For this guide, we’re going to walk through some simple examples for each of the pieces of CRUD, which stands for “Create Read Update Delete”. Each step in this guide will build on the previous and is meant to be followed along.
Before we start, make sure you have one of PostgreSQL, SQLite, or MySQL installed and running. In the project repository, you may find various examples for every supported database.
Initializing a new project
The first thing we need to do is generate our project.
cargo new --lib diesel_demo
cd diesel_demo
First, let’s add Diesel to our dependencies. We’re also going to use a tool called .env
to manage our environment variables for us. We’ll add it to our dependencies as well.
[dependencies]
diesel = { version = "2.2.0", features = ["postgres"] }
# build libpq and openssl as part of the build process
# uncomment these lines if you run into setup issues
# pq-sys = { version = "0.6", features = ["bundled"] }
# openssl-sys = { version = "0.9.100", features = ["vendored"] }
dotenvy = "0.15"
[dependencies]
diesel = { version = "2.2.0", features = ["sqlite", "returning_clauses_for_sqlite_3_35"] }
# build libsqlite3 as part of the build process
# uncomment this line if you run into setup issues
# libsqlite3-sys = { version = "0.30", features = ["bundled"] }
dotenvy = "0.15"
[dependencies]
diesel = { version = "2.2.0", features = ["mysql"] }
# build libmysqlclient as part of the build process
# uncomment this line if you run into setup issues
# mysqlclient-sys = { version = "0.4", features = ["bundled"] }
dotenvy = "0.15"
Installing Diesel CLI
Diesel provides a separate CLI tool to help manage your project. Since it’s a standalone binary, and doesn’t affect your project’s code directly, we don’t add it to Cargo.toml
. Instead, we just install it on our system.
We provide pre-built binaries for diesel cli. You can install the command line tool via:
# Linux/MacOS
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/diesel-rs/diesel/releases/latest/download/diesel_cli-installer.sh | sh
# Windows
powershell -c "irm https://github.com/diesel-rs/diesel/releases/latest/download/diesel_cli-installer.ps1 | iex"
You also can use cargo-binstall
for this. To install diesel cli with cargo binstall
run the following command:
cargo binstall diesel_cli
This automatically finds the right binary for your system.
Alternatively you can manually install diesel cli by using cargo install:
cargo install diesel_cli
Setup Diesel for your project
We need to tell Diesel where to find our database. We do this by setting the DATABASE_URL
environment variable. On our development machines, we’ll likely have multiple projects going, and we don’t want to pollute our environment. We can put the url in a .env
file instead.
echo DATABASE_URL=postgres://username:password@localhost/diesel_demo > .env
echo DATABASE_URL=/path/to/your/sqlite/database.db > .env
echo DATABASE_URL=mysql://username:password@localhost/diesel_demo > .env
Now Diesel CLI can set everything up for us.
diesel setup
This will create our database (if it didn’t already exist) and set up the initial migrations directory, which will contain a generated migration file that establishes the Diesel setup. Note that the migrations directory will not be empty as the initial setup migration is automatically generated.
Now we’re going to write a small CLI that lets us manage a blog (ignoring the fact that we can only access the database from this CLI…). The first thing we’re going to need is a table to store our posts. Let’s create a migration for that:
diesel migration generate create_posts
Diesel CLI will create two empty files for us in the required structure. You’ll see output that looks something like this:
Creating migrations/20160815133237_create_posts/up.sql
Creating migrations/20160815133237_create_posts/down.sql
Migrations allow us to evolve the database schema over time. Each migration consists of an up.sql
file to apply the changes and a down.sql
file to revert them. Applying and immediately reverting a migration should leave your database schema unchanged.
Next, we’ll write the SQL for migrations:
CREATE TABLE posts (
id SERIAL PRIMARY KEY,
VARCHAR NOT NULL,
title body TEXT NOT NULL,
BOOLEAN NOT NULL DEFAULT FALSE
published )
CREATE TABLE posts (
id INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT,
VARCHAR NOT NULL,
title body TEXT NOT NULL,
BOOLEAN NOT NULL DEFAULT 0
published )
CREATE TABLE posts (
id INTEGER AUTO_INCREMENT PRIMARY KEY,
VARCHAR(255) NOT NULL,
title body TEXT NOT NULL,
BOOLEAN NOT NULL DEFAULT FALSE
published );
DROP TABLE posts
DROP TABLE posts
DROP TABLE posts
We can apply our new migration:
diesel migration run
It’s a good idea to ensure that down.sql
is correct. You can quickly confirm that your down.sql
rolls back your migration correctly by redoing
the migration:
diesel migration redo
Write Rust
OK enough SQL, let’s write some Rust. We’ll start by writing some code to show the last five published posts. The first thing we need to do is establish a database connection.
use diesel::prelude::*;
use dotenvy::dotenv;
use std::env;
pub fn establish_connection() -> PgConnection {
.ok();
dotenv()
let database_url = env::var("DATABASE_URL").expect("DATABASE_URL must be set");
PgConnection::establish(&database_url)
.unwrap_or_else(|_| panic!("Error connecting to {}", database_url))
}
use diesel::prelude::*;
use dotenvy::dotenv;
use std::env;
pub fn establish_connection() -> SqliteConnection {
.ok();
dotenv()
let database_url = env::var("DATABASE_URL").expect("DATABASE_URL must be set");
SqliteConnection::establish(&database_url)
.unwrap_or_else(|_| panic!("Error connecting to {}", database_url))
}
use diesel::prelude::*;
use dotenvy::dotenv;
use std::env;
pub fn establish_connection() -> MysqlConnection {
.ok();
dotenv()
let database_url = env::var("DATABASE_URL").expect("DATABASE_URL must be set");
MysqlConnection::establish(&database_url)
.unwrap_or_else(|_| panic!("Error connecting to {}", database_url))
}
We’ll also want to create a Post
struct into which we can read our data, and have diesel generate the names we’ll use to reference tables and columns in our queries.
We’ll add the following lines to the top of src/lib.rs
:
pub mod models;
pub mod schema;
pub mod models;
pub mod schema;
pub mod models;
pub mod schema;
Next we need to create the two modules that we just declared.
use diesel::prelude::*;
#[derive(Queryable, Selectable)]
#[diesel(table_name = crate::schema::posts)]
#[diesel(check_for_backend(diesel::pg::Pg))]
pub struct Post {
pub id: i32,
pub title: String,
pub body: String,
pub published: bool,
}
use diesel::prelude::*;
#[derive(Queryable, Selectable)]
#[diesel(table_name = crate::schema::posts)]
#[diesel(check_for_backend(diesel::sqlite::Sqlite))]
pub struct Post {
pub id: i32,
pub title: String,
pub body: String,
pub published: bool,
}
use diesel::prelude::*;
#[derive(Queryable, Selectable)]
#[diesel(table_name = crate::schema::posts)]
#[diesel(check_for_backend(diesel::mysql::Mysql))]
pub struct Post {
pub id: i32,
pub title: String,
pub body: String,
pub published: bool,
}
#[derive(Queryable)]
will generate all of the code needed to load a Post
struct from a SQL query. #[derive(Selectable)]
will generate code to construct a matching select clause based on your model type based on the table defined via #[diesel(table_name = crate::schema::posts)]
. #[diesel(check_for_backend(diesel::pg::Pg))
(or sqlite::SQLite
or mysql::MySQL
) adds additional compile time checks to verify that all field types in your struct are compatible with their corresponding SQL side expressions. This part is optional, but it greatly improves the generated compiler error messages.
Typically the schema module isn’t created by hand, it gets generated by diesel CLI. When we ran diesel setup
, a file called diesel.toml was created which tells Diesel to maintain a file at src/schema.rs
for us. The file should look like this:
diesel::table! {
{
posts (id) -> Int4,
id -> Varchar,
title -> Text,
body -> Bool,
published }
}
diesel::table! {
{
posts (id) -> Integer,
id -> Text,
title -> Text,
body -> Bool,
published }
}
diesel::table! {
{
posts (id) -> Integer,
id -> Varchar,
title -> Text,
body -> Bool,
published }
}
The exact output might vary slightly depending on your database, but it should be equivalent.
The table!
macro creates a bunch of code based on the database schema to represent all of the tables and columns. We’ll see how exactly to use that in the next example. For a deep dive into the generated code see the Schema in Depth guide.
Any time we run or revert a migration, this file will get automatically updated.
Let’s write the code to actually show us our posts.
use self::models::*;
use diesel::prelude::*;
use diesel_demo::*;
fn main() {
use self::schema::posts::dsl::*;
let connection = &mut establish_connection();
let results = posts
.filter(published.eq(true))
.limit(5)
.select(Post::as_select())
.load(connection)
.expect("Error loading posts");
println!("Displaying {} posts", results.len());
for post in results {
println!("{}", post.title);
println!("-----------\n");
println!("{}", post.body);
}
}
use self::models::*;
use diesel::prelude::*;
use diesel_demo::*;
fn main() {
use self::schema::posts::dsl::*;
let connection = &mut establish_connection();
let results = posts
.filter(published.eq(true))
.limit(5)
.select(Post::as_select())
.load(connection)
.expect("Error loading posts");
println!("Displaying {} posts", results.len());
for post in results {
println!("{}", post.title);
println!("-----------\n");
println!("{}", post.body);
}
}
use self::models::*;
use diesel::prelude::*;
use diesel_demo::*;
fn main() {
use self::schema::posts::dsl::*;
let connection = &mut establish_connection();
let results = posts
.filter(published.eq(true))
.limit(5)
.select(Post::as_select())
.load(connection)
.expect("Error loading posts");
println!("Displaying {} posts", results.len());
for post in results {
println!("{}", post.title);
println!("-----------\n");
println!("{}", post.body);
}
}
The use self::schema::posts::dsl::*
line imports a bunch of aliases so that we can say posts
instead of posts::table
, and published
instead of posts::published
. It’s useful when we’re only dealing with a single table, but that’s not always what we want. It’s always important to keep imports to schema::table::dsl::*
inside of the current function to prevent polluting the module namespace.
We can run our script with cargo run --bin show_posts
. Unfortunately, the results won’t be terribly interesting, as we don’t actually have any posts in the database. Still, we’ve written a decent amount of code, so let’s commit.
The full code for the demo at this point can be found here for PostgreSQL, SQLite and MySQL.
Next, let’s write some code to create a new post. We’ll want a struct to use for inserting a new record.
use crate::schema::posts;
#[derive(Insertable)]
#[diesel(table_name = posts)]
pub struct NewPost<'a> {
pub title: &'a str,
pub body: &'a str,
}
use crate::schema::posts;
#[derive(Insertable)]
#[diesel(table_name = posts)]
pub struct NewPost<'a> {
pub title: &'a str,
pub body: &'a str,
}
use crate::schema::posts;
#[derive(Insertable)]
#[diesel(table_name = posts)]
pub struct NewPost<'a> {
pub title: &'a str,
pub body: &'a str,
}
Now let’s add a function to save a new post.
use self::models::{NewPost, Post};
pub fn create_post(conn: &mut PgConnection, title: &str, body: &str) -> Post {
use crate::schema::posts;
let new_post = NewPost { title, body };
diesel::insert_into(posts::table)
.values(&new_post)
.returning(Post::as_returning())
.get_result(conn)
.expect("Error saving new post")
}
use self::models::{NewPost, Post};
pub fn create_post(conn: &mut SqliteConnection, title: &str, body: &str) -> Post {
use crate::schema::posts;
let new_post = NewPost { title, body };
diesel::insert_into(posts::table)
.values(&new_post)
.returning(Post::as_returning())
.get_result(conn)
.expect("Error saving new post")
}
use self::models::{NewPost, Post};
pub fn create_post(conn: &mut MysqlConnection, title: &str, body: &str) -> Post {
use crate::schema::posts;
let new_post = NewPost { title, body };
.transaction(|conn| {
conndiesel::insert_into(posts::table)
.values(&new_post)
.execute(conn)?;
posts::table
.order(posts::id.desc())
.select(Post::as_select())
.first(conn)
})
.expect("Error while saving post")
}
When we call .get_result
on an insert or update statement, it automatically adds RETURNING *
to the end of the query, and lets us load it into any struct that implements Queryable
for the right types. Neat!
Diesel can insert more than one record in a single query. Just pass a Vec
or slice to insert_into
, and then call get_results
instead of get_result
. If you don’t actually want to do anything with the row that was just inserted, call .execute
instead. The compiler won’t complain at you, that way. :)
Now that we’ve got everything set up, we can create a little script to write a new post.
use diesel_demo::*;
use std::io::{stdin, Read};
fn main() {
let connection = &mut establish_connection();
let mut title = String::new();
let mut body = String::new();
println!("What would you like your title to be?");
.read_line(&mut title).unwrap();
stdin()let title = title.trim_end(); // Remove the trailing newline
println!("\nOk! Let's write {title} (Press {EOF} when finished)\n",);
.read_to_string(&mut body).unwrap();
stdin()
let post = create_post(connection, title, &body);
println!("\nSaved draft {title} with id {}", post.id);
}
#[cfg(not(windows))]
const EOF: &str = "CTRL+D";
#[cfg(windows)]
const EOF: &str = "CTRL+Z";
use diesel_demo::*;
use std::io::{stdin, Read};
fn main() {
let connection = &mut establish_connection();
let mut title = String::new();
let mut body = String::new();
println!("What would you like your title to be?");
.read_line(&mut title).unwrap();
stdin()let title = title.trim_end(); // Remove the trailing newline
println!("\nOk! Let's write {title} (Press {EOF} when finished)\n",);
.read_to_string(&mut body).unwrap();
stdin()
let post = create_post(connection, title, &body);
println!("\nSaved draft {title} with id {}", post.id);
}
#[cfg(not(windows))]
const EOF: &str = "CTRL+D";
#[cfg(windows)]
const EOF: &str = "CTRL+Z";
use diesel_demo::*;
use std::io::{stdin, Read};
fn main() {
let connection = &mut establish_connection();
let mut title = String::new();
let mut body = String::new();
println!("What would you like your title to be?");
.read_line(&mut title).unwrap();
stdin()let title = title.trim_end(); // Remove the trailing newline
println!("\nOk! Let's write {title} (Press {EOF} when finished)\n",);
.read_to_string(&mut body).unwrap();
stdin()
let post = create_post(connection, title, &body);
println!("\nSaved draft {title} with id {}", post.id);
}
#[cfg(not(windows))]
const EOF: &str = "CTRL+D";
#[cfg(windows)]
const EOF: &str = "CTRL+Z";
We can run our new script with cargo run --bin write_post
. Go ahead and write a blog post. Get creative! Here was mine:
Compiling diesel_demo v0.1.0 (file:///Users/sean/Documents/Projects/open-source/diesel_demo)
Running `target/debug/write_post`
What would you like your title to be?
Diesel demo
Ok! Let's write Diesel demo (Press CTRL+D when finished)
You know, a CLI application probably isn't the best interface for a blog demo.
But really I just wanted a semi-simple example, where I could focus on Diesel.
I didn't want to get bogged down in some web framework here.
Saved draft Diesel demo with id 1
Unfortunately, running show_posts
still won’t display our new post, because we saved it as a draft. If we look back to the code in show_posts
, we added .filter(published.eq(true))
, and we had published
default to false in our migration. We need to publish it! But in order to do that, we’ll need to look at how to update an existing record. First, let’s commit. The code for this demo at this point can be found here for PostgreSQL, SQLite, and MySQL.
Now that we’ve got create and read out of the way, update is actually relatively simple. Let’s jump right into the script:
use self::models::Post;
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::{posts, published};
let id = args()
.nth(1)
.expect("publish_post requires a post id")
.parse::<i32>()
.expect("Invalid ID");
let connection = &mut establish_connection();
let post = diesel::update(posts.find(id))
.set(published.eq(true))
.returning(Post::as_returning())
.get_result(connection)
.unwrap();
println!("Published post {}", post.title);
}
use self::models::Post;
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::{posts, published};
let id = args()
.nth(1)
.expect("publish_post requires a post id")
.parse::<i32>()
.expect("Invalid ID");
let connection = &mut establish_connection();
let post = diesel::update(posts.find(id))
.set(published.eq(true))
.returning(Post::as_returning())
.get_result(connection)
.unwrap();
println!("Published post {}", post.title);
}
use self::models::Post;
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::{posts, published};
let id = args()
.nth(1)
.expect("publish_post requires a post id")
.parse::<i32>()
.expect("Invalid ID");
let connection = &mut establish_connection();
let post = connection
.transaction(|connection| {
let post = posts.find(id).select(Post::as_select()).first(connection)?;
diesel::update(posts.find(id))
.set(published.eq(true))
.execute(connection)?;
Ok(post)
})
.unwrap_or_else(|_: diesel::result::Error| panic!("Unable to find post {}", id));
println!("Published post {}", post.title);
}
That’s it! Let’s try it out with cargo run --bin publish_post 1
.
Compiling diesel_demo v0.1.0 (file:///Users/sean/Documents/Projects/open-source/diesel_demo)
Running `target/debug/publish_post 1`
Published post Diesel demo
Additionally, let’s implement a possibility of fetching a single post. We will display the post id with its title. Notice .optional()
call. This returns Option<Post>
instead of throwing an error, which we can then use in our matching pattern. For additional methods to modify the constructed select statements refer to the documentation
of QueryDsl
use self::models::Post;
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::posts;
let post_id = args()
.nth(1)
.expect("get_post requires a post id")
.parse::<i32>()
.expect("Invalid ID");
let connection = &mut establish_connection();
let post = posts
.find(post_id)
.select(Post::as_select())
.first(connection)
.optional(); // This allows for returning an Option<Post>, otherwise it will throw an error
match post {
Ok(Some(post)) => println!("Post with id: {} has a title: {}", post.id, post.title),
Ok(None) => println!("Unable to find post {}", post_id),
Err(_) => println!("An error occured while fetching post {}", post_id),
}
}
use self::models::Post;
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::posts;
let post_id = args()
.nth(1)
.expect("get_post requires a post id")
.parse::<i32>()
.expect("Invalid ID");
let connection = &mut establish_connection();
let post = posts
.find(post_id)
.select(Post::as_select())
.first(connection)
.optional(); // This allows for returning an Option<Post>, otherwise it will throw an error
match post {
Ok(Some(post)) => println!("Post with id: {} has a title: {}", post.id, post.title),
Ok(None) => println!("Unable to find post {}", post_id),
Err(_) => println!("An error occured while fetching post {}", post_id),
}
}
use self::models::Post;
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::posts;
let post_id = args()
.nth(1)
.expect("get_post requires a post id")
.parse::<i32>()
.expect("Invalid ID");
let connection = &mut establish_connection();
let post = posts
.find(post_id)
.select(Post::as_select())
.first(connection)
.optional(); // This allows for returning an Option<Post>, otherwise it will throw an error
match post {
Ok(Some(post)) => println!("Post with id: {} has a title: {}", post.id, post.title),
Ok(None) => println!("Unable to find post {}", post_id),
Err(_) => println!("An error occured while fetching post {}", post_id),
}
}
We can see our post with cargo run --bin get_post 1
.
Compiling diesel_demo v0.1.0 (file:///Users/sean/Documents/Projects/open-source/diesel_demo)
Running `target/debug/get_post 1`
Post with id: 1 has a title: Diesel demo
And now, finally, we can see our post with cargo run --bin show_posts
.
Running `target/debug/show_posts`
Displaying 1 posts
Diesel demo
----------
You know, a CLI application probably isn't the best interface for a blog demo.
But really I just wanted a semi-simple example, where I could focus on Diesel.
I didn't want to get bogged down in some web framework here.
We’ve still only covered three of the four letters of CRUD though. Let’s show how to delete things. Sometimes we write something we really hate, and we don’t have time to look up the ID. So let’s delete based on the title, or even just some words in the title.
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::*;
let target = args().nth(1).expect("Expected a target to match against");
let pattern = format!("%{}%", target);
let connection = &mut establish_connection();
let num_deleted = diesel::delete(posts.filter(title.like(pattern)))
.execute(connection)
.expect("Error deleting posts");
println!("Deleted {} posts", num_deleted);
}
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::*;
let target = args().nth(1).expect("Expected a target to match against");
let pattern = format!("%{}%", target);
let connection = &mut establish_connection();
let num_deleted = diesel::delete(posts.filter(title.like(pattern)))
.execute(connection)
.expect("Error deleting posts");
println!("Deleted {} posts", num_deleted);
}
use diesel::prelude::*;
use diesel_demo::*;
use std::env::args;
fn main() {
use self::schema::posts::dsl::*;
let target = args().nth(1).expect("Expected a target to match against");
let pattern = format!("%{}%", target);
let connection = &mut establish_connection();
let num_deleted = diesel::delete(posts.filter(title.like(pattern)))
.execute(connection)
.expect("Error deleting posts");
println!("Deleted {} posts", num_deleted);
}
We can run the script with cargo run --bin delete_post demo
(at least with the title I chose). Your output should look something like:
Compiling diesel_demo v0.1.0 (file:///Users/sean/Documents/Projects/open-source/diesel_demo)
Running `target/debug/delete_post demo`
Deleted 1 posts
When we try to run cargo run --bin show_posts
again, we can see that the post was in fact deleted. This barely scratches the surface of what you can do with Diesel, but hopefully this tutorial has given you a good foundation to build off of. We recommend exploring the API docs to see more. The final code for this tutorial can be found here for PostgreSQL, SQLite, and MySQL.