Lower memory/cpu usage on reverse proxy #2036

vGimly · 2023-02-16T09:42:01Z

vGimly
Feb 16, 2023

I'm trying to run a small filtering proxy on a VPS with low resources. Something like 5 GB of disk space and 512 MB of RAM in total. But I need to process multiple client transfers of about 50GB in one connection. The client makes a GET request to the server, the server sends a request to an external resource and responds to the client.
On entry, it must filter client request headers, and on exit, it must filter server response headers and keep the payload body unchanged.
After filtering the headers, it should degrade to something as simple as:

char buffer[131072];
sysread(IN_SOCKET,buffer);
syswrite(OUT_SOCKET,buffer);

It should not save the body in memory, or in temporary files (because resources - there is no memory or disk space for it).

It probably needs to dive deeper into the internals of Mojolicious...

#!/usr/bin/env perl
use Mojolicious::Lite -signatures;

# external http server address
our $site='https://speed.hetzner.de';

# 1. Simple and dirty solution. Everything remains in memory.
# * hangs with multiple byte ranges
# ./piece get -v -H Range:bytes=-100 /100MB.bin > /dev/null
get '/100MB.bin' => sub ($c) {
my $hdr=$c->req->headers->to_hash;
delete $hdr->{Host}; # minimal header filter

$c->app->ua->get($site.$c->req->url => $hdr => sub ($ua,$tx) {
     $c->res->headers->from_hash($tx->res->headers->to_hash);
     $c->res->body($tx->res->body);
     $c->rendered($tx->res->code);

     # fix body to keep res->to_string readable
     $tx->res->body("<<body of @{[$tx->res->content->{size}]} bytes here>>");
     say STDERR "Proxy_Request>>\n",$tx->req->to_string,
                "Proxy_Answer >>\n",$tx->res->to_string,
                "\n<<Proxy_Finish\n";
});

$c->render_later;
};

# 2. Try to minimize the memory footprint
# MOJO_MAX_MESSAGE_SIZE=0 ./piece get -v -H Range:bytes=0-100,200-20000000 /10GB.bin | pv -cN pipe -s 20000000 > /dev/null
get '/10GB.bin' => sub ($c) {
my $ua=$c->app->ua;

# filter only allowed request headers
my $hdr_in=$c->req->headers->to_hash;
my %hdr=map {$_=>$hdr_in->{$_}} grep {exists $hdr_in->{$_}}
qw/Range Accept-Encodig TE Accept If-Modified-Since If-None-Match User-Agent/;

# prepare reverse proxy request
my $tx=$ua->max_response_size(0)->build_tx(GET => $site.$c->req->url => \%hdr );

# our answer can be 10GB+, remove limits:
$c->res->max_message_size(0);

# disable content::simple into content::multipart upgrading
# to transparently transfer "Range: bytes=0-2,5-10" requests
$tx->res->content->auto_upgrade(0);
$c->res->content->auto_upgrade(0);

# headers got - send them to client
$tx->res->content->once('body' => sub ($content) {

  # filter allowed response headers
  my $hdr_out=$tx->res->headers->to_hash;
  $c->res->headers->from_hash({map {($_ => $hdr_out->{$_})} grep {exists $hdr_out->{$_}}
     qw/Date Content-Type Content-Range Accept-Ranges Last-Modified Content-Length
        ETag Content-Disposition Location Transfer-Encoding Content-Encoding/});

$c->res->code($tx->res->code);
});

# try to send content ASAP
$tx->res->content->on('read' => sub ($content,$chunk) {
    $content->{content}=''; # try to free memory - we need no response, but not worked?
    $c->write($chunk); # how to safe clear $c->res->content?

# Should be simplified to sysread -> syswrite loop.
# There is no need for buffering, filtering or saving the body of the message.
# Too much of CPU and memory is being used here...
});

# make request
$ua->start($tx, sub ($ua,$tx) { say STDERR "\nRProxy Done: "
    .$tx->res->content->{size}.'/'
    .length($tx->res->body)} # now it keeps full body - should be empty
);

# may be there is a better way to say "answer will be when the Mojo::UserAgent finish?"
$c->render_later;
$c->inactivity_timeout(90);
};

app->start;

kraih · 2023-02-16T10:27:53Z

kraih
Feb 16, 2023
Maintainer

You probably want to take a look at how the proxy helpers work.

mojo/lib/Mojolicious/Plugin/DefaultHelpers.pm

Line 212 in 03d1b64

sub _proxy_start_p {

And this might be a task where mojo.js can be a lot more efficient than Mojolicious.

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Lower memory/cpu usage on reverse proxy #2036

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Replies: 1 comment

{{title}}

Select a reply

Lower memory/cpu usage on reverse proxy #2036

vGimly Feb 16, 2023

Replies: 1 comment

kraih Feb 16, 2023 Maintainer

vGimly
Feb 16, 2023

kraih
Feb 16, 2023
Maintainer