devd: a web daemon for developers

2015-10-23

I've just released devd, a small, self-contained, command-line-only HTTP server for developers. It started as a weekend stress-relief hack (that's a thing where I'm from), but has now become my preferred "daily driver" for most web-ish things. It's simple, direct and does more or less exactly what I need. This isn't terribly surprising, since I wrote it to scratch my own idiosyncratic itch - hopefully other, similarly itchy hackers will find it useful too.

Quick start

Serve the current directory, open it in the browser (-o), and livereload when files change (-l):

devd -ol .

Reverse proxy to http://localhost:8080, and livereload when any file in the src directory changes:

devd -w ./src http://localhost:8080

Features

Cross-platform and self-contained

Devd is a single statically compiled binary with no external dependencies, and is released for OSX, Linux and Windows. Don't want to install Node or Python in that light-weight Docker instance you're hacking in? Just copy over the devd binary and be done with it.

Designed for the terminal

This means no config file, no daemonization, and logs that are designed to be read in the terminal by a developer. Logs are colorized and log entries span multiple lines. Devd's logs are detailed, warn about corner cases that other daemons ignore, and can optionally include things like detailed timing information and full headers.

To make quickly firing up an instance as simple as possible, devd automatically chooses an open port to run on (unless it's specified), and can open a browser window pointing to the daemon root for you (the -o flag in the example above).

Livereload

When livereload is enabled, devd injects a small script into HTML pages, just before the closing head tag. The script listens for change notifications over a websocket connection, and reloads resources as needed. No browser addon is required, and livereload works even for reverse proxied apps. If only changes to CSS files are seen, devd will only reload external CSS resources, otherwise a full page reload is done. This serves the current directory with livereload enabled:

devd -l .

You can also trigger livereload for files that are not being served, letting you reload reverse proxied applications when source files change. So, this command watches the src directory tree, and reverse proxies to a locally running application:

devd -w ./src http://localhost:8888

Reverse proxy + static file server + flexible routing

Modern apps tend to be collections of web servers, and devd caters for this with flexible reverse proxying. You can use devd to overlay a set of services on a single domain, add livereload to services that don't natively support it, add throttling and latency simulation to existing services, and so forth.

Here's a more complicated example showing how all this ties together - it overlays two applications and a tree of static files. Livereload is enabled for the static files (-l) and also triggered whenever source files for reverse proxied apps change:

devd -l \
-w ./src/ \
/=http://localhost:8888 \
/api/=http://localhost:8889 \
/static/=./assets

Light-weight virtual hosting

Devd uses a dedicated domain - devd.io - to do simple virtual hosting. This domain and all its subdomains resolves to 127.0.0.1, which we use to set up virtual hosting without any changes to /etc/hosts or other local configuration. Route specifications that don't start with a leading / are taken to be subdomains of devd.io. So, the following command serves a static site from devd.io, and reverse proxies a locally running app on api.devd.io:

devd ./static api=http://localhost:8888

Check out the docs at the Github repo for the full route specification syntax.

Latency and bandwidth simulation

Want to know what it's like to use your fancy 5mb HTML5 app from a mobile phone in Botswana? Look up the bandwidth and latency here, and invoke devd like so (making sure to convert from kilobits per second to kilobytes per second):

devd -d 114 -u 51 -l 75 .

Devd tries to be reasonably accurate in simulating bandwidth and latency - it uses a token bucket implementation for throttling, properly handles concurrent requests, and chunks traffic up so data flow is smooth.