1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
// Copyright Materialize, Inc. and contributors. All rights reserved.
//
// Use of this software is governed by the Business Source License
// included in the LICENSE file.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0.

//! Optimizer implementation for `SELECT` statements.

use std::fmt::Debug;
use std::sync::Arc;
use std::time::{Duration, Instant};

use mz_compute_types::dataflows::IndexDesc;
use mz_compute_types::plan::Plan;
use mz_compute_types::ComputeInstanceId;
use mz_expr::{MirRelationExpr, MirScalarExpr, OptimizedMirRelationExpr, RowSetFinishing};
use mz_ore::soft_assert_or_log;
use mz_repr::explain::trace_plan;
use mz_repr::{GlobalId, RelationType, Timestamp};
use mz_sql::optimizer_metrics::OptimizerMetrics;
use mz_sql::plan::HirRelationExpr;
use mz_sql::session::metadata::SessionMetadata;
use mz_transform::dataflow::DataflowMetainfo;
use mz_transform::normalize_lets::normalize_lets;
use mz_transform::typecheck::{empty_context, SharedContext as TypecheckContext};
use mz_transform::{StatisticsOracle, TransformCtx};
use timely::progress::Antichain;
use tracing::debug_span;

use crate::catalog::Catalog;
use crate::coord::peek::{create_fast_path_plan, PeekDataflowPlan, PeekPlan};
use crate::optimize::dataflows::{
    prep_relation_expr, prep_scalar_expr, ComputeInstanceSnapshot, DataflowBuilder, EvalTime,
    ExprPrepStyle,
};
use crate::optimize::{
    optimize_mir_local, trace_plan, MirDataflowDescription, Optimize, OptimizeMode,
    OptimizerConfig, OptimizerError,
};
use crate::TimestampContext;

pub struct Optimizer {
    /// A typechecking context to use throughout the optimizer pipeline.
    typecheck_ctx: TypecheckContext,
    /// A snapshot of the catalog state.
    catalog: Arc<Catalog>,
    /// A snapshot of the cluster that will run the dataflows.
    compute_instance: ComputeInstanceSnapshot,
    /// Optional row-set finishing to be applied to the final result.
    finishing: RowSetFinishing,
    /// A transient GlobalId to be used when constructing the dataflow.
    select_id: GlobalId,
    /// A transient GlobalId to be used when constructing a PeekPlan.
    index_id: GlobalId,
    /// Optimizer config.
    config: OptimizerConfig,
    /// Optimizer metrics.
    metrics: OptimizerMetrics,
    /// The time spent performing optimization so far.
    duration: Duration,
}

impl Optimizer {
    pub fn new(
        catalog: Arc<Catalog>,
        compute_instance: ComputeInstanceSnapshot,
        finishing: RowSetFinishing,
        select_id: GlobalId,
        index_id: GlobalId,
        config: OptimizerConfig,
        metrics: OptimizerMetrics,
    ) -> Self {
        Self {
            typecheck_ctx: empty_context(),
            catalog,
            compute_instance,
            finishing,
            select_id,
            index_id,
            config,
            metrics,
            duration: Default::default(),
        }
    }

    pub fn cluster_id(&self) -> ComputeInstanceId {
        self.compute_instance.instance_id()
    }

    pub fn finishing(&self) -> &RowSetFinishing {
        &self.finishing
    }

    pub fn select_id(&self) -> GlobalId {
        self.select_id
    }

    pub fn index_id(&self) -> GlobalId {
        self.index_id
    }

    pub fn config(&self) -> &OptimizerConfig {
        &self.config
    }

    pub fn metrics(&self) -> &OptimizerMetrics {
        &self.metrics
    }

    pub fn duration(&self) -> Duration {
        self.duration
    }
}

// A bogey `Debug` implementation that hides fields. This is needed to make the
// `event!` call in `sequence_peek_stage` not emit a lot of data.
//
// For now, we skip almost all fields, but we might revisit that bit if it turns
// out that we really need those for debugging purposes.
impl Debug for Optimizer {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("OptimizePeek")
            .field("config", &self.config)
            .finish_non_exhaustive()
    }
}

/// Marker type for [`LocalMirPlan`] representing an optimization result without
/// context.
pub struct Unresolved;

/// The (sealed intermediate) result after HIR ⇒ MIR lowering and decorrelation
/// and local MIR optimization.
#[derive(Clone)]
pub struct LocalMirPlan<T = Unresolved> {
    expr: MirRelationExpr,
    df_meta: DataflowMetainfo,
    context: T,
}

/// Marker type for [`LocalMirPlan`] structs representing an optimization result
/// with attached environment context required for the next optimization stage.
pub struct Resolved<'s> {
    timestamp_ctx: TimestampContext<Timestamp>,
    stats: Box<dyn StatisticsOracle>,
    session: &'s dyn SessionMetadata,
}

/// The (final) result after
///
/// 1. embedding a [`LocalMirPlan`] into a `DataflowDescription`,
/// 2. transitively inlining referenced views,
/// 3. timestamp resolution,
/// 4. optimizing the resulting `DataflowDescription` with `MIR` plans.
/// 5. MIR ⇒ LIR lowering, and
/// 6. optimizing the resulting `DataflowDescription` with `LIR` plans.
#[derive(Debug)]
pub struct GlobalLirPlan {
    peek_plan: PeekPlan,
    df_meta: DataflowMetainfo,
    typ: RelationType,
}

impl Optimize<HirRelationExpr> for Optimizer {
    type To = LocalMirPlan;

    fn optimize(&mut self, expr: HirRelationExpr) -> Result<Self::To, OptimizerError> {
        let time = Instant::now();

        // Trace the pipeline input under `optimize/raw`.
        trace_plan!(at: "raw", &expr);

        // HIR ⇒ MIR lowering and decorrelation
        let expr = expr.lower(&self.config, Some(&self.metrics))?;

        // MIR ⇒ MIR optimization (local)
        let mut df_meta = DataflowMetainfo::default();
        let mut transform_ctx =
            TransformCtx::local(&self.config.features, &self.typecheck_ctx, &mut df_meta);
        let expr = optimize_mir_local(expr, &mut transform_ctx)?.into_inner();

        self.duration += time.elapsed();

        // Return the (sealed) plan at the end of this optimization step.
        Ok(LocalMirPlan {
            expr,
            df_meta,
            context: Unresolved,
        })
    }
}

impl LocalMirPlan<Unresolved> {
    /// Produces the [`LocalMirPlan`] with [`Resolved`] contextual information
    /// required for the next stage.
    pub fn resolve(
        self,
        timestamp_ctx: TimestampContext<Timestamp>,
        session: &dyn SessionMetadata,
        stats: Box<dyn StatisticsOracle>,
    ) -> LocalMirPlan<Resolved> {
        LocalMirPlan {
            expr: self.expr,
            df_meta: self.df_meta,
            context: Resolved {
                timestamp_ctx,
                session,
                stats,
            },
        }
    }
}

impl<'s> Optimize<LocalMirPlan<Resolved<'s>>> for Optimizer {
    type To = GlobalLirPlan;

    fn optimize(&mut self, plan: LocalMirPlan<Resolved<'s>>) -> Result<Self::To, OptimizerError> {
        let time = Instant::now();

        let LocalMirPlan {
            expr,
            mut df_meta,
            context:
                Resolved {
                    timestamp_ctx,
                    stats,
                    session,
                },
        } = plan;

        let expr = OptimizedMirRelationExpr(expr);

        // We create a dataflow and optimize it, to determine if we can avoid building it.
        // This can happen if the result optimizes to a constant, or to a `Get` expression
        // around a maintained arrangement.
        let typ = expr.typ();
        let key = typ
            .default_key()
            .iter()
            .map(|k| MirScalarExpr::Column(*k))
            .collect();

        // The assembled dataflow contains a view and an index of that view.
        let mut df_builder = {
            let catalog = self.catalog.state();
            let compute = self.compute_instance.clone();
            DataflowBuilder::new(catalog, compute).with_config(&self.config)
        };

        let debug_name = format!("oneshot-select-{}", self.select_id);
        let mut df_desc = MirDataflowDescription::new(debug_name.to_string());

        df_builder.import_view_into_dataflow(
            &self.select_id,
            &expr,
            &mut df_desc,
            &self.config.features,
        )?;
        df_builder.maybe_reoptimize_imported_views(&mut df_desc, &self.config)?;

        // Resolve all unmaterializable function calls except mz_now(), because
        // we don't yet have a timestamp.
        let style = ExprPrepStyle::OneShot {
            logical_time: EvalTime::Deferred,
            session,
            catalog_state: self.catalog.state(),
        };
        df_desc.visit_children(
            |r| prep_relation_expr(r, style),
            |s| prep_scalar_expr(s, style),
        )?;

        // TODO: Instead of conditioning here we should really
        // reconsider how to render multi-plan peek dataflows. The main
        // difficulty here is rendering the optional finishing bit.
        if self.config.mode != OptimizeMode::Explain {
            df_desc.export_index(
                self.index_id,
                IndexDesc {
                    on_id: self.select_id,
                    key,
                },
                typ.clone(),
            );
        }

        // Set the `as_of` and `until` timestamps for the dataflow.
        df_desc.set_as_of(timestamp_ctx.antichain());

        // Use the opportunity to name an `until` frontier that will prevent
        // work we needn't perform. By default, `until` will be
        // `Antichain::new()`, which prevents no updates and is safe.
        //
        // If `timestamp_ctx.antichain()` is empty, `timestamp_ctx.timestamp()`
        // will return `None` and we use the default (empty) `until`. Otherwise,
        // we expect to be able to set `until = as_of + 1` without an overflow, unless
        // we query at the maximum timestamp. In this case, the default empty `until`
        // is the correct choice.
        if let Some(until) = timestamp_ctx
            .timestamp()
            .and_then(Timestamp::try_step_forward)
        {
            df_desc.until = Antichain::from_elem(until);
        }

        // Construct TransformCtx for global optimization.
        let mut transform_ctx = TransformCtx::global(
            &df_builder,
            &*stats,
            &self.config.features,
            &self.typecheck_ctx,
            &mut df_meta,
        );

        // Let's already try creating a fast path plan. If successful, we don't need to run the
        // whole optimizer pipeline, but just a tiny subset of it. (But we'll need to run
        // `create_fast_path_plan` later again, because, e.g., running `LiteralConstraints` is still
        // ahead of us.)
        let use_fast_path_optimizer = match create_fast_path_plan(
            &mut df_desc,
            self.select_id,
            Some(&self.finishing),
            self.config.features.persist_fast_path_limit,
        ) {
            Ok(maybe_fast_path_plan) => maybe_fast_path_plan.is_some(),
            Err(OptimizerError::UnsafeMfpPlan) => {
                // This is expected, in that `create_fast_path_plan` can choke on `mz_now`, which we
                // haven't removed yet.
                false
            }
            Err(e) => {
                return Err(e);
            }
        };

        // Run global optimization.
        mz_transform::optimize_dataflow(&mut df_desc, &mut transform_ctx, use_fast_path_optimizer)?;

        if self.config.mode == OptimizeMode::Explain {
            // Collect the list of indexes used by the dataflow at this point.
            trace_plan!(at: "global", &df_meta.used_indexes(&df_desc));
        }

        // Get the single timestamp representing the `as_of` time.
        let as_of = df_desc
            .as_of
            .clone()
            .expect("as_of antichain")
            .into_option()
            .expect("unique as_of element");

        // Resolve all unmaterializable function calls including mz_now().
        let style = ExprPrepStyle::OneShot {
            logical_time: EvalTime::Time(as_of),
            session,
            catalog_state: self.catalog.state(),
        };
        df_desc.visit_children(
            |r| prep_relation_expr(r, style),
            |s| prep_scalar_expr(s, style),
        )?;

        // TODO: use the following code once we can be sure that the
        // index_exports always exist.
        //
        // let typ = self.df_desc
        //     .index_exports
        //     .first_key_value()
        //     .map(|(_key, (_desc, typ))| typ.clone())
        //     .expect("GlobalMirPlan type");

        let peek_plan = match create_fast_path_plan(
            &mut df_desc,
            self.select_id,
            Some(&self.finishing),
            self.config.features.persist_fast_path_limit,
        )? {
            Some(plan) if !self.config.no_fast_path => {
                if self.config.mode == OptimizeMode::Explain {
                    // Trace the `used_indexes` for the FastPathPlan.
                    debug_span!(target: "optimizer", "fast_path").in_scope(|| {
                        // Fast path plans come with an updated finishing.
                        let finishing = if !self.finishing.is_trivial(typ.arity()) {
                            Some(&self.finishing)
                        } else {
                            None
                        };
                        trace_plan(&plan.used_indexes(finishing));
                    });
                }
                // Trace the FastPathPlan.
                trace_plan!(at: "fast_path", &plan);

                // Trace the pipeline output under `optimize`.
                trace_plan(&plan);

                // Build the PeekPlan
                PeekPlan::FastPath(plan)
            }
            _ => {
                soft_assert_or_log!(
                    !use_fast_path_optimizer || self.config.no_fast_path,
                    "The fast_path_optimizer shouldn't make a fast path plan slow path."
                );

                // Ensure all expressions are normalized before finalizing.
                for build in df_desc.objects_to_build.iter_mut() {
                    normalize_lets(&mut build.plan.0, &self.config.features)?
                }

                // Finalize the dataflow. This includes:
                // - MIR ⇒ LIR lowering
                // - LIR ⇒ LIR transforms
                let df_desc = Plan::finalize_dataflow(df_desc, &self.config.features)?;

                // Trace the pipeline output under `optimize`.
                trace_plan(&df_desc);

                // Build the PeekPlan
                PeekPlan::SlowPath(PeekDataflowPlan::new(df_desc, self.index_id(), &typ))
            }
        };

        self.duration += time.elapsed();
        let label = match &peek_plan {
            PeekPlan::FastPath(_) => "peek:fast_path",
            PeekPlan::SlowPath(_) => "peek:slow_path",
        };
        self.metrics
            .observe_e2e_optimization_time(label, self.duration);

        Ok(GlobalLirPlan {
            peek_plan,
            df_meta,
            typ,
        })
    }
}

impl GlobalLirPlan {
    /// Unwraps the parts of the final result of the optimization pipeline.
    pub fn unapply(self) -> (PeekPlan, DataflowMetainfo, RelationType) {
        (self.peek_plan, self.df_meta, self.typ)
    }
}