Zap is intended to become the zig replacement for the kind of REST APIs I used to write in python with Flask and mongodb, etc. It can be considered to be a microframework for web applications.
What I needed for that was a blazingly fast, robust HTTP server that I could use with Zig. While facil.io supports TLS, I don't care about HTTPS support. In production, I use nginx as a reverse proxy anyway.
Zap wraps and patches facil.io - the C web application framework.
⚡ZAP⚡ IS SUPER ALPHA
Under the hood, everything is super robust and fast. My zig wrappers are fresh, juicy, and alpha.
After having used ZAP in production for weeks, I can confidently assert that it proved to be:
- ⚡ blazingly fast ⚡
- 💪 extremely robust 💪
Exactly the goals I set out to achieve!
Here's what works:
- Super easy build process: Zap's
build.zig
now uses the up-and-coming Zig package manager for its C-dependencies, no git submodules anymore.- tested on Linux and macOS (arm, M1)
- hello: welcomes you with some static HTML
- routes: a super easy example dispatching on the HTTP path
- serve: the traditional static web server with optional dynamic request handling
- sendfile: simple example of how to send a file, honoring compression headers, etc.
- bindataformpost: example to receive binary files via form post.
- hello_json: serves you json dependent on HTTP path
- endpoint: a simple JSON REST API example featuring a
/users
endpoint for performing PUT/DELETE/GET/POST operations and listing users, together with a simple frontend to play with. It also introduces a/stop
endpoint that shuts down Zap, so memory leak detection can be performed in main(). - mustache: a simple example using mustache templating.
- endpoint authentication: a simple authenticated endpoint. Read more about authentication here.
- http parameters: a simple example sending itself query parameters of all supported types.
- cookies: a simple example sending itself a cookie and responding with a session cookie.
- websockets: a simple websockets chat for the browser.
- Username/Password Session Authentication: A convenience authenticator that redirects un-authenticated requests to a login page and sends cookies containing session tokens based on username/password pairs received via POST request.
- MIDDLEWARE support: chain together request handlers in middleware style. Provide custom context structs, totally type-safe, using ZIG-CEPTION. If you come from GO this might appeal to you.
- MIDDLEWARE with endpoint
support:
Same as the example above, but this time we use an endpoint at the end of the
chain, by wrapping it via
zap.Middleware.EndpointHandler
. Mixing endpoints in your middleware chain allows for usage of Zap's authenticated endpoints and your custom endpoints. Since Endpoints use a simpler API, you have to user.setUserContext()
andr.getUserContext()
with the request if you want to access the middleware context from a wrapped endpoint. Since this mechanism uses an*anyopaque
pointer underneath (to not break the Endpoint API), it is less type-safe thanzap.Middleware
's use of contexts. - Per Request Contexts : With the introduction of
setUserContext()
andgetUserContext()
, you can, of course use those two in projects that don't usezap.SimpleEndpoint
orzap.Middleware
, too, if you really, really, absolutely don't find another way to solve your context problem. We recommend using azap.SimpleEndpoint
inside of a struct that can provide all the context you need instead. You get access to your struct in the callbacks via the@fieldParentPtr()
trick that is used extensively in Zap's examples, like the endpoint example. - Error Trace Responses: You can now
call
r.sendError(err, status_code)
when you catch an error and a stack trace will be returned to the client / browser.
I'll continue wrapping more of facil.io's functionality and adding stuff to zap to a point where I can use it as the JSON REST API backend for real research projects, serving thousands of concurrent clients.
Claiming to be blazingly fast is the new black. At least, Zap doesn't slow you down and if your server performs poorly, it's probably not exactly Zap's fault. Zap relies on the facil.io framework and so it can't really claim any performance fame for itself. In this initial implementation of Zap, I didn't care about optimizations at all.
But, how fast is it? Being blazingly fast is relative. When compared with a simple GO HTTP server, a simple Zig Zap HTTP server performed really good on my machine (x86_64-linux):
- Zig Zap was nearly 30% faster than GO
- Zig Zap had over 50% more throughput than GO
Update: Thanks to @felipetrz, I got to test against more realistic Python
and Rust examples. Both python sanic
and rust axum
were easy enough to
integrate.
Update: I have automated the benchmarks. See blazingly-fast.md for more information. Also, thanks to @alexpyattaev, the benchmarks are fairer now, pinning server and client to specific CPU cores.
Update: I have consolidated the benchmarks to one good representative per language. See more details in blazingly-fast.md. It contains rust implementations that come pretty close to Zap's performance in the simplistic testing scenario.
So, being somewhere in the ballpark of basic GO performance, zig zap seems to be ... of reasonable performance 😎.
I can rest my case that developing ZAP was a good idea because it's faster than both alternatives: a) staying with Python, and b) creating a GO + Zig hybrid.
See more details in blazingly-fast.md.
ZAP is very robust. In fact, it is so robust that I was confidently able to only work with in-memory data (RAM) in all my initial ZAP projects so far: 3 online research experiments. No database, no file persistence, until I hit "save" at the end 😊.
So I was able to postpone my cunning data persistence strategy that's similar to a mark-and-sweep garbage collector and would only persist "dirty" data when traffic is low, in favor of getting stuff online more quickly. But even if implemented, such a persistence strategy is risky because when traffic is not low, it means the system is under (heavy) load. Would you confidently NOT save data when load is high and the data changes most frequently -> the potential data loss is maximized?
To answer that question, I just skipped it. I skipped saving data until receiving a "save" signal. And it worked. ZAP kept on zapping. When traffic calmed down or all experiment participants had finished, I hit "save" and went on analyzing the data.
Handling all errors does pay off after all. No hidden control flow, no hidden errors or exceptions is one of Zig's strengths.
To be honest: There are still pitfalls. E.g. if you request large stack sizes for worker threads, Zig won't like that and panic. So make sure you don't have local variables that require tens of megabytes of stack space.
See the StopEndpoint in the
endpoint example. That example uses ZIG's awesome
GeneralPurposeAllocator
to report memory leaks when ZAP is shut down. The
StopEndpoint
just stops ZAP when receiving a request on the /stop
route.
You can use the same strategy in your debug builds and tests to check if your code leaks memory.
Make sure you have the latest zig release (0.11.0) installed. Fetch it from here.
$ git clone https://github.com/zigzap/zap.git
$ cd zap
$ zig build run-hello
$ # open http://localhost:3000 in your browser
... and open http://localhost:3000 in your browser.
Make sure you have the latest zig release (0.11.0) installed. Fetch it from here.
If you don't have an existing zig project, create one like this:
$ mkdir zaptest && cd zaptest
$ zig init-exe
$ git init ## (optional)
Note: Nix/NixOS users are lucky; you can use the existing flake.nix
and run
nix develop
to get a development shell providing zig and all
dependencies to build and run the GO, python, and rust examples for the
wrk
performance tests. For the mere building of zap projects,
nix develop .#build
will only fetch zig 0.11.0.
With an existing Zig project, adding Zap to it is easy:
- Add zap to your
build.zig.zon
- Add zap to your
build.zig
To add zap to build.zig.zon
:
.{
.name = "My example project",
.version = "0.0.1",
.dependencies = .{
// zap v0.1.14-pre
.zap = .{
.url = "https://github.com/zigzap/zap/archive/refs/tags/v0.1.14-pre.tar.gz",
.hash = "122067d12bc7abb93f7ce623f61b94cadfdb180cef12da6559d092e6b374202acda3",
}
}
}
Then, in your build.zig
's build
function, add the following before
`b.installArtifact(exe)``:
const zap = b.dependency("zap", .{
.target = target,
.optimize = optimize,
});
exe.addModule("zap", zap.module("zap"));
exe.linkLibrary(zap.artifact("facil.io"));
From then on, you can use the Zap package in your project. Check out the examples to see how to use Zap.
You can change the URL to Zap in your build.zig.zon
- easiest: use a tagged release
- or to one of the tagged versions, e.g.
0.0.9
- or to the latest commit of
zap
Go to the release page. Every release
will state its version number and also provide instructions for changing
build.zig.zon
and build.zig
.
See here.
At the current time, I can only add to zap what I need for my personal and professional projects. While this happens blazingly fast, some if not all nice-to-have additions will have to wait. You are very welcome to help make the world a blazingly fast place by providing patches or pull requests, add documentation or examples, or interesting issues and bug reports - you'll know what to do when you receive your calling 👼.
Check out CONTRIBUTING.md for more details.
See also introducing.md for more on the state and progress of this project.
We now have our own ZAP discord server!!!
You can also reach me on the zig showtime discord server under the handle renerocksai (renerocksai#1894).
Being blazingly fast requires a constant feed of caffeine. I usually manage to provide that to myself for myself. However, to support keeping the juices flowing and putting a smile on my face and that warm and cozy feeling into my heart, you can always buy me a coffee ☕. All donations are welcomed 🙏 blazingly fast! That being said, just saying "hi" also works wonders with the smiles, warmth, and coziness 😊.
You build and run the examples via:
$ zig build [EXAMPLE]
$ ./zig-out/bin/[EXAMPLE]
... where [EXAMPLE]
is one of hello
, routes
, or serve
.
Example: building and running the hello example:
$ zig build hello
$ ./zig-out/bin/hello
To just run an example, like routes
, without generating an executable, run:
$ zig build run-[EXAMPLE]
Example: building and running the routes example:
$ zig build run-routes
const std = @import("std");
const zap = @import("zap");
fn on_request(r: zap.SimpleRequest) void {
if (r.path) |the_path| {
std.debug.print("PATH: {s}\n", .{the_path});
}
if (r.query) |the_query| {
std.debug.print("QUERY: {s}\n", .{the_query});
}
r.sendBody("<html><body><h1>Hello from ZAP!!!</h1></body></html>") catch return;
}
pub fn main() !void {
var listener = zap.SimpleHttpListener.init(.{
.port = 3000,
.on_request = on_request,
.log = true,
});
try listener.listen();
std.debug.print("Listening on 0.0.0.0:3000\n", .{});
// start worker threads
zap.start(.{
.threads = 2,
.workers = 2,
});
}