mz_adapter/optimize/
index.rs

1// Copyright Materialize, Inc. and contributors. All rights reserved.
2//
3// Use of this software is governed by the Business Source License
4// included in the LICENSE file.
5//
6// As of the Change Date specified in that file, in accordance with
7// the Business Source License, use of this software will be governed
8// by the Apache License, Version 2.0.
9
10//! Optimizer implementation for `CREATE INDEX` statements.
11//!
12//! Note that, in contrast to other optimization pipelines, timestamp selection is not part of
13//! index optimization. Instead users are expected to separately set the as-of on the optimized
14//! `DataflowDescription` received from `GlobalLirPlan::unapply`. Reasons for choosing to exclude
15//! timestamp selection from the index optimization pipeline are:
16//!
17//!  (a) Indexes don't support non-empty `until` frontiers, so they don't provide opportunity for
18//!      optimizations based on the selected timestamp.
19//!  (b) We want to generate dataflow plans early during environment bootstrapping, before we have
20//!      access to all information required for timestamp selection.
21//!
22//! None of this is set in stone though. If we find an opportunity for optimizing indexes based on
23//! their timestamps, we'll want to make timestamp selection part of the index optimization again
24//! and find a different approach to bootstrapping.
25//!
26//! See also MaterializeInc/materialize#22940.
27
28use std::sync::Arc;
29use std::time::{Duration, Instant};
30
31use mz_compute_types::dataflows::IndexDesc;
32use mz_compute_types::plan::Plan;
33use mz_repr::GlobalId;
34use mz_repr::explain::trace_plan;
35use mz_sql::names::QualifiedItemName;
36use mz_sql::optimizer_metrics::OptimizerMetrics;
37use mz_transform::TransformCtx;
38use mz_transform::dataflow::DataflowMetainfo;
39use mz_transform::normalize_lets::normalize_lets;
40use mz_transform::notice::{IndexAlreadyExists, IndexKeyEmpty};
41use mz_transform::reprtypecheck::{
42    SharedContext as ReprTypecheckContext, empty_context as empty_repr_context,
43};
44
45use crate::optimize::dataflows::{
46    ComputeInstanceSnapshot, DataflowBuilder, ExprPrepStyle, prep_relation_expr, prep_scalar_expr,
47};
48use crate::optimize::{
49    LirDataflowDescription, MirDataflowDescription, Optimize, OptimizeMode, OptimizerCatalog,
50    OptimizerConfig, OptimizerError, trace_plan,
51};
52
53pub struct Optimizer {
54    /// A representation typechecking context to use throughout the optimizer pipeline.
55    repr_typecheck_ctx: ReprTypecheckContext,
56    /// A snapshot of the catalog state.
57    catalog: Arc<dyn OptimizerCatalog>,
58    /// A snapshot of the cluster that will run the dataflows.
59    compute_instance: ComputeInstanceSnapshot,
60    /// A durable GlobalId to be used with the exported index arrangement.
61    exported_index_id: GlobalId,
62    /// Optimizer config.
63    config: OptimizerConfig,
64    /// Optimizer metrics.
65    metrics: OptimizerMetrics,
66    /// The time spent performing optimization so far.
67    duration: Duration,
68}
69
70impl Optimizer {
71    pub fn new(
72        catalog: Arc<dyn OptimizerCatalog>,
73        compute_instance: ComputeInstanceSnapshot,
74        exported_index_id: GlobalId,
75        config: OptimizerConfig,
76        metrics: OptimizerMetrics,
77    ) -> Self {
78        Self {
79            repr_typecheck_ctx: empty_repr_context(),
80            catalog,
81            compute_instance,
82            exported_index_id,
83            config,
84            metrics,
85            duration: Default::default(),
86        }
87    }
88}
89
90/// A wrapper of index parts needed to start the optimization process.
91pub struct Index {
92    name: QualifiedItemName,
93    on: GlobalId,
94    keys: Vec<mz_expr::MirScalarExpr>,
95}
96
97impl Index {
98    /// Construct a new [`Index`]. Arguments are recorded as-is.
99    pub fn new(name: QualifiedItemName, on: GlobalId, keys: Vec<mz_expr::MirScalarExpr>) -> Self {
100        Self { name, on, keys }
101    }
102}
103
104/// The (sealed intermediate) result after:
105///
106/// 1. embedding an [`Index`] into a [`MirDataflowDescription`],
107/// 2. transitively inlining referenced views, and
108/// 3. jointly optimizing the `MIR` plans in the [`MirDataflowDescription`].
109#[derive(Clone, Debug)]
110pub struct GlobalMirPlan {
111    df_desc: MirDataflowDescription,
112    df_meta: DataflowMetainfo,
113}
114
115impl GlobalMirPlan {
116    pub fn df_desc(&self) -> &MirDataflowDescription {
117        &self.df_desc
118    }
119}
120
121/// The (final) result after MIR ⇒ LIR lowering and optimizing the resulting
122/// `DataflowDescription` with `LIR` plans.
123#[derive(Clone, Debug)]
124pub struct GlobalLirPlan {
125    df_desc: LirDataflowDescription,
126    df_meta: DataflowMetainfo,
127}
128
129impl GlobalLirPlan {
130    pub fn df_desc(&self) -> &LirDataflowDescription {
131        &self.df_desc
132    }
133
134    pub fn df_meta(&self) -> &DataflowMetainfo {
135        &self.df_meta
136    }
137}
138
139impl Optimize<Index> for Optimizer {
140    type To = GlobalMirPlan;
141
142    fn optimize(&mut self, index: Index) -> Result<Self::To, OptimizerError> {
143        let time = Instant::now();
144
145        let on_entry = self.catalog.get_entry(&index.on);
146        let full_name = self
147            .catalog
148            .resolve_full_name(&index.name, on_entry.conn_id());
149        let on_desc = on_entry
150            .relation_desc()
151            .expect("can only create indexes on items with a valid description");
152
153        let mut df_builder = {
154            let compute = self.compute_instance.clone();
155            DataflowBuilder::new(&*self.catalog, compute).with_config(&self.config)
156        };
157        let mut df_desc = MirDataflowDescription::new(full_name.to_string());
158
159        df_builder.import_into_dataflow(&index.on, &mut df_desc, &self.config.features)?;
160        df_builder.maybe_reoptimize_imported_views(&mut df_desc, &self.config)?;
161
162        let index_desc = IndexDesc {
163            on_id: index.on,
164            key: index.keys.clone(),
165        };
166        df_desc.export_index(self.exported_index_id, index_desc, on_desc.typ().clone());
167
168        // Prepare expressions in the assembled dataflow.
169        let style = ExprPrepStyle::Maintained;
170        df_desc.visit_children(
171            |r| prep_relation_expr(r, style),
172            |s| prep_scalar_expr(s, style),
173        )?;
174
175        // Construct TransformCtx for global optimization.
176        let mut df_meta = DataflowMetainfo::default();
177        let mut transform_ctx = TransformCtx::global(
178            &df_builder,
179            &mz_transform::EmptyStatisticsOracle, // TODO: wire proper stats
180            &self.config.features,
181            &self.repr_typecheck_ctx,
182            &mut df_meta,
183            Some(&mut self.metrics),
184        );
185        // Run global optimization.
186        mz_transform::optimize_dataflow(&mut df_desc, &mut transform_ctx, false)?;
187
188        if self.config.mode == OptimizeMode::Explain {
189            // Collect the list of indexes used by the dataflow at this point.
190            trace_plan!(at: "global", &df_meta.used_indexes(&df_desc));
191        }
192
193        // Emit a notice if we are trying to create an empty index.
194        if index.keys.is_empty() {
195            df_meta.push_optimizer_notice_dedup(IndexKeyEmpty);
196        }
197
198        // Emit a notice for each available index identical to the one we are
199        // currently optimizing.
200        for (index_id, idx) in df_builder
201            .indexes_on(index.on)
202            .filter(|(_id, idx)| idx.keys.as_ref() == &index.keys)
203        {
204            df_meta.push_optimizer_notice_dedup(IndexAlreadyExists {
205                index_id,
206                index_key: idx.keys.to_vec(),
207                index_on_id: idx.on,
208                exported_index_id: self.exported_index_id,
209            });
210        }
211
212        self.duration += time.elapsed();
213
214        // Return the (sealed) plan at the end of this optimization step.
215        Ok(GlobalMirPlan { df_desc, df_meta })
216    }
217}
218
219impl Optimize<GlobalMirPlan> for Optimizer {
220    type To = GlobalLirPlan;
221
222    fn optimize(&mut self, plan: GlobalMirPlan) -> Result<Self::To, OptimizerError> {
223        let time = Instant::now();
224
225        let GlobalMirPlan {
226            mut df_desc,
227            df_meta,
228        } = plan;
229
230        // Ensure all expressions are normalized before finalizing.
231        for build in df_desc.objects_to_build.iter_mut() {
232            normalize_lets(&mut build.plan.0, &self.config.features)?
233        }
234
235        // Finalize the dataflow. This includes:
236        // - MIR ⇒ LIR lowering
237        // - LIR ⇒ LIR transforms
238        let df_desc = Plan::finalize_dataflow(df_desc, &self.config.features)?;
239
240        // Trace the pipeline output under `optimize`.
241        trace_plan(&df_desc);
242
243        self.duration += time.elapsed();
244        self.metrics
245            .observe_e2e_optimization_time("index", self.duration);
246
247        // Return the plan at the end of this `optimize` step.
248        Ok(GlobalLirPlan { df_desc, df_meta })
249    }
250}
251
252impl GlobalLirPlan {
253    /// Unwraps the parts of the final result of the optimization pipeline.
254    pub fn unapply(self) -> (LirDataflowDescription, DataflowMetainfo) {
255        (self.df_desc, self.df_meta)
256    }
257}
mz_adapter/optimize/index.rs

mz_adapter/optimize/
index.rs
