NGINX with enabled image_filter
and secure_links
modules allows to quickly build an image server. The server can resize and cache images. It acts as a reverse-proxy between the external world and an internal storage with full-size image versions. Additionally, it protects images using MD5 hash values. It's a basic mechanism of resource authorization. It allows to access an image only if its hash value is known to the requester.
NGINX needs to be compiled with those two modules. This is not a default configuration. You can check enabled modules by running nginx -V
. nginx-extras
package in Ubuntu, however, comes with image_filter
and secure_links
already enabled.
Starting from NGINX 1.9.11 you can load modules dynamically. You can install the module and enable it. Here are Debian/Ubuntu specific instructions.
sudo apt-get install nginx-module-image-filter
Then, in nginx.conf
use load_module
directive to enable it:
load_module modules/ngx_http_image_filter_module.so;
Let's start with the cache. It is a file-based mechanism that stores resized versions of images. It should be large enough to reduce the number of resizing operations. In our example, the cache will be holding up to 10 GB of images. The max_size
parameter defines the point at which NGINX starts to remove the least recently requested images from the cache to make room for new items.
proxy_cache_path /var/cache/nginx/images levels=1:2 keys_zone=images:64M inactive=60d max_size=10GB;
Let's specify two server
blocks: an internal server for resizing images acting as a reverse-proxy to the actual storage (in our case S3) and a public server that proxies authorized requests to the internal server and stores requested images in the cache.
Here's the public server:
server {
listen 80;
server_name images.zaiste.net;
location /resize {
secure_link $arg_hash;
secure_link_md5 "$uri your-secret-goes-here";
if ($secure_link = "") {
return 404;
}
proxy_cache images;
proxy_cache_lock on;
proxy_cache_valid 30d;
proxy_cache_use_stale error timeout invalid_header updating;
expires 30d;
proxy_pass http://localhost:11337;
}
}
The secure_link
module works by generating a hash, which is the concatenation of the URL of the requested image and a secret string. This hash is long, so it is more convenient to append it to the URL rather than prepending it. This mechanisms prevents people from requesting arbitrary images.
Here's an example of a valid URL:
https://images.zaiste.net/resize/100x100/polyconf.jpg?hash=159c3isu11W1TBRm0YwBN
Internal server fetches images from S3 and resizes them on-the-fly based on dimensions specified in the URL. You can also define other image transformation such us cropping or rotating.
server {
listen 11337;
server_name localhost;
set $backend 's3.eu-central-1.amazonaws.com/your-bucket-name';
resolver 8.8.8.8;
resolver_timeout 30s;
proxy_buffering off;
proxy_pass_request_body off;
proxy_pass_request_headers off;
proxy_hide_header "x-amz-id-2";
proxy_hide_header "x-amz-request-id";
proxy_hide_header "x-amz-storage-class";
proxy_hide_header "Set-Cookie";
proxy_ignore_headers "Set-Cookie";
proxy_set_header Host $backend;
image_filter_jpeg_quality 95;
image_filter_buffer 50M;
image_filter_interlace on;
location ~ ^/resize/(?<width>\d+)x(?<height>\d+)/(?<name>.*)$ {
image_filter resize $width $height;
proxy_pass http://$backend/$name;
}
location ~ ^/resize/(?<width>\d+)/(?<name>.*)$ {
image_filter resize $width -;
proxy_pass http://$backend/$name;
}
location ~ ^/crop/(?<width>\d+)x(?<height>\d+)/(?<name>.*)$ {
image_filter crop $width $height;
proxy_pass http://$backend/$name;
}
}
NGINX needs an explicit definition for DNS resolution when using variables in proxy_pass
directive i.e. $backend
. Otherwise, you will get no resolver defined to resolve localhost
error.
You can clean AWS specific headers using proxy_hide_header
.
Original images may be large. We need to allocate enough memory for image_filter
module so it can load and resize them. In our example we set the image_filter_buffer
directive to handle image files of up to 50 MB. The dash ‑
in the second example tells NGINX to maintain the original aspect ratio.
This setup, if used in production, should also provide a rate limiting mechanism. This can be done in NGINX using limit_req
module.