The Patchify crate is an auto-update library, providing the ability for Rust applications to automatically update themselves.
It includes functionality to embed into both the application itself and into a web-based API server, to give the application the ability to check for updates, and for the server to fully handle those updates. The library modules are extremely easy to use, with simple configuration and minimal code required to get up and running.
The application to equip with auto-update powers can potentially be anything that runs persistently: a web server, a desktop application, a command-line tool, etc.
There is a roadmap for the project, which sets out the planned releases and their associated functionality, and indicates current status according to the intended goals.
There is full test coverage, with unit tests, integration tests, and end-to-end tests, utilising various types of mock. There are also helpful examples.
The main sections in this README are:
The main high-level points of note are:
- Client application
- Fully-autonomous update checking and upgrade process
- Configurable checking intervals
- Automatic application restart
- Ability to register and manage critical actions to choreograph upgrades
- Update status broadcaster for application-wide status updates
- Verification of release files using SHA256 hashes
- Verification of HTTP response signatures using public keys
- API server
- Webserver-agnostic, but with full integration for Axum
- Logging of HTTP requests and events using Tokio Tracing
- Streaming of large release files for memory efficiency
- Signing of HTTP responses using private keys
- Full yet minimal examples working out of the box
- Configuration from config files and env vars using Figment
- High-performance asynchronous HTTP server using Tokio Hyper
- Based on the robust and ergonomic web framework Axum
- Fully-documented codebase
- Full test coverage with unit and integration tests
HTTP responses will be signed with the server's private key, allowing connecting clients to verify that they have not been tampered with. The key format used is Ed25519, which is faster and more secure than RSA.
Release files are checked against SHA256 hashes when the server starts up, and the client verifies the hashes of the files it downloads. This ensures that the files are an accurate replica of the original, and have not been tampered with.
The server will stream large release files to clients, which is more efficient than reading the entire file into memory before sending it. This is fully-configurable.
Currently, the following modules are provided:
The modules are designed to be used independently.
The client
module provides functionality to embed into an application, to give it
auto-update abilities.
Use of the client module is essentially as follows:
- Bring in the
patchify::client
module. - Create a new
Updater
instance, passing in the appropriate configuration via aConfig
instance.
That's it! The Updater
instance will spawn threads to check for updates in the
background, at the specified intervals. It will also log its activity, and you
can listen in to that activity from inside your application using the provided
Updater.subscribe()
method.
An example of this in use is provided as examples/cli-app-*.rs
. There are two
versions, so that the upgrade capabilities can be seen in action.
To run the examples, use the following commands:
cargo run --example cli-app-v1
cargo run --example cli-app-v2
The following points are worth noting:
- The
tokio::signal::ctrl_c()
function is used to wait for aCtrl-C
signal, to keep the application running so that the update process can be observed. - The example allows configuration of the application name, the location of the API server, the API server's public key, and the update interval.
This is covered in more depth in the end-to-end example guide.
The server
module provides functionality to embed into a web-based API server, to give it
the ability to serve updates and related information to applications. As well as
the critical functionality, it also provides endpoint implementations
suitable for use with Axum.
A fairly simple, yet comprehensive example of how to use the server module is
provided as examples/axum-server.rs
. This example is essentially an expanded
version of the testbins/standard-api-server
integration test, and relies upon
the same common code used by all the integration tests, but adds the ability to
accept configuration. It is a good starting point for a real-world server
implementation.
To run the example, use the following command:
cargo run --example axum-server
The following points are worth noting:
- The full implementation logic can be found in
tests/common/server.rs
, which is used by the integration tests. This provides theinitialize()
,create_patchify_api_server()
, andpatchify_api_routes()
functions. - The
tokio::signal::ctrl_c()
function is used to wait for aCtrl-C
signal, which will help the server to shut down cleanly. This is not strictly required, but is a common pattern, and good practice. It is used in the integration tests to ensure that the test server continues running in the background while it is still needed, and can then be told to shut down when the tests are complete. - The example allows configuration of the application name, the host and port
to run on, the folder containing the release files, and a list of versions
with their associated hashes. In order to use a randomly-assigned port, set
the value to
0
.
By default there will be no versions configured for the example. To add some, release files should be created in the configured releases folder, named appropriately, and the hashes should be calculated and added to the list. The example will then serve these files and hashes to clients that request them. This is covered in more depth in the end-to-end example.
The steps to set up a project using Patchify 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.
There are some key points to note about the environment you choose:
- Debian and Ubuntu are the Linux distros of choice, although other distros should also work just fine, as there are no special requirements.
- Running natively on Windows is not currently targeted or tested, but there are plans to support it properly in future. Running on WSL does work fine.
- Running natively on MacOS is untested, although there is no known technical reason why it would not work.
Patchify is configured using Config
structs that are passed in to the
Updater
instance on the client side, and to the Core
instance on the server
side. These are documented here:
For the client, the creation of an Updater
instance with a suitable Config
is all that is required. For the server, in addition to creating a Core
instance, there is also a choice about whether to integrate with Axum or
call the Core
methods directly from your own endpoint functions.
Notably, the client functionality provides two main touchpoints to help
orchestrate the upgrade process: the Updater.subscribe()
method, and the
critical actions counter.
The critical actions counter is a simple counter that can be incremented and decremented, and is used to establish when it's safe to restart the application after an upgrade. It is a simple way to ensure that the application is not in the middle of a critical operation when it is restarted.
The basic premise is that if an application is about to do something that
should not be interrupted, the counter can be incremented, and then decremented
when the operation is complete. The Updater
instance will then only restart
the application when the counter is zero. If the updater is about to restart the
application, then it will deny the start of any new critical actions until the
restart has completed.
This makes it very easy to integrate the upgrade process into an application with confidence around when exactly a restart will occur.
If more control is needed over the manner in which the restart occurs, then it is advisable to register a critical action when the application starts, and never deregister it, instead relying upon the status change events to detect when a restart is needed, and handle it in a customised way.
The Updater.subscribe()
method allows the application to listen in to the
activity of the Updater
instance, and to react to the status changes that it
broadcasts. This is useful for updating the application's UI, or for logging
purposes. It also provides a way of manually controlling the upgrade process, if
necessary.
It's not possible to have a 100%-working end-to-end example available right out of the box, because the release files need to be generated and registered with the server. However, all of the ingredients are provided, and by following the few short steps in this section you can have a fully-working example up and running in no time.
You will need to have a Rust environment set up, on Linux, along with a clone of the Patchify repository. The steps in this example assume that commands are being run from the root of the repository. This is for demonstration purposes only, and in a real-world scenario you would be working with your own application repository and including Patchify as a dependency, using it as described in the rest of this README.
Assuming a fresh clone, create a new directory for the server to serve releases from:
mkdir -p /tmp/patchify-releases
We now need to compile the client examples. Although you can run them directly, they also get compiled when the tests are run, so that's the simplest way to build them, and also ensures that all the tests are passing in your environment.
cargo test
Once the tests have completed, you need to copy the compiled binaries to the releases directory:
cp target/debug/examples/cli-app-v1 /tmp/patchify-releases/cli-app-1.0.0
cp target/debug/examples/cli-app-v2 /tmp/patchify-releases/cli-app-2.0.0
Note the change in filename. This is because Cargo wants each example to have a
different crate, and what we want to do here is actually create two different
versions of the same application. Therefore the examples are named v1
and
v2
, corresponding to versions 1.0.0
and 2.0.0
respectively.
In order to be able to run the client application later, copy the application binary to your current directory:
cp target/debug/examples/cli-app-v1 ./cli-app
Note again the change in filename. This is because we don't care what version the application is, we just want to run it. The file we copied will be replaced with the new version when the updater runs.
Note that if your local Cargo is set up to use a different directory for builds, you will need to adjust the paths accordingly.
Now we need to run the server example. This will serve the release files that we just created, and will also provide the API endpoints for the client application to interact with.
But first, we need some configuration. Copy the examples/axum-server.toml
file
to your current working directory:
cp examples/axum-server.toml .
Now edit it and set the releases
value to /tmp/patchify-releases
, and add
the two versions in, with their associated hashes. The file should look
something like this:
appname = "cli-app"
host = "127.0.0.1"
port = 8000
releases = "/tmp/patchify-releases"
[versions]
"1.0.0" = "beef1a2b3c4d5e6f7a8b9c0d1e2f3a4b5c6d7e8f9a0b1c2d3e4f5a6b7c8d9e0f"
"2.0.0" = "cafe1a2b3c4d5e6f7a8b9c0d1e2f3a4b5c6d7e8f9a0b1c2d3e4f5a6b7c8d9e0f"
In order to obtain the hashes, you can use the sha256sum
command:
sha256sum /tmp/patchify-releases/cli-app-1.0.0
sha256sum /tmp/patchify-releases/cli-app-2.0.0
By default the server will run on port 8000
, and will serve from localhost.
Feel free to change these values to suit your environment.
Now you can run the server example. Note that you will need to keep this open in a different terminal window, so from this point on you will have two terminal windows open.
cargo run --example axum-server
The server will check the validity of the release files when it starts up, and exit if there are any errors. Note that this may take a few seconds, as it needs to read the entirety of each file to calculate the hash.
Copy the examples/cli-app.toml
file to your current working directory:
cp examples/cli-app.toml .
Now edit it and set the updater_api_server
value to whatever the server is
running on. This should be as you configured, but the server will print out the
address when it starts up, along with the public key. You will need to add that
key to the client configuration as well, as updater_api_key
:
appname = "cli-app"
updater_api_server = "http://127.0.0.1:8000/api/"
updater_api_key = "beef1a2b3c4d5e6f7a8b9c0d1e2f3a4b5c6d7e8f9a0b1c2d3e4f5a6b7c8d9e0f"
update_on_startup = false
update_interval = 10
The example server generates a new key each time it starts up, so you will need to copy it from the server's output each time you restart the server. This is to make the process more robust for demonstration purposes than accepting a private key in the server configuration, in case it is wrongly generated, which can cause frustration.
Note that in the configuration example above, the update_on_startup
value is
set to false
. This is because we want to see the update process in action, so
we don't want the application to update itself when it starts up. Additionally,
the update_interval
is set to 10
, so that the application will check for
updates every 10 seconds, which is not too long to wait. Feel free to experiment
with these values and observe the differences in behaviour.
We're now ready to run the client application!
./cli-app
The application will start up and check for updates at the interval you specified in the configuration. If there are updates available, it will download them, verify them, install them, and restart itself. You should see the status change events being printed to the console, and you should also see the printed version number change when the application restarts.