The easiest way to write Fastly Compute services in Zig.

• • What is Compute@Edge?
  • What is Zigly?
  • Usage
    • Adding Zigly as a dependency
    • A minimal WebAssembly program
    • Testing Compute@Edge modules
    • Using Zigly
      • Hello world!
      • Inspecting incoming requests
      • Making HTTP queries
      • Cache override
      • Pipes
    • Proxying
    • Redirects
    • Response decompression
      • Dictionaries
      • Logging
  • Deployment to Fastly's platform

Compute@Edge is Fastly's service to run custom code directly on CDN nodes.

The service runs anything that can be compiled to WebAssembly, and exports a convenient set of functions to interact with the platform.

Zigly is a library that makes it easy to write Compute@Edge modules in Zig.

Beyond the functions exported by the Fastly platform, Zigly will eventually include additional utility functions (cookie manipulation, JWT tokens, tracing...) to make application development as simple as possible.

Zigly is written for Zig 0.11.x.

Add the following to your build.zig.zon file:

.{
    .name = "project",
    .version = "0.0.1",
    .dependencies = .{
        .zigly = .{
            .url = "https://github.com/jedisct1/zigly/archive/refs/tags/0.1.3.tar.gz",
            .hash = "122025e93467c3e3855b323fcf6cbd5f569603f17eb72cd626354a7ed11609e900ab",
        },
    },
}

And the following to your build.zig file:

    const zigly = b.dependency("zigly", .{
        .target = target,
        .optimize = optimize,
    });
    exe.addModule("zigly", zigly.module("zigly"));
    exe.linkLibrary(zigly.artifact("zigly"));

The zigly structure can be imported in your application with:

const zigly = @import("zigly");

const std = @import("std");

pub fn main() !void
    std.debug.print("Hello from WebAssembly and Zig!\n", .{});
}

The program can be compiled with (replace example.zig with the source file name):

zig build-exe -target wasm32-wasi example.zig

Happy with the result? Add -Doptimize=ReleaseSmall or -Doptimize=ReleaseFast to get very small or very fast module:

zig build-exe -target wasm32-wasi -Doptimize=ReleaseSmall example.zig

The example above should not compile to more than 411 bytes.

If you are using a build file instead, define the target as wasm32-wasi in the build.zig file:

const target = try std.zig.CrossTarget.parse(.{ .arch_os_abi = "wasm32-wasi" });

...and build with zig build -Doptimize=ReleaseSmall or -Doptimize=ReleaseFast to get optimized modules.

The easiest way to test the resulting modules is to use Viceroy, a reimplementation of the Fastly API that runs locally.

var downstream = try zigly.downstream();
var response = downstream.response;
try response.body.writeAll("Hello world!");
try response.finish();

downstream() returns a type representing the initial connection, from a client to the proxy.

That type includes response, that can be used to send a response, as well as request, that can be used to inspect the incoming request.

Every function call may fail with an error from the FastlyError set.

Slightly more complicated example:

var downstream = try zigly.downstream();
var response = downstream.response;

response.setStatus(201);
response.headers.set("X-Example", "Header");

try response.body.writeAll("Partial");
try response.flush();
try response.body.writeAll("Response");
try response.finish();

var logger = Logger.open("logging_endpoint");
logger.write("Operation sucessful!");

Note that calling finish() is always required in order to actually send a response to the client.

But realistically, most responses will either be simple redirects:

var downstream = try zigly.downstream();
try downstream.redirect(302, "https://www.perdu.com");

or responding directly from the cache, proxying to the origin if the cached entry is nonexistent or expired:

var downstream = try zigly.downstream();
try downstream.proxy("google", "www.google.com");

Applications can read the body of an incoming requests as well as other informations such as the headers:

const request = downstream.request;
const user_agent = try request.headers.get(allocator, "user-agent");
if (request.isPost()) {
    // method is POST, read the body until the end, up to 1000000 bytes
    const body = try request.body.readAll(allocator, 1000000);
}

As usual in Zig, memory allocations are never hidden, and applications can choose the allocator they want to use for individual function calls.

Making HTTP queries is easy:

var query = try zigly.Request.new("GET", "https://example.com");
var response = try query.send("backend");
const body = try response.body.readAll(allocator, 0);

Arbitrary headers can be added the the outgoing query:

try query.headers.set("X-Custom-Header", "Custom value");

Body content can also be pushed, even as chunks:

try query.body.write("X");
try query.body.write("Y");
try query.body.close();

And the resulting response contains headers and body properties, that can be inspected the same way as a downstream query.

Caching can be disabled or configured on a per-query basis with setCachingPolicy():

try query.setCachingPolicy(.{ .serve_stale = 600, .pci = true });

Attributes include:

• no_cache
• ttl
• serve_stale
• pci
• surrogate_key

With pipe(), the response sent to a client can be a direct copy of another response. The application will then act as a proxy, optionally also copying the original status and headers.

var query = try zigly.Request.new("GET", "https://google.com");
var upstream_response = try query.send("google");
var downstream = try zigly.downstream();
try downstream.response.pipe(&upstream_response, true, true);

Proxying is even easier to use than pipes when a query should be sent unmodified (with the exception of the Host header) to the origin:

var downstream = try zigly.downstream();
try downstream.proxy("google", "www.google.com");

The second parameter is optional. If null, the original Host header will not be modified.

Redirecting the client to another address can be done with a single function call on the downstream object:

var downstream = try zigly.downstream();
try downstream.redirect(302, "https://www.perdu.com");

By default, responses are left as-is. Which means that if compression (Content-Encoding) was accepted by the client, the response can be compressed.

Calling setAutoDecompressResponse(true) on a Request object configures the Compute@Edge runtime to decompress gzip-encoded responses before streaming them to the application.

const dict = try zigly.Dictionary.open("name");
const value = try dict.get(allocator, "key");

const logger = try zigly.Logger.open("endpoint);
try logger.write("Log entry");

The fastly command-line tool only supports compilation of Rust and AssemblyScript at the moment. However, it can still be used to upload pre-compiled code written in other languages, including Zig.

1. Create a new project:

fastly compute init

For the language, select Other (pre-compiled WASM binary).

2. Add a build script:

Add the following lines to the fastly.toml file:

[scripts]
build = "zig build -Doptimize=ReleaseSmall -Dtarget=wasm32-wasi && mkdir -p bin && fastly compute pack --wasm-binary zig-out/bin/*"

3. Package the Compute@Edge module, passing in your compiled WebAssembly module.

fastly compute pack --path zig-out/bin/main.wasm

4. Test locally

fastly compute serve --skip-build --file zig-out/bin/main.wasm

5. Deploy!

fastly compute deploy

In order to deploy new versions, repeat steps 3 and 5.