In Chapter 4, Palmieri focuses on logging and telemetry. Axum is very different
from Actix, and my first foray into trying to understand it led me to
[Tower](https://docs.rs/tower/latest/tower/), the Rust community's de-facto
standards for modular networking development and design.
I completely short-circuited much of what the book recommended and, instead,
just went with the most basic implementation possible. I added the tracing
libraries as recommended by the Axum developers, and then implemented the first
level of tracing as recommended by Tower:
``` sh
$ cargo add --features tower_http/trace,tracing tower tower_http tracing tracing_subscriber
```
And then I updated the app startup code to include it:
``` rust
pub async fn app(configuration: &Settings) -> Router {
tracing_subscriber::fmt::init();
let pool = PgPoolOptions::new()
.max_connections(50)
.connect(&configuration.database.url())
.await
.expect("could not connect to database_url");
routes().layer(Extension(pool)).layer(TraceLayer::new_for_http())
}
```
That is literally all that was needed. And the output is:
``` plaintext
2023-03-25T16:49:06.385563Z DEBUG request{method=GET uri=/ version=HTTP/1.1}:
tower_http::trace::on_request: started processing request
2023-03-25T16:49:06.386270Z DEBUG request{method=GET uri=/ version=HTTP/1.1}:
tower_http::trace::on_response: finished processing request latency=0 ms
status=200
```
That's not great logging, but it's a start. As I understand it,
`tracing_subscriber::fmt::init()` initializes the formatter, but I'm confused as
to where this is saved or stored, since it seems to be... nowhere. The deeper
Rust gets, the wilder it seems.
What I did manage was to create, [as recommended by Chris
Allen](https://bitemyapp.com/blog/notes-on-zero2prod-rust/), a very simple Layer
that shoves a new object into the collection of data being passed around by the
request. That object contains a unique UUID for the session being processed.
Since Tokio is a multi-threaded system, having a UUID allows us to trace each
individual request from beginning to end... provided I've hooked up by handlers
just right.
I learned most of this by reading the [Axum Session source
code](https://docs.rs/axum-sessions/latest/src/axum_sessions/session.rs.html),
which implements something much more complex. Since we're at a deeper level of
the service handling I need a function takes a Request and returns a Response,
and in the middle inserts a SessionId into the Request passed in; by giving the
type a name any handlers can now find and use that SessionId:
``` rust
/// In file `session_id.rs`
pub struct SessionId(pub Uuid);
pub async fn session_id<B>(mut req: Request<B>, next: Next<B>)
-> Result<Response, StatusCode> {
req.extensions_mut().insert(SessionId(Uuid::new_v4()));
Ok(next.run(req).await)
}
```
With that, I now need to add it to the layers initialized with the app object:
```rust
/// In lib.rs:pub async fn app()`:
routes()
.layer(Extension(pool))
.layer(TraceLayer::new_for_http())
.layer(middleware::from_fn(session_id::session_id))
```
And with that, the SessionId is available. Since it's the outermost layer, it
can now be used by anything deeper in. Let's add it to the `subscribe`
function:
``` rust
/// In routes/subscribe.rs/subscribe()
pub(crate) async fn subscribe(
Extension(session): Extension<SessionId>,
Extension(pool): Extension<PgPool>,
payload: Option<Form<NewSubscription>>,
) -> Result<(StatusCode, ()), ZTPError> {
if let Some(payload) = payload {
// Multi-line strings in Rust. Ugly. Would have preferred a macro.
let sql = r#"INSERT INTO subscriptions
(id, email, name, subscribed_at)
VALUES ($1, $2, $3, $4);"#.to_string();
let subscription: Subscription = (&(payload.0)).into();
tracing::info!(
"request_id {} - Adding '{}' as a new subscriber.",
session.0.to_string(),
subscription.name
);
// ...
```
And with that, every Request now has a strong ID associated with it:
``` plaintext
2023-03-26T22:19:23.305421Z INFO ztp::routes::subscribe:
request_id d0f4a6e7-de0d-48bc-902b-713901c1d63b -
Adding 'Elf M. Sternberg' as a new subscriber.
```
That's a very noisy trace; I'd like to start knocking it down to something more
like a responsible log, or give me permission to format it the way I like. I'm
also getting incredibly noisy messages from the `sqlx::query` call, including
the text of the SQL template (the `let sql = ...` line above), which I really
don't need every time someone makes a request, and is horribly formatted for
principled analytics.
Configuring it to return JSON turned out to be easy, although my first pass
puzzled me. I had to turn `json` formatting on as a feature:
``` sh
$ cargo add --features=json tracing_subscriber
```
And then it was possible to configure the format:
``` rust
// in lib.rs:app()
// ...
let format = tracing_subscriber::fmt::format()
.with_level(false) // don't include levels in formatted output
.with_thread_names(true)
.json(); // include the name of the current thread
tracing_subscriber::fmt().event_format(format).init();
// ...
```
``` json
{
"timestamp":"2023-03-26T22:53:13.091366Z",
"fields": {
"message":"request_id 479014e2-5f13-4e12-8401-34d8f8bf1a18 - "
"Adding 'Elf M. Sternberg' as a new subscriber."},
"target":"ztp::routes::subscribe",
"threadName":"tokio-runtime-worker"
}
```
This pretty much concludes my week-long foray into Palmieri's book; I'm not
going to worry too much about the deployment stuff, since that's part of my
daytime job and I'm not interesting in going over it again.
Overall, this was an excellent book for teaching me many of the basics, and
provides a really good introduction into the way application servers can be
written in Rust. I disagree with the premise that "the language doesn't mean
anything to the outcome," as I've heard some people say, nor do I think using
Rust is some kind of badge of honor. Instead, I think it's a mark of a
responsible developer, one who can produce code that works well the first time,
and with some hard thinking about how types work (and some heavy-duty exposure
to Haskell), Rust development can be your first thought, not your "I need
speed!" thought, when developing HTTP-based application servers.
This chapter introduces the Actix "Extractors" for retrieving form data. I've
added tests to the `./tests` folder to attempt to interact with those
extractors; as of this commit, a89cbe5b, they fail because the example code
isn't there.
What is there is a variant of the "Hello, World!" code from the previous
exercises (section 3.5), which uses the Actix extractor:
``` rust
// Actix, *not* Axum. Does not work with the current framework.
fn index(form: web::Form<FormData>) -> String {
format!('Welcome {}!', form.username)
}
```
Translated and polished into Axum, it translates to:
``` rust
pub async fn index(payload: Option<Form<FormData>>) -> impl IntoResponse {
let username = payload.map_or("World".to_string(), move |index| -> String {
String::from(&(index.username))
});
(StatusCode::OK, format!("Hello, {}!\n", &username))
}
```
The Axum version is a little smarter, providing a default "World!" if you don't
specify a name. That's what `.map_or` does, although the `or` part actually
comes first in the function. So the result is:
``` sh
$ curl http://localhost:3000/
Hello, World!
$ curl 'http://localhost:3000/?username=Spock'
Hello, Spock!
```
Which is more or less the version we want.
**Section 3.7.3** then goes into some detail on the implementation of a Trait.
A Trait in Rust is like an Interface in other languages; it describes a
collection of functions for manipulating the values found in a defined Type.
**Types**: A Type is just a description of the value: `u16` is a sixteen-bit
unsigned integer; `char` is a Unicode character, and it's size is always 32bits,
but a `String` is a bunch of things: it's an array of characters (which are not
always `char`!), a length for that string and a capacity. If the String is
manipulated to exceed the capacity, the array is re-allocated to a new capacity
and the old array copied into it. A `Vec<String>` is an array of Strings; as a
Type, it is considered to have a single value: whatever is in it at the moment
you use it.
**Trait**: A Trait defines a collection of one or more functions that can
operate on a value.
The truly nifty thing about Traits is that they can be implemented after the
fact. By importing a Trait and an implementation of that trait specific to a
type into a module containing that type, you can extend the behavior of a type
in a deterministic way without having to modify or inherit the code, as you
would in an object-oriented language.
Axum has a valuable trait, `FromRequest`. For any structure you can imagine
passing from the client to the server, you can implement `FromRequest` for that
object and any content in the body of the message will be transformed into that
structure.
We've seen a trait before: `IntoResponse`, written as `impl IntoResponse`, and
is the output (the return type) of many of the functions that produce return
values for our application server. In this case the return type instructs Rust
to look in the current lexical scope and, for the value returned by that
function, determine if an `IntoResponse` trait has been defined for it. If it
has, the value will be returned because Axum has now been assured that there
exists a function to convert that value into something streamable and usable as
an HTTP response.
Fortunately for us, Axum has already implemented `FromRequest` for all the
native data types, as well as some structures and arrays. Even better, it has
implemented `FromRequest` for the Serde serialization/deserialization library.
So in this example:
``` rust
pub struct FormData {
username: String,
}
pub async fn index(payload: Option<Form<FormData>>) -> impl IntoResponse { ...
```
A `Form` (something using the `application/x-www-form-urlencoded` protocol) of
`FormData` will automatically be converted into a `payload` object of `{
username: "Spock" )`, and in this case wrapped in a `Some()` handler. (Or
`None`, if there was no form included.) <aside>So far, there's not too much
bloat in this product; with all the debugging symbols, it's 60MB or so, but
stripped to the bone it's only 3.1MB, tolerable for modern deployments.</aside>
First, though, we must adjust our `valid_subscription` test:
``` rust
let body = "name=le%20guin&email=ursula_le_guin%40gmail.com";
let response = hyper::Client::new()
.request(
Request::builder()
.method("POST")
.header("content-type", "application/x-www-form-urlencoded")
.uri(format!("http://{}/subscriptions", addr))
.body(Body::from(body))
.unwrap(),
)
.await
.expect("Failed to execute request.");
```
Two updates from the book: first, we're sending it via POST instead of GET. This
is the correct way to do things; a GET should never (and I mean *never*) cause a
change of state on the back-end. To send something new that the server will
process and store, you use a POST. (To update something, or to send something to
a known *and unique* URI, PUT is better.) Secondly, since we're using a generic
form-data object, we need to set the content-type on the client so that the
server is informed of how to unpack this payload. The '%20' and '%40' markers in
the `body` are the space and the `@` respectively.
I completely ignored the advice in the book and went instead with
[Dbmate](https://github.com/amacneil/dbmate); Dbmate is a bit cranky; your SQL
must be very much nestled against the 'up' and 'down' markers in the migration
file, and it seems to be quite opinionated about everything being lowercase.
That said, it was trivial to create a database with it:
``` sh
$ dbmate new create_subscriptions_table
```
This will create the folder
`db/migrations/20230322174957_create_subscriptions_table.sql`,
(The timestamp will be different, obviously), and in this file you put the
following, as specified in the book:
``` sql
-- migrate:up
create table subscriptions (
id uuid not null,
primary key (id),
email text not null unique,
name text not null,
subscribed_at timestamptz not null
);
-- migrate:down
drop table subscriptions;
```
To use Dbmate, you have to specify how to connect. I'm using Postgres, so let's
start with creating a database and a user for it:
``` sh
$ sudo -u postgres psql
[sudo] possword for user: ...................
postgres=# create database newsletter;
CREATE DATABASE
postgres=# create user newletter with encrypted password 'redacted';
CREATE USER
postgres=# grant all privileges on database newsletter to newsletter;
GRANT
postgres=# exit
```
In your project root, create a `.env` file to specify your connection:
``` sh
DATABASE_URL="postgres://newsletter:redacted@127.0.0.1:5432/newsletter?sslmode=disable"
```
The `sslmode` flag there is necessary for localhost connections, as Dbmate
assumes an encrypted connection by default, but we're isolating to a local
connection that is, usually, safe.
With the new entry in your `.env` file, you can now run a migration:
``` sh
$ dbmate up
Writing: ./db/schema.sql
```
Running `dbmate up` will automatically create the database for you if it hasn't
already; `dbmate migrate` also performs migrations, but it will not create the
database.
Now you can re-connect to Postgres as the newsletter user and see what you've
got:
``` sh
$ psql --user newsletter -h localhost --password
Password:
psql (14.7 (Ubuntu 14.7-0ubuntu0.22.04.1), server 11.7 (Ubuntu 11.7-0ubuntu0.19.10.1))
newsletter=> \d
List of relations
Schema | Name | Type | Owner
--------+-------------------+-------+------------
public | schema_migrations | table | newsletter
public | subscriptions | table | newsletter
(2 rows)
newsletter=> \d subscriptions
Table "public.subscriptions"
Column | Type | Collation | Nullable | Default
---------------+--------------------------+-----------+----------+---------
id | uuid | | not null |
email | text | | not null |
name | text | | not null |
subscribed_at | timestamp with time zone | | not null |
Indexes:
"subscriptions_pkey" PRIMARY KEY, btree (id)
"subscriptions_email_key" UNIQUE CONSTRAINT, btree (email)
```
Note that Dbmate has allocated a table to itself, `schema_migrations`, for
tracking what it's done to your system and when. Try not to conflict with it,
okay?
Every complex app has a configuration, and there are plenty of different ways
the configuration can be specified. Environment variables, internal defaults,
and configuration files-- the last of which comes in so many different flavors.
Rust has a well-known [config](https://docs.rs/config/latest/config/index.html)
crate that supports all the most common configurations: YAML, JSON, TOML; you
can even add your own by writing something that implements the `config::Format`
trait. Add it to Cargo.toml:
``` sh
$ cargo add config
```
For the meantime, we're just going to create a new file, called
`configuration.rs`, and put our configuration details in there. Right now we
have a single configuration detail: the port.
I'm going to go above and beyond Lucas here and configure some internal defaults
for my code. It will have expectations. First, you have to tell Serde that
there will be default values:
``` rust
use config::Config;
pub struct Settings {
pub port: u16,
}
```
Then, you have to set those default values. Fortunately, Rust provides a "set
default values" trait named, sensibly enough, Default:
``` rust
impl Default for Settings {
fn default() -> Self {
Settings { port: 3001 }
}
}
```
Again, exceeding the book's parameters, I'm going to say that if the file is
missing the default parameters should hold:
``` rust
pub(crate) fn get_configuration() -> Result<Settings, config::ConfigError> {
Config::builder()
.add_source(config::File::with_name("./ztd.config").required(false))
.build()?
.try_deserialize()
}
```
And since this is the first time I'm doing this, I'm going to write a test to
assert that my understanding of how this all works is correct:
``` rust
mod tests {
use super::*;
#[test]
fn test_for_defaults() {
let maybe_config = get_configuration();
assert!(!maybe_config.is_err());
let config = maybe_config.unwrap();
assert_eq!(config.port, 3001);
}
}
```
Elf M. Sternberg