image-resizer is a Node.js application that sits as a custom origin to your CDN and will resize/optimise images on-the-fly. It is Heroku ready, but can also be deployed easily to any cloud provider (has been used with success on AWS).
The primary goal for this project was to abstract the need to set image dimensions during the upload and storage phase of images in a modern web application.
Building and deploying your own version of image-resizer is as easy as running the cli tool (image-resizer new), setting your Heroku configs and firing it up!
Based on Express.js image-resizer uses sharp under the hood to modify and optimise your images.
There is also a plugin architecture that allows you to add your own image sources. Out of the box it supports: S3, Facebook, Twitter, Youtube, Vimeo (and local file system in development mode).
When a new image size is requested of image-resizer via the CDN, it will pull down the original image from the cloud. It will then resize according to the requested dimensions, optimize according to file type and optionally filter the image. All responses are crafted with custom responses to maximise the facility of the CDN.
$ npm install -g tripviss/image-resizer gulp
$ mkdir my_fancy_image_server
$ cd my_fancy_image_server
$ image-resizer new
$ npm install
$ gulp watch
This will create a new directory structure including all the necessary files needed to run image-resizer. The money file is server.js which loads the express configuration and routes.
image-resizer can also simply be added as a node_module to any project and the streams interfaces used standalone. ./test.js has a good example of how the app should work running behind Express.
There is a RubyGem of helpers (both Ruby and Javascript) to assist you in building the endpoints for your image-resizer instance.
The codebase takes advantage of node streams.
Inspired a lot by Gulp, image-resizer passes around an Image object between each of the streams that contains information about the request and the image data (either as a buffer or stream).
Modified images are not sent back to s3 for storage. Instead, use a CDN to cache the modified images. This greatly improves performance both on the server side and client side. Google PageSpeed does not like 302 redirects.
Removing the need to push data to s3 also helps server response times, as the network performance can be wildly inconsistent.
image-resizer supports a range of custom plugins for both image sources and filters. As mentioned above a number of sources are supported out of the box but each of these can be over written as needed.
The directory structure created via $ image-resizer new will include a plugins directory where the initialization script will pick up any scripts and insert them into the application.
image-resizer only requires a working node/npm environment and libvips. The necessary buildpack information to load your Heroku environment is included.
Configuration of image-resizer is done via environment variables. This is done to be compatible with Heroku deployments.
To set environment variables in your Heroku console.
heroku config:set AWS_ACCESS_KEY_ID=abcd1234
For Heroku deployment the minimum required variables are:
AWS_ACCESS_KEY_ID
AWS_SECRET_ACCESS_KEY
AWS_REGION
S3_BUCKET
NODE_ENV
If you choose to change your default source to be something other than S3 then the NODE_ENV variable is the only required one (and whatever you need for your default source).
For convenience in local and non-Heroku deployments the variables can be loaded from a .env file. Sensible local defaults are included in src/config/environment_vars.js.
The available variables are as follows:
NODE_ENV: 'development',
PORT: 3001,
DEFAULT_SOURCE: 's3',
EXCLUDE_SOURCES: null, // add comma delimited list
// Restrict to named modifiers strings only
NAMED_MODIFIERS_ONLY: false,
// AWS keys
AWS_ACCESS_KEY_ID: null,
AWS_SECRET_ACCESS_KEY: null,
AWS_REGION: null,
S3_BUCKET: null,
// Resize options
RESIZE_PROCESS_ORIGINAL: true,
AUTO_ORIENT: true,
REMOVE_METADATA: true,
// Protect original files by specifying a max image width or height - limits
// max height/width in parameters
MAX_IMAGE_DIMENSION: null,
// Color used when padding an image with the 'pad' crop modifier.
IMAGE_PADDING_COLOR: 'white',
// Optimization options
IMAGE_QUALITY: 80,
IMAGE_PROGRESSIVE: true,
// Cache expiries
IMAGE_EXPIRY: 60 * 60 * 24 * 90,
IMAGE_EXPIRY_SHORT: 60 * 60 * 24 * 2,
JSON_EXPIRY: 60 * 60 * 24 * 30,
// Logging options
LOG_PREFIX: 'resizer',
QUEUE_LOG: true,
// Response settings
CACHE_DEV_REQUESTS: false,
// Twitter settings
TWITTER_CONSUMER_KEY: null,
TWITTER_CONSUMER_SECRET: null,
TWITTER_ACCESS_TOKEN: null,
TWITTER_ACCESS_TOKEN_SECRET: null,
// Where are the local files kept?
LOCAL_FILE_PATH: process.cwd(),
// Display an image if a 404 request is encountered from a source
IMAGE_404: null
// Whitelist arbitrary HTTP source prefixes using EXTERNAL_SOURCE_*
EXTERNAL_SOURCE_WIKIPEDIA: 'https://upload.wikimedia.org/wikipedia/'Optimization of images is done via sharp. The environment variables to set are:
IMAGE_QUALITY: 1 - 100IMAGE_PROGRESSIVE: true | false
You may also adjust the image quality setting per request with the q quality modifier described below.
While image-resizer will work as a standalone app, almost all of its facility is moot unless you run it behind a CDN. This has only been run behind AWS Cloudfront at this point and consequently all of the response headers are customized to work best in that environment. However other CDN's can not operate much differently, any pull requests in this regard would be most appreciated ;-)
A couple of routes are included with the default app, but the most important is the image generation one, which is as follows:
http://my.cdn.com/:modifiers/path/to/image.png[:format][:metadata]
Modifiers are a dash delimited string of the requested modifications to be made, these include:
Supported modifiers are:
- height: eg. h500
- width: eg. w200
- square: eg. s50
- crop: eg. cfill
- top: eg. y12
- left: eg. x200
- gravity: eg. gs, gne
- filter: eg. fsepia
- external: eg. efacebook
- quality: eg. q90
Crop modifiers:
- fit
- maintain original proportions
- resize so image fits wholly into new dimensions
- eg: h400-w500 - 400x600 -> 333x500
- default option
- fill
- maintain original proportions
- resize via smallest dimension, crop the largest
- crop image all dimensions that dont fit
- eg: h400-w500 - 400x600 -> 400x500
- cut
- maintain original proportions
- no resize, crop to gravity or x/y
- scale
- do not maintain original proportions
- force image to be new dimensions (squishing the image)
- pad
- maintain original proportions
- resize so image fits wholly into new dimensions
- padding added on top/bottom or left/right as needed (color is configurable)
Examples:
http://my.cdn.com/s50/path/to/image.pnghttp://my.cdn.com/h50/path/to/image.pnghttp://my.cdn.com/h50-w100/path/to/image.pnghttp://my.cdn.com/s50-gne/path/to/image.pnghttp://my.cdn.com/path/to/image.png- original image request, will be optimized but not resized
It is worthy of note that this application will not scale images up, we are all about keeping images looking good. So a request for h400 on an image of only 200px in height will not scale it up.
By default image-resizer will use s3 as the image source. To access an s3 object the full path of the image within the bucket is used, minus the bucket name eg:
https://s3.amazonaws.com/sample.bucket/test/image.png
translates to:
http://my.cdn.com/test/image.png
It is possible to bring images in from external sources and store them behind your own CDN. This is very useful when it comes to things like Facebook or Vimeo which have very inconsistent load times. Each external source can still enable any of the modification parameters list above.
In addition to the provided external sources, you can easily add your own external sources using EXTERNAL_SOURCE_* environment variables. For example, to add Wikipedia as an external source, set the following environment variable:
EXTERNAL_SOURCE_WIKIPEDIA: 'https://upload.wikimedia.org/wikipedia/'
Then you can request images beginning with the provided path prefix using the ewikipedia modifier, eg:
http://my.cdn.com/ewikipedia/en/7/70/Example.png
translates to:
https://upload.wikimedia.org/wikipedia/en/7/70/Example.png
You may also use Regular Expressions. Just use the same syntax as regex literals in JS. For example, to match any URLs matching *.googleusercontent.com:
EXTERNAL_SOURCE_GOOGLEUSERCONTENT=/^https?:\/\/.+\.googleusercontent\.com\//i
Or to simply allow any URL (below is a really naive regex for matching URLs):
EXTERNAL_SOURCE_URL=/^https?:\/\//i
It is worth noting that Twitter requires a full set of credentials as you need to poll their API in order to return profile pics.
A shorter expiry on images from social sources can also be set via IMAGE_EXPIRY_SHORT env var so they expiry at a faster rate than other images.
It is also trivial to write new source streams via the plugins directory. Examples are in src/streams/sources/.
You can convert images to another image format by appending an extra extension to the image path:
http://my.cdn.com/path/to/image.png.webp
JPEG (.jpg/.jpeg), PNG (.png), and WEBP (.webp) output formats are supported.
image-resizer can return the image metadata as a json endpoint:
http://my.cdn.com/path/to/image.png.json
Metadata is removed in all other image requests by default, unless the env var REMOVE_METADATA is set to false.
Included are both a .buildpacks file and a Procfile ready for Heroku deployment. Run the following cmd in your Heroku console to enable the correct buildpacks (copied from here).
heroku config:set BUILDPACK_URL=https://github.com/ddollar/heroku-buildpack-multi
The .buildpacks file will then take care of the installation process.
As mentioned above there is a minimum set of config vars that need to be set before image-resizer runs correctly.
It is also of note that due to some issues with GCC, sharp can not be used on the older Heroku stacks. Currently it requires cedar-14 stack.
To run image-resizer locally:
npm install -g gulp
npm install
gulp watchSee install docs for sharp if you get errors related to sharp / libvips.
The gulp setup includes nodemon which runs the app nicely, restarting between code changes. PORT can be set in the .env file if you need to run on a port other than 3001.
Tests can be run with: gulp test