#373 Zero-Downtime Deployment pro

Thanks very much for this, Ryan - great timing considering the recent release of Phusion's new premium Passenger Enterprise product, which also allows for rolling restarts. It's good to learn that zero-downtime updates are also doable with 100% open source tools via Unicorn, albeit with increased memory requirements during restart.

Also thanks for including the bit about how to deploy a temporary site maintenance page - a very helpful tip!

Asset precompile can hinder zero downtime too, however it's relatively easy to fix too, if you use nginx:

  location ~ ^/assets/  {
    gzip_static on;
    try $uri /old$uri;
    expires max;
    add_header  Cache-Control public;
  }

and if you've done generating your public/assets, throw away public/oldassets, and move public/assets to its place!

Very good the part about maintenance! I loved heroku's simplicity of such tasks, however I didn't know it was this easy to accomplish about the same thing just with capistrano and a quick Nginx config.

Sometimes I am having with asset pipeline not get reloaded and the code itself.
It happend when I use this zero-downtime restart also regular restart.

To solve it I need to: cap unicorn:stop; sleep 10; cap unicorn:start

Has onyone encountered this?

Thanks Ryan for this episode. Yet to watch this. I would like to know whether this work with Ruby1.8.7?

For some strange reasons i keep getting
the task `unicorn:setup' does not exist
when i type in
cap unicorn:setup unicorn:stop unicorn:start

I'm a little confused about these line:

  old_pid = "#{server.config[:pid]}.oldbin"
  if File.exists?(old_pid) && server.pid != old_pid 

Wouldn't that set the old_pid variable to the filename of the unicorn.pid.oldpid file, and then compare that filename with the pid of the new server? For example, it would expand to:

if File.exists?('/tmp/unicorn.pid.oldpid') && 13274 != '/tmp/unicorn.pid.oldpid'

Or am I misunderstanding something?

Hi guys, It seems that Capistrano has in fact now removed the cap deploy:web:[action] tasks. Being that I'm very new to Capistrano, could anyone point me to some info on how to get that functionality back?

Why does this file https://github.com/railscasts/373-zero-downtime-deployment/blob/master/blog-after/config/recipes/templates/nginx_unicorn.erb contain this twice:

if (-f $document_root/system/maintenance.html) {
return 503;
}
error_page 503 @maintenance;
location @maintenance {
rewrite ^(.*)$ /system/maintenance.html last;
break;
}

I ran into this issue after 5 deployments where the new master process would report a Gemfile not found (Bundler::GemfileNotFound) error and quit. It turn's out that the new master process was using a environment variable setting BUNDLE_GEMFILE which referred to a older release that has been cleaned up and no longer exists.

The only way to fix this at the time was to stop and start the unicorn process which defeat's the purpose of zero downtime deployment. To prevent this from happening in your unicorn config template add this block

before_exec do |server|
  ENV['BUNDLE_GEMFILE'] = "<%= current_path %>/Gemfile"
end

you do have to run unicorn:setup and maybe stop and start the service to make sure the new config setting is picked up.

hi, great! I am trying to figure out how to get the right syntax highlighting for unicorn_init.erb as shown in the screencast?