Preforked Plack Handler for performance freaks
C++ Perl XS
Latest commit 6720c5d Sep 30, 2016 @kazeburo committed on GitHub Merge pull request #29 from ilmari/listen-sock
Allow passing an already-open socket to ->new

README.md

NAME

Gazelle - a Preforked Plack Handler for performance freaks

SYNOPSIS

$ plackup -s Gazelle --port 5003 --max-reqs-per-child 50000 \
     -E production -a app.psgi

DESCRIPTION

Gazelle is a PSGI Handler. It is derivied from Starlet. A lot of its code was rewritten or optimized by converting it to XS code.

Gazelle supports following features:

  • Supports HTTP/1.1. (Without Keepalive support.)
  • Ultra fast HTTP processing using picohttpparser.
  • Uses accept4(2) if the operating system supports it.
  • Uses writev(2) for output responses.
  • Prefork and graceful shutdown using Parallel::Prefork.
  • Hot deploy and unix domain socket using Server::Starter.

Gazelle is suitable for running HTTP application servers behind a reverse proxy such as nginx.

One can find a Benchmark here: https://github.com/kazeburo/Gazelle/wiki/Benchmark .

SAMPLE CONFIGURATION WITH NGINX

nginx.conf:

http {
  upstream app {
    server unix:/path/to/app.sock;
  }
  server {
    location / {
      proxy_pass http://app;
    }
    location ~ ^/(stylesheets|images)/ {
      root /path/to/webapp/public;
    }
  }
}

command line of running Gazelle

$ start_server --path /path/to/app.sock --backlog 16384 -- plackup -s Gazelle \
  -workers=20 --max-reqs-per-child 1000 --min-reqs-per-child 800 -E production -a app.psgi

start_server is bundled with Server::Starter

COMMAND LINE OPTIONS

In addition to the options supported by plackup, Gazelle accepts the following options:

--max-workers=

Number of worker processes (default: 10).

--timeout=

Seconds until timeout (default: 300).

--max-reqs-per-child=

Maximal number of requests to be handled before a worker process exits (default: 1000).

--min-reqs-per-child=

If set, randomize the number of requests handled by a single worker process between this value and the one supplied by --max-reqs-per-child (default: none).

--spawn-interval=

If set, worker processes will not be spawned more than once than every number of seconds given in the parameter. Furthermore, when a SIGHUP is being received, no more than one worker processes will be collected during this interval. This feature is useful for doing a "slow-restart". See http://blog.kazuhooku.com/2011/04/web-serverstarter-parallelprefork.html for more information. (default: none)

--child-exit=s

the subroutine code to be executed right before a child process exits. e.g. --child-exit='sub { POSIX::_exit(0) }'. (default: none)

SEE ALSO

Starlet Parallel::Prefork Server::Starter https://github.com/h2o/picohttpparser

LICENSE of Starlet

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

See http://www.perl.com/perl/misc/Artistic.html

LICENSE

Copyright (C) Masahiro Nagano.

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

AUTHOR

Masahiro Nagano kazeburo@gmail.com