An HTTP server for Node designed to serve PRPL apps in production.
⚠️ IMPORTANT⚠️ prpl-serveris in maintenance mode, and is no longer recommended. Reports and PRs for critical bugs and security issues will be accepted, but we will no longer accept new feature requests or PRs.For differential serving, we now recommend a simple two-build configuration using
nomodulefor client-side capability sniffing. See https://jasonformat.com/modern-script-loading/ for more details on this pattern.If you are looking for a modern development sever, we recommend @web/dev-server.
In addition, note that Chromium is considering removing support for Server Push in a future version. Consider using
<link rel="preload">tags as a simpler alternative with similar performance benefits.
- Usage
- Differential Serving
- Entrypoint
- Base paths
- HTTP/2 Server Push
- Service Workers
- HTTPS
- Caching
- HTTP Errors
- Rendering for Bots
- Google App Engine Quickstart
$ npm install -g prpl-server
$ prpl-server --root . --config polymer.json$ npm install --save prpl-serverprpl = require('prpl-server');
express = require('express');
const app = express();
app.get('/api/launch', (req, res, next) => res.send('boom'));
app.get('/*', prpl.makeHandler('.', {
builds: [
{name: 'modern', browserCapabilities: ['es2015', 'push']},
{name: 'fallback'},
],
}));
app.listen(8080);Modern browsers offer great features that improve performance, but most applications need to support older browsers too. prpl-server can serve different versions of your application to different browsers by detecting browser capabilities using the user-agent header.
prpl-server understands the notion of a build, a variant of your application optimized for a particular set of browser capabilities.
Builds are specified in a JSON configuration file. This format is compatible with polymer.json, so if you are already using polymer-cli for your build pipeline, you can annotate your existing builds with browser capabilities, and copy the configuration to your server root. prpl-server will look for a file called polymer.json in the server root, or you can specify it directly with the --config flag.
In this example we define two builds, one for modern browsers that support ES2015 and HTTP/2 Push, and a fallback build for other browsers:
{
"entrypoint": "index.html",
"builds": [
{"name": "modern", "browserCapabilities": ["es2015", "push"]},
{"name": "fallback"}
]
}
The browserCapabilities field defines the browser features required for that build. prpl-server analyzes the request user-agent header and picks the best build for which all capabilities are met. If multiple builds are compatible, the one with more capabilities is preferred. If there is a tie, the build that comes earlier in the configuration file wins.
You should always include a fallback build with no capability requirements. If you don't, prpl-server will warn at startup, and will return a 500 error on entrypoint requests to browsers for which no build can be served.
The following keywords are supported. See also the browser-capabilities library which prpl-server uses.
| Keyword | Description |
|---|---|
| push | HTTP/2 Server Push |
| serviceworker | Service Worker API |
| modules | JavaScript Modules (including dynamic import() and import.meta) |
| es2015 | ECMAScript 2015 (aka ES6) |
| es2016 | ECMAScript 2016 |
| es2017 | ECMAScript 2017 |
| es2018 | ECMAScript 2018 |
In the PRPL pattern, the entrypoint is a small HTML file that acts as the application bootstrap.
prpl-server will serve the entrypoint from the best compatible build from /, and from any path that does not have a file extension and is not an existing file.
prpl-server expects that each build subdirectory contains its own entrypoint file. By default it is index.html, or you can specify another name with the entrypoint configuration file setting.
Note that because the entrypoint is served from many URLs, and varies by user-agent, cache hits for the entrypoint will be minimal, so it should be kept as small as possible.
Since prpl-server serves resources from build subdirectories, your application source can't know the absolute URLs of build-specific resources upfront.
For most documents in your application, the solution is to use relative URLs to refer to other resources in the build, and absolute URLs to refer to resources outside of the build (e.g. static assets, APIs). However, since the entrypoint is served from URLs that do not match its location in the build tree, relative URLs will not resolve correctly.
The solution we recommend is to place a <base> tag in your entrypoint to anchor its relative URLs to the correct build subdirectory, regardless of the URL the entrypoint was served from. You may then use relative URLs to refer to build-specific resources from your entrypoint, as though you were in your build subdirectory. Put <base href="/"> in your source entrypoint, so that URLs resolve when serving your source directly during development. In your build pipeline, update each entrypoint's base tag to match its build subdirectory (e.g. <base href="/modern/">).
If you are using polymer-cli, set {"autoBasePath": true} in your polymer.json to perform this base tag update automatically.
Note that <base> tags only affect relative URLs, so to refer to resources outside of the build from your entrypoint, use absolute URLs as you normally would.
Server Push allows an HTTP/2 server to preemptively send additional resources alongside a response. This can improve latency by eliminating subsequent round-trips for dependencies such as scripts, CSS, and HTML imports.
prpl-server looks for a file called push-manifest.json in each build subdirectory, and uses it to map incoming request paths to the additional resources that should be pushed with it. The original push manifest file format is described here. Tools for generating a push manifest include http2-push-manifest and polymer-cli.
Each key in the push manifest is a regular expression pattern that will be matched against the incoming request path. Patterns are forced to match exactly (e.g. foo.html is equivalent to ^foo.html$). You can use wildcard patterns to push resources for client-side application routes (e.g. /articles/.*). In the case of the entrypoint, the resolved filename (e.g. index.html) is used as a key to the push manifest, in addition to the application route.
Resources in the push manifest can be specified as absolute or relative paths. Absolute paths are interpreted relative to the server root directory. Relative paths are interpreted relative to the location of the push manifest file itself (i.e. the build subdirectory), so that they do not need to know which build subdirectory they are being served from. Push manifests generated by polymer-cli always use relative paths.
prpl-server is designed to be used behind an HTTP/2 reverse proxy, and currently does not generate push responses itself. Instead it sets preload link headers, which are intercepted by cooperating reverse proxy servers and upgraded into push responses. Servers that implement this upgrading behavior include Apache, nghttpx, and Google App Engine.
If a build with a push manifest is served to a browser that does not support push according to the browser-capabilities support matrix, then a nopush attribute is added to the generated preload link headers.
To confirm your push manifest is working during local development, you can look for Link: <URL>; rel=preload response headers in your browser dev tools.
To see genuine push locally, you will need to run a local HTTP/2 reverse proxy such as nghttpx:
- Install nghttpx (Homebrew, Ubuntu, source).
- Generate a self-signed TLS certificate, e.g.
openssl req -newkey rsa:2048 -x509 -nodes -keyout server.key -out server.crt - Start prpl-server (assuming default
127.0.0.1:8080). - Start nghttpx:
nghttpx -f127.0.0.1,8443 -b127.0.0.1,8080 server.key server.crt --no-ocsp - Visit
https://localhost:8443. In Chrome, Push responses will show up in the Network tab as Initiator: Push / Other.
Note that Chrome will not allow a service worker to be registered over HTTPS with a self-signed certificate. You can enable chrome://flags/#allow-insecure-localhost to bypass this check. See this page for more tips on developing service workers in Chrome.
prpl-server sets the Service-Worker-Allowed header to / for any request path ending with service-worker.js. This allows a service worker served from a build subdirectory to be registered with a scope outside of that directory, e.g. register('service-worker.js', {scope: '/'}).
prpl-server automatically serves a tiny self-unregistering service worker for any request path ending with service-worker.js that would otherwise have had a 404 Not Found response. To disable this behavior, set unregisterMissingServiceWorkers: false in your configuration file.
This can be useful when the location of a service worker has changed, as it will prevent clients from getting stuck with an old service worker indefinitely.
This problem arises because when a service worker updates, a 404 is treated as a failed update. It does not cause the service worker to be unregistered. See w3c/ServiceWorker#204 for more discussion of this problem.
Your apps should always be served over HTTPS. It protects your user's data, and is required for features like service workers and HTTP/2.
If the --https-redirect flag is set, prpl-server will redirect all HTTP requests to HTTPS. It sends a 301 Moved Permanently redirect to an https:// address with the same hostname on the default HTTPS port (443).
prpl-server trusts X-Forwarded-Proto and X-Forwarded-Host headers from your reverse proxy to determine the client's true protocol and hostname. Most reverse proxies automatically set these headers, but if you encounter issues with redirect loops, missing or incorrect X-Forwarded-* headers may be the cause.
You should always use --https-redirect in production, unless your reverse proxy already performs HTTPS redirection.
By default, prpl-server sets the Cache-Control header to max-age=60 (1 minute), except for the entrypoint and service worker which gets max-age=0. ETag headers are also sent, so resources that have not changed on the server can be re-validated efficiently.
To change this default for non-entrypoint resources, set the cacheControl property in your configuration file, or the --cache-control command-line flag, to the desired Cache-Control header value. You may want to set --cache-control=no-cache during development.
For more advanced caching behavior, use prpl-server as a library with Express and register a middleware that sets the Cache-Control header before registering the prpl-server middleware. If prpl-server sees that the Cache-Control header has already been set, it will not modify it. For example, to set year-long caching for images:
app.get('/images/*', (req, res, next) => {
res.setHeader('Cache-Control', 'public, max-age=31536000');
next();
});
app.get('/*', prpl.makeHandler('.', config))Choosing the right cache headers for your application can be complex. See Caching best practices & max-age gotchas for one starting point.
By default, if a 404 Not Found or other HTTP server error occurs, prpl-server will serve a minimal text/plain response. To serve custom errors, use prpl-server as a library with Express, set forwardErrors: true in your configuration object, and register an error-handling middleware after registering the prpl-server handler:
app.get('/*', prpl.makeHandler('.', {
builds: [ ... ],
forwardErrors: true
}));
app.use((err, req, res, next) => {
if (err.status === 404) {
res.status(404).sendFile('my-custom-404.html', {root: rootDir});
} else {
next();
}
});Many bots don't execute JavaScript when processing your application. This can cause your application to not render correctly when crawled by some search engines, social networks, and link rendering bots.
One solution to this problem is Rendertron. Rendertron is a server which runs headless Chrome to render and serialize web pages for these bots, so all the content is contained in one network request. Use the --bot-proxy flag to instruct prpl-server to proxy requests from a known list of bots through a Rendertron server.
Note that you can also use the Rendertron middleware directly if you have a custom Express server.
Google App Engine is a managed server platform that supports Node.js. You can deploy prpl-server to App Engine standard environment with a few steps:
-
Follow these instructions to set up a Google Cloud project and install the Google Cloud SDK. As instructed, run the
gcloud initcommand to authenticate and choose your project ID. -
cdto the directory you want to serve (e.g. your app'sbuild/directory if you are using polymer-cli). -
Run
npm initand follow the prompts to create yourpackage.json. -
Run
npm install --save prpl-serverto add prpl-server as a dependency. -
Edit your
package.jsonto add astartscript. This is the command App Engine runs when your app starts. Configureprpl-serverto listen on all hosts, and to redirect HTTP connections to HTTPS. You should also specify the version of Node your app requires via theenginessection.
{
"scripts": {
"start": "prpl-server --host 0.0.0.0 --https-redirect"
},
"engines": {
"node": "8.x.x"
}
}- Create an
app.yamlfile. This tells App Engine that you want to use Node.js in the App Engine standard environment:
runtime: nodejs8- Run
gcloud app deployto deploy to your App Engine project.gcloudwill tell you the URL your app is being served from. For next steps, check out the Node.js in the App Engine standard environment documentation.