[SQLite] Add option to execute PRAGMA optimize; on close of a connection (#2116)

* CHANGELOG: mention that users should upgrade CLI

* [SQLite] Add option to execute `PRAGMA optimize;` on close of a connection

* Update sqlx-sqlite/src/options/mod.rs

* Update sqlx-sqlite/src/options/mod.rs

* Update sqlx-sqlite/src/options/mod.rs

---------

Co-authored-by: Austin Bonander <austin.bonander@gmail.com>

sqlx-sqlite/src/connection/mod.rs

@@ -11,8 +11,11 @@ use std::ptr::NonNull;
 
 use crate::connection::establish::EstablishParams;
 use crate::connection::worker::ConnectionWorker;
+use crate::options::OptimizeOnClose;
 use crate::statement::VirtualStatement;
 use crate::{Sqlite, SqliteConnectOptions};
+use sqlx_core::executor::Executor;
+use std::fmt::Write;
 
 pub(crate) use sqlx_core::connection::*;
 
@@ -39,6 +42,7 @@ mod worker;
 /// You can explicitly call [`.close()`][Self::close] to ensure the database is closed successfully
 /// or get an error otherwise.
 pub struct SqliteConnection {
+    optimize_on_close: OptimizeOnClose,
     pub(crate) worker: ConnectionWorker,
     pub(crate) row_channel_size: usize,
 }
@@ -70,6 +74,7 @@ impl SqliteConnection {
         let params = EstablishParams::from_options(options)?;
         let worker = ConnectionWorker::establish(params).await?;
         Ok(Self {
+            optimize_on_close: options.optimize_on_close.clone(),
             worker,
             row_channel_size: options.row_channel_size,
         })
@@ -102,6 +107,14 @@ impl Connection for SqliteConnection {
 
     fn close(mut self) -> BoxFuture<'static, Result<(), Error>> {
         Box::pin(async move {
+            if let OptimizeOnClose::Enabled { analysis_limit } = self.optimize_on_close {
+                let mut pragma_string = String::new();
+                if let Some(limit) = analysis_limit {
+                    write!(pragma_string, "PRAGMA analysis_limit = {}; ", limit).ok();
+                }
+                pragma_string.push_str("PRAGMA optimize;");
+                self.execute(&*pragma_string).await?;
+            }
             let shutdown = self.worker.shutdown();
             // Drop the statement worker, which should
             // cover all references to the connection handle outside of the worker thread

sqlx-sqlite/src/options/mod.rs

@@ -80,10 +80,18 @@ pub struct SqliteConnectOptions {
     pub(crate) serialized: bool,
     pub(crate) thread_name: Arc<DebugFn<dyn Fn(u64) -> String + Send + Sync + 'static>>,
 
+    pub(crate) optimize_on_close: OptimizeOnClose,
+
     #[cfg(feature = "regexp")]
     pub(crate) register_regexp_function: bool,
 }
 
+#[derive(Clone, Debug)]
+pub enum OptimizeOnClose {
+    Enabled { analysis_limit: Option<u32> },
+    Disabled,
+}
+
 impl Default for SqliteConnectOptions {
     fn default() -> Self {
         Self::new()
@@ -170,6 +178,9 @@ impl SqliteConnectOptions {
 
         pragmas.insert("auto_vacuum".into(), None);
 
+        // Soft limit on the number of rows that `ANALYZE` touches per index.
+        pragmas.insert("analysis_limit".into(), None);
+
         Self {
             filename: Cow::Borrowed(Path::new(":memory:")),
             in_memory: false,
@@ -188,6 +199,7 @@ impl SqliteConnectOptions {
             thread_name: Arc::new(DebugFn(|id| format!("sqlx-sqlite-worker-{}", id))),
             command_channel_size: 50,
             row_channel_size: 50,
+            optimize_on_close: OptimizeOnClose::Disabled,
             #[cfg(feature = "regexp")]
             register_regexp_function: false,
         }
@@ -464,6 +476,54 @@ impl SqliteConnectOptions {
         self
     }
 
+    /// Execute `PRAGMA optimize;` on the SQLite connection before closing.
+    ///
+    /// The SQLite manual recommends using this for long-lived databases.
+    ///
+    /// This will collect and store statistics about the layout of data in your tables to help the query planner make better decisions.
+    /// Over the connection's lifetime, the query planner will make notes about which tables could use up-to-date statistics so this
+    /// command doesn't have to scan the whole database every time. Thus, the best time to execute this is on connection close.
+    ///
+    /// `analysis_limit` sets a soft limit on the maximum number of rows to scan per index.
+    /// It is equivalent to setting [`Self::analysis_limit`] but only takes effect for the `PRAGMA optimize;` call
+    /// and does not affect the behavior of any `ANALYZE` statements made during the connection's lifetime.
+    ///
+    /// If not `None`, the `analysis_limit` here overrides the global `analysis_limit` setting,
+    /// but only for the `PRAGMA optimize;` call.
+    ///
+    /// Not enabled by default.
+    ///
+    /// See [the SQLite manual](https://www.sqlite.org/lang_analyze.html#automatically_running_analyze) for details.
+    pub fn optimize_on_close(
+        mut self,
+        enabled: bool,
+        analysis_limit: impl Into<Option<u32>>,
+    ) -> Self {
+        self.optimize_on_close = if enabled {
+            OptimizeOnClose::Enabled {
+                analysis_limit: (analysis_limit.into()),
+            }
+        } else {
+            OptimizeOnClose::Disabled
+        };
+        self
+    }
+
+    /// Set a soft limit on the number of rows that `ANALYZE` touches per index.
+    ///
+    /// This also affects `PRAGMA optimize` which is set by [Self::optimize_on_close].
+    ///
+    /// The value recommended by SQLite is `400`. There is no default.
+    ///
+    /// See [the SQLite manual](https://www.sqlite.org/lang_analyze.html#approx) for details.
+    pub fn analysis_limit(mut self, limit: impl Into<Option<u32>>) -> Self {
+        if let Some(limit) = limit.into() {
+            return self.pragma("analysis_limit", limit.to_string());
+        }
+        self.pragmas.insert("analysis_limit".into(), None);
+        self
+    }
+
     /// Register a regexp function that allows using regular expressions in queries.
     ///
     /// ```