diesel/connection/statement_cache/
mod.rs

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
//! Helper types for prepared statement caching
//!
//! A primer on prepared statement caching in Diesel
//! ------------------------------------------------
//!
//! Diesel uses prepared statements for virtually all queries. This is most
//! visible in our lack of any sort of "quoting" API. Values must always be
//! transmitted as bind parameters, we do not support direct interpolation. The
//! only method in the public API that doesn't require the use of prepared
//! statements is [`SimpleConnection::batch_execute`](super::SimpleConnection::batch_execute).
//!
//! In order to avoid the cost of re-parsing and planning subsequent queries,
//! by default Diesel caches the prepared statement whenever possible, but
//! this an be customized by calling [`Connection::set_cache_size`](super::Connection::set_cache_size).
//! Queries will fall into one of three buckets:
//!
//! - Unsafe to cache
//! - Cached by SQL
//! - Cached by type
//!
//! A query is considered unsafe to cache if it represents a potentially
//! unbounded number of queries. This is communicated to the connection through
//! [`QueryFragment::is_safe_to_cache_prepared`]. While this is done as a full AST
//! pass, after monomorphisation and inlining this will usually be optimized to
//! a constant. Only boxed queries will need to do actual work to answer this
//! question.
//!
//! The majority of AST nodes are safe to cache if their components are safe to
//! cache. There are at least 4 cases where a query is unsafe to cache:
//!
//! - queries containing `IN` with bind parameters
//!     - This requires 1 bind parameter per value, and is therefore unbounded
//!     - `IN` with subselects are cached (assuming the subselect is safe to
//!        cache)
//!     - `IN` statements for postgresql are cached as they use `= ANY($1)` instead
//!        which does not cause a unbound number of binds
//! - `INSERT` statements with a variable number of rows
//!     - The SQL varies based on the number of rows being inserted.
//! - `UPDATE` statements
//!     - Technically it's bounded on "number of optional values being passed to
//!       `SET` factorial" but that's still quite high, and not worth caching
//!       for the same reason as single row inserts
//! - `SqlLiteral` nodes
//!     - We have no way of knowing whether the SQL was generated dynamically or
//!       not, so we must assume that it's unbounded
//!
//! For queries which are unsafe to cache, the statement cache will never insert
//! them. They will be prepared and immediately released after use (or in the
//! case of PG they will use the unnamed prepared statement).
//!
//! For statements which are able to be cached, we then have to determine what
//! to use as the cache key. The standard method that virtually all ORMs or
//! database access layers use in the wild is to store the statements in a
//! hash map, using the SQL as the key.
//!
//! However, the majority of queries using Diesel that are safe to cache as
//! prepared statements will be uniquely identified by their type. For these
//! queries, we can bypass the query builder entirely. Since our AST is
//! generally optimized away by the compiler, for these queries the cost of
//! fetching a prepared statement from the cache is the cost of [`HashMap<u32,
//! _>::get`], where the key we're fetching by is a compile time constant. For
//! these types, the AST pass to gather the bind parameters will also be
//! optimized to accessing each parameter individually.
//!
//! Determining if a query can be cached by type is the responsibility of the
//! [`QueryId`] trait. This trait is quite similar to `Any`, but with a few
//! differences:
//!
//! - No `'static` bound
//!     - Something being a reference never changes the SQL that is generated,
//!       so `&T` has the same query id as `T`.
//! - `Option<TypeId>` instead of `TypeId`
//!     - We need to be able to constrain on this trait being implemented, but
//!       not all types will actually have a static query id. Hopefully once
//!       specialization is stable we can remove the `QueryId` bound and
//!       specialize on it instead (or provide a blanket impl for all `T`)
//! - Implementors give a more broad type than `Self`
//!     - This really only affects bind parameters. There are 6 different Rust
//!       types which can be used for a parameter of type `timestamp`. The same
//!       statement can be used regardless of the Rust type, so [`Bound<ST, T>`](crate::expression::bound::Bound)
//!       defines its [`QueryId`] as [`Bound<ST, ()>`](crate::expression::bound::Bound).
//!
//! A type returning `Some(id)` or `None` for its query ID is based on whether
//! the SQL it generates can change without the type changing. At the moment,
//! the only type which is safe to cache as a prepared statement but does not
//! have a static query ID is something which has been boxed.
//!
//! One potential optimization that we don't perform is storing the queries
//! which are cached by type ID in a separate map. Since a type ID is a u64,
//! this would allow us to use a specialized map which knows that there will
//! never be hashing collisions (also known as a perfect hashing function),
//! which would mean lookups are always constant time. However, this would save
//! nanoseconds on an operation that will take microseconds or even
//! milliseconds.

use std::any::TypeId;
use std::borrow::Cow;
use std::hash::Hash;
use std::ops::{Deref, DerefMut};

use strategy::{StatementCacheStrategy, WithCacheStrategy, WithoutCacheStrategy};

use crate::backend::Backend;
use crate::connection::InstrumentationEvent;
use crate::query_builder::*;
use crate::result::QueryResult;

use super::{CacheSize, Instrumentation};

/// Various interfaces and implementations to control connection statement caching.
#[allow(unreachable_pub)]
pub mod strategy;

/// A prepared statement cache
#[allow(missing_debug_implementations, unreachable_pub)]
#[cfg_attr(
    docsrs,
    doc(cfg(feature = "i-implement-a-third-party-backend-and-opt-into-breaking-changes"))
)]
pub struct StatementCache<DB: Backend, Statement> {
    cache: Box<dyn StatementCacheStrategy<DB, Statement>>,
    // increment every time a query is cached
    // some backends might use it to create unique prepared statement names
    cache_counter: u64,
}

/// A helper type that indicates if a certain query
/// is cached inside of the prepared statement cache or not
///
/// This information can be used by the connection implementation
/// to signal this fact to the database while actually
/// preparing the statement
#[derive(Debug, Clone, Copy)]
#[cfg_attr(
    docsrs,
    doc(cfg(feature = "i-implement-a-third-party-backend-and-opt-into-breaking-changes"))
)]
#[allow(unreachable_pub)]
pub enum PrepareForCache {
    /// The statement will be cached    
    Yes {
        /// Counter might be used as unique identifier for prepared statement.
        #[allow(dead_code)]
        counter: u64,
    },
    /// The statement won't be cached
    No,
}

#[allow(clippy::new_without_default, unreachable_pub)]
impl<DB, Statement> StatementCache<DB, Statement>
where
    DB: Backend + 'static,
    Statement: 'static,
    DB::TypeMetadata: Clone,
    DB::QueryBuilder: Default,
    StatementCacheKey<DB>: Hash + Eq,
{
    /// Create a new prepared statement cache using [`CacheSize::Unbounded`] as caching strategy.
    #[allow(unreachable_pub)]
    pub fn new() -> Self {
        StatementCache {
            cache: Box::new(WithCacheStrategy::default()),
            cache_counter: 0,
        }
    }

    /// Set caching strategy from predefined implementations
    pub fn set_cache_size(&mut self, size: CacheSize) {
        if self.cache.cache_size() != size {
            self.cache = match size {
                CacheSize::Unbounded => Box::new(WithCacheStrategy::default()),
                CacheSize::Disabled => Box::new(WithoutCacheStrategy::default()),
            }
        }
    }

    /// Setting custom caching strategy. It is used in tests, to verify caching logic
    #[allow(dead_code)]
    pub(crate) fn set_strategy<Strategy>(&mut self, s: Strategy)
    where
        Strategy: StatementCacheStrategy<DB, Statement> + 'static,
    {
        self.cache = Box::new(s);
    }

    /// Prepare a query as prepared statement
    ///
    /// This functions returns a prepared statement corresponding to the
    /// query passed as `source` with the bind values passed as `bind_types`.
    /// If the query is already cached inside this prepared statement cache
    /// the cached prepared statement will be returned, otherwise `prepare_fn`
    /// will be called to create a new prepared statement for this query source.
    /// The first parameter of the callback contains the query string, the second
    /// parameter indicates if the constructed prepared statement will be cached or not.
    /// See the [module](self) documentation for details
    /// about which statements are cached and which are not cached.
    #[allow(unreachable_pub)]
    pub fn cached_statement<T, F>(
        &mut self,
        source: &T,
        backend: &DB,
        bind_types: &[DB::TypeMetadata],
        mut prepare_fn: F,
        instrumentation: &mut dyn Instrumentation,
    ) -> QueryResult<MaybeCached<'_, Statement>>
    where
        T: QueryFragment<DB> + QueryId,
        F: FnMut(&str, PrepareForCache) -> QueryResult<Statement>,
    {
        Self::cached_statement_non_generic(
            self.cache.as_mut(),
            T::query_id(),
            source,
            backend,
            bind_types,
            &mut |sql, is_cached| {
                if is_cached {
                    instrumentation.on_connection_event(InstrumentationEvent::CacheQuery { sql });
                    self.cache_counter += 1;
                    prepare_fn(
                        sql,
                        PrepareForCache::Yes {
                            counter: self.cache_counter,
                        },
                    )
                } else {
                    prepare_fn(sql, PrepareForCache::No)
                }
            },
        )
    }

    /// Reduce the amount of monomorphized code by factoring this via dynamic dispatch
    fn cached_statement_non_generic<'a>(
        cache: &'a mut dyn StatementCacheStrategy<DB, Statement>,
        maybe_type_id: Option<TypeId>,
        source: &dyn QueryFragmentForCachedStatement<DB>,
        backend: &DB,
        bind_types: &[DB::TypeMetadata],
        prepare_fn: &mut dyn FnMut(&str, bool) -> QueryResult<Statement>,
    ) -> QueryResult<MaybeCached<'a, Statement>> {
        let cache_key = StatementCacheKey::for_source(maybe_type_id, source, bind_types, backend)?;
        if !source.is_safe_to_cache_prepared(backend)? {
            let sql = cache_key.sql(source, backend)?;
            return prepare_fn(&sql, false).map(MaybeCached::CannotCache);
        }
        cache.get(cache_key, backend, source, prepare_fn)
    }
}

/// Implemented for all `QueryFragment`s, dedicated to dynamic dispatch within the context of
/// `statement_cache`
///
/// We want the generated code to be as small as possible, so for each query passed to
/// [`StatementCache::cached_statement`] the generated assembly will just call a non generic
/// version with dynamic dispatch pointing to the VTABLE of this minimal trait
///
/// This preserves the opportunity for the compiler to entirely optimize the `construct_sql`
/// function as a function that simply returns a constant `String`.
#[allow(unreachable_pub)]
#[cfg_attr(
    docsrs,
    doc(cfg(feature = "i-implement-a-third-party-backend-and-opt-into-breaking-changes"))
)]
pub trait QueryFragmentForCachedStatement<DB> {
    /// Convert the query fragment into a SQL string for the given backend
    fn construct_sql(&self, backend: &DB) -> QueryResult<String>;
    /// Check whether it's safe to cache the query
    fn is_safe_to_cache_prepared(&self, backend: &DB) -> QueryResult<bool>;
}
impl<T, DB> QueryFragmentForCachedStatement<DB> for T
where
    DB: Backend,
    DB::QueryBuilder: Default,
    T: QueryFragment<DB>,
{
    fn construct_sql(&self, backend: &DB) -> QueryResult<String> {
        let mut query_builder = DB::QueryBuilder::default();
        self.to_sql(&mut query_builder, backend)?;
        Ok(query_builder.finish())
    }

    fn is_safe_to_cache_prepared(&self, backend: &DB) -> QueryResult<bool> {
        <T as QueryFragment<DB>>::is_safe_to_cache_prepared(self, backend)
    }
}

/// Wraps a possibly cached prepared statement
///
/// Essentially a customized version of [`Cow`]
/// that does not depend on [`ToOwned`]
#[allow(missing_debug_implementations, unreachable_pub)]
#[cfg_attr(
    docsrs,
    doc(cfg(feature = "i-implement-a-third-party-backend-and-opt-into-breaking-changes"))
)]
#[non_exhaustive]
pub enum MaybeCached<'a, T: 'a> {
    /// Contains a not cached prepared statement
    CannotCache(T),
    /// Contains a reference cached prepared statement
    Cached(&'a mut T),
}

impl<T> Deref for MaybeCached<'_, T> {
    type Target = T;

    fn deref(&self) -> &Self::Target {
        match *self {
            MaybeCached::CannotCache(ref x) => x,
            MaybeCached::Cached(ref x) => x,
        }
    }
}

impl<T> DerefMut for MaybeCached<'_, T> {
    fn deref_mut(&mut self) -> &mut Self::Target {
        match *self {
            MaybeCached::CannotCache(ref mut x) => x,
            MaybeCached::Cached(ref mut x) => x,
        }
    }
}

/// The lookup key used by [`StatementCache`] internally
///
/// This can contain either a at compile time known type id
/// (representing a statically known query) or a at runtime
/// calculated query string + parameter types (for queries
/// that may change depending on their parameters)
#[allow(missing_debug_implementations, unreachable_pub)]
#[derive(Hash, PartialEq, Eq)]
#[cfg_attr(
    docsrs,
    doc(cfg(feature = "i-implement-a-third-party-backend-and-opt-into-breaking-changes"))
)]
pub enum StatementCacheKey<DB: Backend> {
    /// Represents a at compile time known query
    ///
    /// Calculated via [`QueryId::QueryId`]
    Type(TypeId),
    /// Represents a dynamically constructed query
    ///
    /// This variant is used if [`QueryId::HAS_STATIC_QUERY_ID`]
    /// is `false` and [`AstPass::unsafe_to_cache_prepared`] is not
    /// called for a given query.
    Sql {
        /// contains the sql query string
        sql: String,
        /// contains the types of any bind parameter passed to the query
        bind_types: Vec<DB::TypeMetadata>,
    },
}

impl<DB> StatementCacheKey<DB>
where
    DB: Backend,
    DB::QueryBuilder: Default,
    DB::TypeMetadata: Clone,
{
    /// Create a new statement cache key for the given query source
    // Note: Intentionally monomorphic over source.
    #[allow(unreachable_pub)]
    pub fn for_source(
        maybe_type_id: Option<TypeId>,
        source: &dyn QueryFragmentForCachedStatement<DB>,
        bind_types: &[DB::TypeMetadata],
        backend: &DB,
    ) -> QueryResult<Self> {
        match maybe_type_id {
            Some(id) => Ok(StatementCacheKey::Type(id)),
            None => {
                let sql = source.construct_sql(backend)?;
                Ok(StatementCacheKey::Sql {
                    sql,
                    bind_types: bind_types.into(),
                })
            }
        }
    }

    /// Get the sql for a given query source based
    ///
    /// This is an optimization that may skip constructing the query string
    /// twice if it's already part of the current cache key
    // Note: Intentionally monomorphic over source.
    #[allow(unreachable_pub)]
    pub fn sql(
        &self,
        source: &dyn QueryFragmentForCachedStatement<DB>,
        backend: &DB,
    ) -> QueryResult<Cow<'_, str>> {
        match *self {
            StatementCacheKey::Type(_) => source.construct_sql(backend).map(Cow::Owned),
            StatementCacheKey::Sql { ref sql, .. } => Ok(Cow::Borrowed(sql)),
        }
    }
}