const LOGICAL_TABLE_ID: u32 = 1024;
577    const PHYSICAL_TABLE_ID: u32 = 2048;
578
579    struct UnusedRegionQueryHandler;
580
581    #[async_trait]
582    impl RegionQueryHandler for UnusedRegionQueryHandler {
583        async fn select_target(
584            &self,
585            _read_preference: ReadPreference,
586            _region_id: RegionId,
587        ) -> crate::error::Result<RegionQueryTarget> {
588            unreachable!("get_regions does not select region query targets")
589        }
590
591        async fn do_get(
592            &self,
593            _target: &RegionQueryTarget,
594            _request: QueryRequest,
595        ) -> crate::error::Result<SendableRecordBatchStream> {
596            unreachable!("get_regions does not query regions")
597        }
598
599        async fn handle_remote_dyn_filter_update(
600            &self,
601            _target: &RegionQueryTarget,
602            _query_id: String,
603            _update: RemoteDynFilterUpdate,
604        ) -> crate::error::Result<()> {
605            unreachable!("get_regions does not update dynamic filters")
606        }
607
608        async fn handle_remote_dyn_filter_unregister(
609            &self,
610            _target: &RegionQueryTarget,
611            _query_id: String,
612            _unregister: RemoteDynFilterUnregister,
613        ) -> crate::error::Result<()> {
614            unreachable!("get_regions does not unregister dynamic filters")
615        }
616    }
617
618    fn table_info(
619        table_id: u32,
620        name: &str,
621        columns: &[&str],
622        partition_keys: Vec<usize>,
623    ) -> TableInfo {
624        let schema = Arc::new(Schema::new(
625            columns
626                .iter()
627                .map(|name| ColumnSchema::new(*name, ConcreteDataType::string_datatype(), true))
628                .collect(),
629        ));
630        let meta = TableMeta {
631            schema,
632            primary_key_indices: vec![],
633            value_indices: vec![],
634            engine: "metric".to_string(),
635            next_column_id: columns.len() as u32,
636            options: Default::default(),
637            created_on: Default::default(),
638            updated_on: Default::default(),
639            partition_key_indices: partition_keys,
640            column_ids: (0..columns.len() as u32).collect(),
641        };
642        TableInfoBuilder::default()
643            .table_id(table_id)
644            .table_version(0)
645            .name(name.to_string())
646            .catalog_name(DEFAULT_CATALOG_NAME.to_string())
647            .schema_name(DEFAULT_SCHEMA_NAME.to_string())
648            .desc(None)
649            .table_type(TableType::Base)
650            .meta(meta)
651            .build()
652            .unwrap()
653    }
654
655    fn region_route(region_number: u32, expression: Option<PartitionExpr>) -> RegionRoute {
656        RegionRoute {
657            region: Region {
658                id: RegionId::new(PHYSICAL_TABLE_ID, region_number),
659                partition_expr: expression
660                    .map(|expression| expression.as_json_str().unwrap())
661                    .unwrap_or_default(),
662                ..Default::default()
663            },
664            ..Default::default()
665        }
666    }
667
668    async fn planner_and_plan(
669        physical_partition_keys: Vec<usize>,
670        expressions: Vec<Option<PartitionExpr>>,
671    ) -> (DistExtensionPlanner, LogicalPlan, TableName) {
672        let logical_info = table_info(LOGICAL_TABLE_ID, "logical", &["host"], vec![0]);
673        let physical_info = table_info(
674            PHYSICAL_TABLE_ID,
675            "physical",
676            &["host", "rack"],
677            physical_partition_keys,
678        );
679        let logical_table = EmptyTable::from_table_info(&logical_info);
680        let physical_table = EmptyTable::from_table_info(&physical_info);
681        let catalog_manager = MemoryCatalogManager::with_default_setup();
682        for table in [&logical_table, &physical_table] {
683            let info = table.table_info();
684            catalog_manager
685                .register_table_sync(RegisterTableRequest {
686                    catalog: info.catalog_name.clone(),
687                    schema: info.schema_name.clone(),
688                    table_name: info.name.clone(),
689                    table_id: info.table_id(),
690                    table: table.clone(),
691                })
692                .unwrap();
693        }
694
695        let backend = Arc::new(MemoryKvBackend::default());
696        let metadata_manager = TableMetadataManager::new(backend.clone());
697        let routes = expressions
698            .into_iter()
699            .enumerate()
700            .map(|(index, expression)| region_route(index as u32 + 1, expression))
701            .collect();
702        metadata_manager
703            .create_table_metadata(
704                physical_info,
705                TableRouteValue::physical(routes),
706                HashMap::new(),
707            )
708            .await
709            .unwrap();
710        metadata_manager
711            .create_table_metadata(
712                logical_info,
713                TableRouteValue::logical(PHYSICAL_TABLE_ID),
714                HashMap::new(),
715            )
716            .await
717            .unwrap();
718
719        let table_route_cache = Arc::new(new_table_route_cache(
720            "planner-test-routes".to_string(),
721            CacheBuilder::new(16).build(),
722            backend.clone(),
723        ));
724        let partition_info_cache = Arc::new(new_partition_info_cache(
725            "planner-test-partitions".to_string(),
726            CacheBuilder::new(16).build(),
727            table_route_cache.clone(),
728        ));
729        let partition_rule_manager = Arc::new(PartitionRuleManager::new(
730            backend,
731            table_route_cache,
732            partition_info_cache,
733        ));
734        let (resolved_physical_id, physical_route) = partition_rule_manager
735            .find_physical_table_route_with_id(PHYSICAL_TABLE_ID)
736            .await
737            .unwrap();
738        assert_eq!(PHYSICAL_TABLE_ID, resolved_physical_id);
739        let (resolved_logical_id, logical_route) = partition_rule_manager
740            .find_physical_table_route_with_id(LOGICAL_TABLE_ID)
741            .await
742            .unwrap();
743        assert_eq!(PHYSICAL_TABLE_ID, resolved_logical_id);
744        assert_eq!(physical_route.region_routes, logical_route.region_routes);
745        let catalog_manager: CatalogManagerRef = catalog_manager;
746        let planner = DistExtensionPlanner::new(
747            catalog_manager,
748            partition_rule_manager,
749            Arc::new(UnusedRegionQueryHandler),
750            false,
751        );
752        let table_source = Arc::new(DefaultTableSource::new(Arc::new(
753            DfTableProviderAdapter::new(logical_table),
754        )));
755        let plan = LogicalPlanBuilder::scan_with_filters("logical", table_source, None, vec![])
756            .unwrap()
757            .filter(df_col("host").eq(lit("a")))
758            .unwrap()
759            .build()
760            .unwrap();
761        (
762            planner,
763            plan,
764            TableName::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, "logical"),
765        )
766    }
767
768    fn physical_partition_expressions() -> Vec<Option<PartitionExpr>> {
769        vec![
770            Some(partition_col("host").lt(Value::String("m".into()))),
771            Some(
772                partition_col("host")
773                    .gt_eq(Value::String("m".into()))
774                    .and(partition_col("rack").lt(Value::String("n".into()))),
775            ),
776            Some(
777                partition_col("host")
778                    .gt_eq(Value::String("m".into()))
779                    .and(partition_col("rack").gt_eq(Value::String("n".into()))),
780            ),
781        ]
782    }
783
784    #[tokio::test]
785    async fn region_routing_uses_captured_table() {
786        let (mut planner, plan, table_name) = planner_and_plan(vec![0], vec![None]).await;
787        planner.catalog_manager = MemoryCatalogManager::with_default_setup();
788        assert_eq!(
789            vec![RegionId::new(LOGICAL_TABLE_ID, 1)],
790            planner.get_regions(&table_name, &plan).await.unwrap()
791        );
792    }
793
794    #[tokio::test]
795    async fn logical_table_pruning_uses_physical_partition_datatypes() {
796        let (planner, plan, table_name) =
797            planner_and_plan(vec![0, 1], physical_partition_expressions()).await;
798
799        assert_eq!(
800            vec![RegionId::new(LOGICAL_TABLE_ID, 1)],
801            planner.get_regions(&table_name, &plan).await.unwrap()
802        );
803    }
804
805    #[tokio::test]
806    async fn missing_physical_partition_datatype_falls_back_to_all_logical_regions() {
807        let (planner, plan, table_name) =
808            planner_and_plan(vec![0], physical_partition_expressions()).await;
809
810        assert_all_logical_regions(planner.get_regions(&table_name, &plan).await.unwrap());
811    }
812
813    #[tokio::test]
814    async fn missing_route_partition_expression_falls_back_to_all_logical_regions() {
815        let mut expressions = physical_partition_expressions();
816        expressions[1] = None;
817        let (planner, plan, table_name) = planner_and_plan(vec![0, 1], expressions).await;
818
819        assert_all_logical_regions(planner.get_regions(&table_name, &plan).await.unwrap());
820    }
821
822    fn assert_all_logical_regions(mut regions: Vec<RegionId>) {
823        regions.sort_unstable();
824        assert_eq!(
825            vec![
826                RegionId::new(LOGICAL_TABLE_ID, 1),
827                RegionId::new(LOGICAL_TABLE_ID, 2),
828                RegionId::new(LOGICAL_TABLE_ID, 3),
829            ],
830            regions
831        );
832    }
833}
query/dist_plan/planner.rs

query/dist_plan/
planner.rs
