imaginary

command module

v0.0.0-...-9cb131c Latest Latest Go to latest Published: Nov 6, 2015 License: MIT Imports: 22 Imported by: 0

Details

Valid go.mod file

The Go module system was introduced in Go 1.11 and is the official dependency management solution for Go.
Redistributable license

Redistributable licenses place minimal restrictions on how software can be used, modified, and redistributed.
Tagged version

Modules with tagged versions give importers more predictable builds.
Stable version

When a project reaches major version v1 it is considered stable.
Learn more about best practices

Repository

github.com/cnaize/imaginary

Links

Open Source Insights

README ¶

imaginary

Fast HTTP microservice written in Go for high-level image processing backed by bimg and libvips. imaginary can be used as private or public HTTP service for massive image processing. It's almost dependency-free and only uses net/http native package for better performance.

Supports multiple image operations exposed as a simple HTTP API, with additional optional features such as API token authorization, gzip compression, HTTP traffic throttle strategy and CORS support for web clients.

imaginary can read images from HTTP payloads, server local path or remote HTTP servers, supporting JPEG, PNG, WEBP and TIFF formats and it's able to output to JPEG, PNG and WEBP, including conversion between them.

It uses internally libvips, a powerful and efficient library written in C for image processing which requires a low memory footprint and it's typically 4x faster than using the quickest ImageMagick and GraphicsMagick settings or Go native image package, and in some cases it's even 8x faster processing JPEG images.

To get started, take a look the installation steps, usage cases and API docs.

imaginary is currently used in production processing thousands of images per day.

Supported image operations

Resize
Enlarge
Crop
Rotate (with auto-rotate based on EXIF orientation)
Flip (with auto-flip based on EXIF metadata)
Flop
Zoom
Thumbnail
Extract area
Watermark (customizable by text)
Custom output color space (RGB, black/white...)
Format conversion (with additional quality/compression settings)
Info (image size, format, orientation, alpha...)

Prerequisites

libvips v7.40.0+ (7.42.0+ recommended)
C compatible compiler such as gcc 4.6+ or clang 3.0+
Go 1.3+

Installation

go get -u github.com/h2non/imaginary

libvips

Run the following script as sudo (supports OSX, Debian/Ubuntu, Redhat, Fedora, Amazon Linux):

curl -s https://raw.githubusercontent.com/lovell/sharp/master/preinstall.sh | sudo bash -

The install script requires curl and pkg-config

Docker

See Dockerfile for image details.

Fetch the image (comes with latest stable Go and libvips versions)

docker pull h2non/imaginary

Start the container with optional flags (default listening on port 9000)

docker run -p 9000:9000 h2non/imaginary -cors -gzip

Start the container in debug mode:

docker run -p 9000:9000 -e "DEBUG=*" h2non/imaginary

Enter to the interactive shell in a running container

sudo docker exec -it <containerIdOrName> bash

Stop the container

docker stop h2non/imaginary

You can see all the Docker tags here.

Heroku

Click on the Heroku button to easily deploy your app:

Or alternatively you can follow the manual steps:

Clone this repository:

git clone https://github.com/h2non/imaginary.git

Set the buildpack for your application

heroku config:add BUILDPACK_URL=https://github.com/h2non/heroku-buildpack-imaginary.git

Add Heroku git remote:

heroku git:remote -a your-application

Deploy it!

git push heroku master

Recommended resources

Given the multithreaded native nature of Go, in term of CPUs, most cores means more concurrency and therefore, a better performance can be achieved. From the other hand, in terms of memory, 512MB of RAM is usually enough for small services with low concurrency (<5 request/second). Up to 2GB for high-load HTTP service processing potentially large images or exposed to an eventual high concurrency.

If you need to expose imaginary as public HTTP server, it's highly recommended to protect the service against DDoS-like attacks. imaginary has built-in support for HTTP concurrency throttle strategy to deal with this in a more convenient way and mitigate possible issues limiting the number of concurrent requests per second and caching the awaiting requests, if necessary.

Production notes

In production focused environments it's highly recommended to enable the HTTP concurrency throttle strategy in your imaginary servers.

The recommended concurrency limit per server to guarantee a good performance is up to 20 requests per second.

You can enable it simply passing a flag to the binary:

$ imaginary -concurrency 20

Scalability

If you're looking for a large scale solution based on imaginary, you should scale it horizontally and distribute the HTTP load over a pool of imaginary servers.

Assuming that you want to provide a high availability to deal efficiently with about 100 concurrent req/sec, you should probably use a front balancer (e.g: HAProxy) to delegate the request control flow and quality of service distributing the HTTP load across a pool of server:

        |==============|
        |  Dark World  |
        |==============|
              ||||
        |==============|
        |   Balancer   |
        |==============|
           |       |   
          /         \
         /           \
        /             \
 /-----------\   /-----------\
 | imaginary |   | imaginary | (*N)
 \-----------/   \-----------/

Clients

node.js/io.js

Performance

libvips is probably the faster open source solution for image processing. Here you can see some performance test comparisons for multiple scenarios:

libvips speed and memory usage
sharp performance tests
bimg (Go library with C bindings to libvips)

Benchmark

See benchmark.sh for more details

Environment: Go 1.4.2. libvips-7.42.3. OSX i7 2.7Ghz

Requests  [total]       200
Duration  [total, attack, wait]   10.030639787s, 9.949499515s, 81.140272ms
Latencies [mean, 50, 95, 99, max]   83.124471ms, 82.899435ms, 88.948008ms, 95.547765ms, 104.384977ms
Bytes In  [total, mean]     23443800, 117219.00
Bytes Out [total, mean]     175517000, 877585.00
Success   [ratio]       100.00%
Status Codes  [code:count]      200:200

Conclusions

imaginary can deal efficiently with up to 20 request per second running in a multicore machine, where it crops a JPEG image of 5MB and spending per each request less than 100 ms

The most expensive image operation under high concurrency scenarios (> 20 req/sec) is the image enlargement, which requires a considerable amount of math operations to scale the original image. In this kind of operation the required processing time usually grows over the time if you're stressing the server continuously. The advice here is as simple as taking care about the number of concurrent enlarge operations to avoid server performance bottle necks.

Usage

imaginary server

Usage:
  imaginary -p 80
  imaginary -cors -gzip
  imaginary -concurrency 10
  imaginary -enable-url-source
  imaginary -h | -help
  imaginary -v | -version

Options:
  -a <addr>                 bind address [default: *]
  -p <port>                 bind port [default: 8088]
  -h, -help                 output help
  -v, -version              output version
  -cors                     Enable CORS support [default: false]
  -gzip                     Enable gzip compression [default: false]
  -key <key>                Define API key for authorization
  -mount <path>             Mount server local directory
  -http-cache-ttl <num>     The TTL in seconds. Adds caching headers to locally served files.
  -http-read-timeout <num>  HTTP read timeout in seconds [default: 30]
  -http-write-timeout <num> HTTP write timeout in seconds [default: 30]
  -enable-url-source        Enable remote HTTP URL image source processing [default: false]
  -certfile <path>          TLS certificate file path
  -keyfile <path>           TLS private key file path
  -concurreny <num>         Throttle concurrency limit per second [default: disabled]
  -burst <num>              Throttle burst max cache size [default: 100]
  -mrelease <num>           OS memory release inverval in seconds [default: 30]
  -cpus <num>               Number of used cpu cores.
                            (default for current machine is 8 cores)

Start the server in a custom port

imaginary -p 8080

Also, you can pass the port as environment variable

PORT=8080 imaginary

Enable HTTP server throttle strategy (max 10 request/second)

imaginary -p 8080 -concurrency 10

Enable remote URL image fetching (then you can do GET request passing the url=http://server.com/image.jpg query param)

imaginary -p 8080 -enable-url-source

Mount local directory (then you can do GET request passing the file=image.jpg query param)

imaginary -p 8080 -mount ~/images

Send caching headers (only possible with the -mount option). The headers can be set in either "cache nothing" or "cache for N seconds". By specifying 0 Imaginary will send the "don't cache" headers, otherwise it sends headers with a TTL. The following example informs the client to cache the result for 1 year.

imaginary -mount ~/images -http-cache-ttl 31556926

Increase libvips threads concurrency (experimental)

VIPS_CONCURRENCY=10 imaginary -p 8080 -concurrency 10

Enable debug mode

DEBUG=* imaginary -p 8080

Or filter debug output by package

DEBUG=imaginary imaginary -p 8080

Examples

Reading a local image (you must pass the -mount=<directory> flag):

curl -O "http://localhost:8088/crop?width=500&height=400&file=foo/bar/image.jpg"

Fetching the image from a remote server (you must pass the -enable-url-source flag):

curl -O "http://localhost:8088/crop?width=500&height=400&url=https://raw.githubusercontent.com/h2non/imaginary/master/fixtures/large.jpg"

Playground

imaginary exposes an ugly HTML form for playground purposes in: http://localhost:8088/form

HTTP API

Authorization

imaginary supports a simple token-based API authorization. To enable it, you should specific the flag -key secret when you call the binary.

API token can be defined as HTTP header (API-Key) or query param (key).

Example request with API key:

POST /crop HTTP/1.1
Host: localhost:8088
API-Key: secret

Errors

imaginary will always reply with the proper HTTP status code and JSON body with error details.

Here an example response error when the payload is empty:

{
  "message": "Cannot read payload: no such file",
  "code": 1
}

See all the predefined supported errors here.

Params

Complete list of available params. Take a look to each specific endpoint to see which params are supported. Image measures are always in pixels, unless otherwise indicated.

width int - Width of image area to extract/resize
height int - Height of image area to extract/resize
top int - Top edge of area to extract. Example: 100
left int - Left edge of area to extract. Example: 100
areawidth int - Height area to extract. Example: 300
areaheight int - Width area to extract. Example: 300
quality int - JPEG image quality between 1-100. Default 80
compression int - PNG compression level. Default: 6
rotate int - Image rotation angle. Must be multiple of 90. Example: 180
factor int - Zoom factor level. Example: 2
margin int - Text area margin for watermark. Example: 50
dpi int - DPI value for watermark. Example: 150
textwidth int - Text area width for watermark. Example: 200
opacity float - Opacity level for watermark text. Default: 0.2
force bool - Force image transformation size. Default: false
nocrop bool - Disable crop transformation enabled by default by some operations. Default: false
noreplicate bool - Disable text replication in watermark. Default false
norotation bool - Disable auto rotation based on EXIF orientation. Default false
noprofile bool - Disable adding ICC profile metadata. Default false
text string - Watermark text content. Example: copyright (c) 2189
font string - Watermark text font type and format. Example: sans bold 12
color string - Watermark text RGB decimal base color. Example: 255,200,150
type string - Specify the image format to output. Possible values are: jpeg, png and webp
gravity string - Define the crop operation gravity. Supported values are: north, south, centre, west and east. Defaults to centre.
file string - Use image from server local file path. In order to use this you must pass the -mount=<dir> flag.
url string - Fetch the image from a remove HTTP server. In order to use this you must pass the -enable-url-source flag.
colorspace string - Use a custom color space for the output image. Allowed values are: srgb or bw (black&white)

GET /

Content-Type: application/json

Serves as JSON the current imaginary, bimg and libvips versions.

GET /health

Content-Type: application/json

Provides some useful statistics about the server stats with the following structure:

uptime number - Server process uptime in seconds.
allocatedMemory number - Currently allocated memory in megabytes.
totalAllocatedMemory number - Total allocated memory over the time in megabytes.
gorouting number - Number of running gorouting.
cpus number - Number of used CPU cores.

Example response:

{
  "uptime": 1293,
  "allocatedMemory": 5.31,
  "totalAllocatedMemory": 34.3,
  "goroutines": 19,
  "cpus": 8
}

GET /form

Content Type: text/html

Serves an ugly HTML form, just for testing/playground purposes

GET | POST /info

Accepts: image/*, multipart/form-data. Content-Type: application/json

Returns the image metadata as JSON:

{
  "width": 550,
  "height": 740,
  "type": "jpeg",
  "space": "srgb",
  "hasAlpha": false,
  "hasProfile": true,
  "channels": 3,
  "orientation": 1
}