Terracotta

This repository provides a boilerplate webserver application, based on Axum, to act as a foundation and springboard for building web applications and APIs. The intention is for it to be used as a template for new projects.

The name Terracotta was chosen because it's rusty in colour, and clay represents something that can be moulded into different shapes.

It is intended to be easy to use and understand, easy to set up and extend, and easy to deploy.

Terracotta exists as a crate on crates.io to establish presence and gain visibility and awareness, and because there are plans to add command-line functionality to help with setup. It is not intended to be used as a library, and is not published as such. (See the Usage section for more information.) It may also be useful to be able to run it and see it working before then using it as a foundation for a new project.

Terracotta was created in response to the lack of full examples of how to use Axum, and the fact that many tutorials are out-of-date, lacking important elements, or just plain wrong. You may not need everything provided — and you also may well not agree with how some parts are implemented — but if you are wanting a leg-up to save some time, it's not a bad place to start!

The main sections in this README are:

Features

The main high-level points of note are:

Authentication

Terracotta features a custom-rolled authentication system, to demonstrate how to implement a basic session-based setup. Although storing a user list in plain text on a server is okay for small, limited projects and experiments, it is highly recommended to store the credentials securely in a database. That is currently outside the scope of this project, for a number of reasons.

In a real system you will probably also want to store the sessions in a database instead of in memory.

It is also worth noting that the pattern implemented is the best and most ideal for an application that serves HTML to a browser. If you are creating an API then you will want to change some of the behaviour to return HTTP status codes to tell the client that the request is unauthorised, rather than sending a login page. Equally, you will likely want to implement JWT or similar. That is also outside of scope at present, partly because there are various options to choose from.

The authentication system is set up to make it easy to configure routes as either public or protected, and is fully-implemented including a login page, logout action, and handling of every part of the authentication journey and the possible situations.

Error-handling

Terracotta has an opinionated approach to handling errors, including both HTTP errors and "true" (Rust) errors. This serves as a baseline to build on or to change as required.

Databases

Terracotta very purposefully does not include any kind of database integration. There are so many, and such a plethora of crates to choose from, that this is best left to the application developer to decide. Database interaction is very straightforward and so this is a simple addition to make.

Templates

The choice of Tera is unlikely to upset anyone, but if there is a preferred option then it is easy to change or remove. Tera has been implemented in a slightly opinionated manner, but it should be clear what has been changed from the defaults and how.

Usage

The Terracotta repository is designed so that it can be used as a template for new projects, and then customised and extended. You will naturally rename the project and tailor it to your needs, and as you implement your own features it will get harder and harder to merge in any upstream changes. It is therefore likely best to consider this a starting point only, and an upgrade reference, rather than an on-going contributing source.

Note that Terracotta is not designed to be used as a library, and its existence on crates.io is as a binary. This is to establish presence, but also there are plans for command-line tools to be added.

Getting started

The Terracotta repository is set up as a template repository on GitHub, so that you can easily click the "Use this template" button to create a new repository based on it. You can then clone your new repository and start working on it. This will give you a starting point where you have all the project files, but none of the commit history. This is the currently-recommended way to start a new project using Terracotta (there are plans to also have a command-line tool at a later date).

You may instead decide that you want to fork the repository, or clone it and then push it to a new repository. This is also fine, but you should be aware of the following points:

For these reasons, forking in the GitHub-recognised sense is not recommended, and cloning and pushing to a new repository is only recommended if you are comfortable with the caveats mentioned above.

To see an example of a project that has been created based on Terracotta, you can look at the Rustmark application. This shows how Terracotta can be used as a starting point, and then extended to create a more complex application. Note that in the case of Rustmark, the decision was made to actually fork Terracotta at a stage before its initial release, as the commit history was considered useful, and there were no release commits or tags to cause the conflict issues mentioned above. However, after that point of inception, all Terracotta updates have been applied manually, and it is not a "true" fork in GitHub terms.

Structure

The code in this repository follows a simple and straightforward layout, which is intended to be easy to understand and extend. You should absolutely modify the file structure to suit the shape of your own web application.

The basic folder structure is:

The layout of each folder should be fairly self-explanatory, but it is worth mentioning that the src folder represents the simplest sensible minimum. In a proper application it is likely that the handlers should be split out into more files, and there would also be various other supporting files too. Rather than dictate a layout, it is left as an exercise for the reader to implement their preferred approach.

Any images and other files that need to be protected by authentication should be placed in the content directory. Public images should be placed in static/img, and will be served from the /img URL path, similar to the CSS, JS, and WebFont files.

All of the content and static material is included in the compiled binary, making it very straightforward to deploy.

Coding standards

The code in this repository follows some specific and opinionated [coding standards][]. These mostly follow typical community conventions, but notable differences are the use of tabs for indentation, the alignment of various terms to aid readability, the use of comment headers to separate sections of code, and the usage of Nerd Font symbols in those headers to belay semantic meaning in order to apply highlighting.

You may well dislike aspects of the coding style, which is fine — feel free to change things, and make the code your own! Individuality is important.

Setup

The steps to set up a Terracotta project are simple and standard. You need a reasonably-recent Rust environment, on a Linux machine. There are currently no special requirements beyond what is needed to build a standard Rust project.

Note that these instructions are for building the application yourself, which will usually be in context of having used Terracotta as a template for a new project. In this case these steps will apply for your project too. You can also download the crate using cargo install terracotta, which will install the latest version of Terracotta from crates.io, but this currently is not particularly useful beyond letting you poke at the default, running application without having to clone the repository and build it yourself, to see if you like it. See the Getting started section for more information on creating your project using Terracotta as a template.

Environment

There are some key points to note about the environment you choose:

Typically, you will set up Rust using rustup, which is the recommended way to install Rust. The stable toolchain is targeted, as the focus is on stability and correctness, rather than bleeding-edge features.

Once you have Rust installed, you can build the project using cargo build. This will download and compile all dependencies, and build the project. You can then run the project using cargo run.

Configuration

Terracotta is configured using a TOML file. The default configuration file is Config.toml, which should be placed in the same directory as the binary. The configuration settings (and file) are optional, and if not provided, Terracotta will use default values for all configuration options.

It is also possible to pass configuration parameters from the command line, as environment variables. The environment variables take precedence over the configuration file options.

General options

The following options should be specified without any heading:

As shown here:

toml host = "127.0.0.1" port = 8000 logdir = "log" title = "Terracotta"

Local loading options

By default, all resources are baked into the binary, and served from there. This is the most efficient way to run the application, but it is also possible to load resources from the local filesystem, which can be useful for development and testing, and when there are large content files.

It is possible to supplement or override static assets. Static assets are subdivided into protected and public.

The following options should be specified under a [local_loading] heading:

Each of these options can be one of the following values:

As shown here:

toml [local_loading] protected_assets = "Override" # default is "Deny" public_assets = "Override" # default is "Deny"

For those options that allow loading from the local filesystem, the following options can be specified under a [local_paths] heading:

As shown here:

toml [local_paths] protected_assets = "content" public_assets = "static"

An example is provided, rustacean-flat-happy.png, which is available through http://localhost:8000/rustacean-flat-happy.png if using the settings in the example configuration file. This is a protected asset, and so will only be served to logged-in users.

Static file options

When static files are requested, the method by which they are served depends upon their source and size. All files baked into the binary are served directly from memory, and so these options do not apply to them. Files loaded from the local filesystem are loaded into memory and served all once if they are small enough, but past a certain (configurable) size they are streamed to the client.

The sizes of the stream buffer and read buffer are hugely important to performance, with smaller buffers greatly impacting download speeds. The default values have been carefully chosen based on extensive testing, and should not generally need to be changed. However, on a system with lots of users and very few large files it may be worth decreasing the buffer sizes to reduce memory usage when those files are requested, and on a system with very few users and lots of large files it may be worth increasing the buffer sizes to improve throughput. However, the chosen values are already within 5-10% of the very best possible speeds, so any increase should be made with caution. It is more likely that they would need to be decreased a little on a very busy system with a lot of large files, where the memory usage could become a problem and the raw speed of each download becomes a secondary concern.

The following options should be specified under a [static_files] heading:

Each of these options accepts an integer value.

As shown here:

toml [static_files] stream_threshold = 1000 # 1MiB — files above this size will be streamed stream_buffer = 256 # 256KB read_buffer = 128 # 128KB

User list

A list of user credentials can be specified under a [users] heading:

As shown here:

toml [users] joe = "1a2b3c"

This is a simple list of username/password pairs, where the username is the key and the password is the value. The password is stored in plain text, so be aware of the security implications of this (ideally you would implement an integration with your preferred database instead). The username and password are both case-sensitive.

Running

Terracotta can be run using the cargo run command, or by running the compiled binary directly. The server will listen on port 8000 by default, and will serve content from the static directory, plus any request handlers that you define. The static directory contains the static files to be served.

Testing

You can run the test suite using cargo test. This will run all unit and integration tests.

Note that, at present, there are no tests written specifically for this project, as it is mostly a combination of other crates from the Rust ecosystem. Tests might be added when the project is more mature and sensible things to test have been clearly identified.

Documentation

This is the first release, so there is not much in the way of documentation just yet. A few things may change when Axum 0.7 comes out, so documentation will be written once Terracotta has been updated to be compatible.

You can build the developer documentation using cargo doc. This will generate HTML files and place them into target/doc. You can then open the documentation in your browser by opening target/doc/terracotta/index.html.

Building the documentation for local development use will also provide you with links to the source code.

Deployment

Building

You can build the project in release mode by using cargo build --release. Everything required for deployment will be contained in the single binary file produced. It is recommended to run upx on the executable before deployment, to reduce the file size.

You can optionally supplement the compiled system with additional files from the local filesystem, as described in the Local loading options section above.

The resulting binary file can then be copied to the deployment environment, and run directly. This will often be in a Docker or Kubernetes container.

Examples

A typical build script might look like this:

sh cargo build --release upx --best target/release/terracotta scp target/release/terracotta you@yourserver:/path/to/deployment/directory

Docker

A common deployment scenario is to use Docker. The Terracotta repository includes a Dockerfile, which can be used to build a Docker image. This image is based on Alpine, and so is very small. It is also built using multi-stage builds, so the final image is even smaller.

It is worth noting that the Alpine build uses the musl C library, which is not compatible with the glibc C library used by most other Linux distributions and Docker images. The advantage of using Alpine is that the resulting image is very small, and everything is compiled statically. If you have any compatibility problems then you may want to use the distroless build instead, which is based on glibc.

The Docker image can be built using the following command:

sh docker build -t terracotta .

By default, this will build a release image, and compress the binary using upx. The setup is optimised for executable speed, build speed, and image size.

Profiles

You can specific the dev profile by passing the --build-arg profile=dev option to the docker build command. This will build an image that is not compressed, and is optimised for build speed but not image size.

Build arguments

Additionally, there are two other build arguments that can be passed in:

Running

It's worth noting that the host IP to serve on needs to be set to 0.0.0.0 to allow outside traffic to connect. In other words, the host entry in the Config.toml file should be set to "0.0.0.0" for a Docker setup:

toml host = "0.0.0.0" port = 8000

By default, Terracotta will run on port 8000, and this is expected by the Dockerfile. It is therefore advisable to keep this configured as such in the Config.toml file (or omitted), and instead use port mapping to map the container port to a host port. This can be achieved by specifying the -p option when calling the docker run command, for instance:

sh docker run -p 8000:8000 terracotta

This will make the Terracotta server available on port 8000 on the host machine, so that, on that machine, you will be able to visit it at http://localhost:8000 or http://127.0.0.1:8000 in your browser.

If you run Terracotta on a different port, you will need to specify that port in the Dockerfile.

Volumes

It is possible to mount volumes into the Docker container, to provide access to local files. This can be useful for development, and also for providing additional content and static assets. The following volumes are available:

These paths, and the options controlling them, can be overridden using the local loading options described above.

To mount a volume, use the -v option when calling the docker run command, for instance:

sh docker run -v /path/to/markdown:/usr/src/content:ro terracotta

It is advisable to specify the ro (read-only) option, as shown above, as there is no reason for Terracotta to need to write to the content files.

Examples

Default build, generating a compressed release image:

sh docker build -t terracotta .

Default build, generating an uncompressed release image:

sh docker build -t terracotta --build-arg upx=0 .

Dev build, generating an uncompressed dev image:

sh docker build -t terracotta --build-arg profile=dev .

Adjusting the opt-level for the release build:

sh docker build -t terracotta --build-arg cargo_opts="--config opt-level=z" .

Running the image:

sh docker run terracotta

Running the image and exposing the default port:

sh docker run -p 8000:8000 terracotta

Mounting volumes:

sh docker run \ -v /path/to/markdown:/usr/src/content:ro \ -v /path/to/templates:/usr/src/html:ro \ -v /path/to/assets:/usr/src/static:ro \ terracotta

Attributions

This project uses the Rust logo as a default, due to being written in Rust. The logo is freely usable under the CC-BY (Creative Commons Attribution) license.

An image of Ferris the crab (the Rust mascot) is used as an example of protected content. This image is sourced from rustacean.net and is in the Public Domain, so can be freely used.

This project uses the Bulma CSS framework, which is published under the MIT license and free to use without restriction.

The Font Awesome icons are published under the CC-BY (Creative Commons Attribution) license, and the webfonts under the SIL OFL (Open Font License). They are freely usable, along with the CSS code used to display them, which is released under the MIT license.