Optimising concurrent ImageMagick Requests using redis/php-resque

Question

I am working on a site that uses ImageMagick to generate images. The site will get hundreds of request every minute and using ImageMagick to do this causes the site to crash.

So we implemented Redis and Php-resque to do the ImageMagick generating in the background on a seperate server so that it doesn't crash our main one. The problem is that it's still taking a very long time to get images done. A user might expect to wait up to 2-3 minutes for an image request because the server is so busy processing these images.

I am not sure what information to give you, but I'm more looking for advice. I think if we can cut down the initial process time for the ImageMagick request, then obviously this will help speed up the amount of images we can process.

Below is a sample of the ImageMagick script that we use:

convert -size 600x400 xc:none \( ".$path."assets/images/bases/base_image_69509021433289153_8_0.png -fill rgb\(255,15,127\) -colorize 100% \) -composite \( ".$path."assets/images/bases/eye_image_60444011438514404_8_0.png -fill rgb\(15,107,255\) -colorize 100% \) -composite \( ".$path."assets/images/markings/marking_clan_8_marking_10_1433289499.png -fill rgb\(255,79,79\) -colorize 100% \) -composite \( ".$path."assets/images/bases/shading_image_893252771433289153_8_0.png -fill rgb\(135,159,255\) -colorize 100% \) -compose Multiply -composite \( ".$path."assets/images/highlight_image_629750231433289153_8_0.png -fill rgb\(27,35,36\) -colorize 100% \) -compose Overlay -composite \( ".$path."assets/images/lineart_image_433715161433289153_8_0.png -fill rgb\(0,0,0\) -colorize 100% \) -compose Over -composite ".$path."assets/generated/queue/tempt_preview_27992_userid_0_".$filename."_file.png

My theory is that the reason this takes quite a long time is due to the process of colouring the images. Is there a way to optimise this process at all?

Anyone who has some experience with handling heavy loads of imagemagick processes or can see some glaringly easy ways to optimise our requests, I'd be very greatful.

Thank you :)

Answer 1

The queue is being processed one at a time? Have you tried to make concurrent jobs, that will keep running in parallel so you work more than one element at once if that is the case?

Answer 2

Your command actually boils down to this:

convert -size 600x400 xc:none                                 \
    \( 1.png -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 2.png -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 3.png -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 4.png -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 5.png -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 6.png -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    result.png

My thoughts are as follows:

Point 1:

The first -composite onto a blank canvas seems pointless - presumably 1.png is a 600x400 PNG with transparency, so your first line can avoid the compositing operation and save 16% of the processing time by changing to:

convert -background none 1.png -fill ... -colorize 100% \
   \( 2.png ..
   \( 3.png ...

Point 2

I put the equivalent of your command into a loop and did 100 iterations and it takes 15 seconds. I then changed all your reads of PNG files into reads of MPC files - or Magick Pixel Cache files. That reduced the processing time to just under 10 seconds, i.e. by 33%. A Magic Pixel Cache is just a pre-decompressed, pre-decoded file that can be read directly into memory without any CPU effort. You could pre-create them whenever your catalogue changes and store them alongside the PNG files. To make one you do

convert image.png image.mpc

and you will get out image.mpc and image.cache. Then you would simply change your code to look like this:

convert -size 600x400 xc:none                                 \
    \( 1.mpc -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 2.mpc -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 3.mpc -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 4.mpc -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 5.mpc -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    \( 6.mpc -fill rgb\(x,y,z\) -colorize 100% \) -composite  \
    result.png

Point 3

Unfortunately you haven't answered my questions yet, but if your assets catalogue is not too big, you could put that (or the MPC equivalents above) onto a RAM disk at system startup.

Point 4

You should definitely run in parallel - that will yield the biggest gains of all. It is very simple with GNU Parallel - example here.

If you are using REDIS, it is actually easier than that. Just LPUSH your MIME-encoded images into a REDIS list like this:

#!/usr/bin/perl
################################################################################
# generator.pl <number of images> <image size in bytes>
# Mark Setchell
# Base64 encodes and sends "images" of specified size to REDIS
################################################################################
use strict;
use warnings FATAL => 'all';
use Redis;
use MIME::Base64;
use Time::HiRes qw(time);

my $Debug=0;    # set to 1 for debug messages

my $nargs = $#ARGV + 1;
if ($nargs != 2) {
    print "Usage: generator.pl <number of images> <image size in bytes>\n";
    exit 1;
}

my $nimages=$ARGV[0];
my $imsize=$ARGV[1];

# Our "image"
my $image="x"x$imsize;

printf "DEBUG($$): images: $nimages, size: $imsize\n" if $Debug;

# Connection to REDIS
my $redis = Redis->new;
my $start=time;

for(my $i=0;$i<$nimages;$i++){
   my $encoded=encode_base64($image,'');
   $redis->rpush('images'=>$encoded);
   print "DEBUG($$): Sending image $i\n" if $Debug;
}
my $elapsed=time-$start;
printf "DEBUG($$): Sent $nimages images of $imsize bytes in %.3f seconds, %d images/s\n",$elapsed,int($nimages/$elapsed);

and then run multiple workers that all sit there doing BLPOPs of jobs to do

#!/usr/bin/perl
################################################################################
# worker.pl
# Mark Setchell
# Reads "images" from REDIS and uudecodes them as fast as possible
################################################################################
use strict;
use warnings FATAL => 'all';
use Redis;
use MIME::Base64;
use Time::HiRes qw(time);

my $Debug=0;    # set to 1 for debug messages
my $timeout=1;  # number of seconds to wait for an image
my $i=0;

# Connection to REDIS
my $redis = Redis->new;

my $start=time;

while(1){
   #my $encoded=encode_base64($image,'');
   my (undef,$encoded)=$redis->blpop('images',$timeout);
   last if !defined $encoded;
   my $image=decode_base64($encoded);
   my $l=length($image);
   $i++; 
   print "DEBUG($$): Received image:$i, $l bytes\n" if $Debug;
}

my $elapsed=time-$start-$timeout; # since we waited that long for the last one
printf "DEBUG($$): Received $i images in %.3f seconds, %d images/s\n",$elapsed,int($i/$elapsed);

If I run one generator process as above and have it generate 100,000 images of 200kB each, and read them out with 4 worker processes on my reasonable spec iMac, it takes 59 seconds, or around 1,700 images/s can pass through REDIS.