Nginx rolling restart of Rails app with capistrano

For the life of me I can't figure out how to make this work properly.

The problem is similar to what others have, such as: How to do a rolling restart of a cluster of mongrels

We, however, are using Nginx/Passenger instead of Mongrel.

The issue is that on a deploy if we use this standard :restart task:

task :restart, :roles => [:app], :except => {:no_release => true} do
  run "cd #{deploy_to}/current && touch tmp/restart.txt"
end

It touches the restart.txt file across every web server, but any passenger instances currently serving requests need to finish before the new ones are spawned it seems. This creates a serious delay and causes our app to be unavailable for up to 2 minutes while everything is coming back up.

In order to get around that the plan is to do the following:

1. deploy code
2. go to server 1, remove it from the load balancer
3. restart nginx-passenger on server 1
4. wait 60 seconds
5. add server 1 back to load balancer
6. go to server 2 (repeat steps 3 - 开发者_如何转开发5)

To accomplish this, I attempted this:

(lb.txt is the file that the load balancer looks for)

task :restart, :roles => [:app], :except => {:no_release => true} do
  servers = find_servers_for_task(current_task)
  servers.map do |s|
    run "cd #{deploy_to}/current && echo '' > public/lb.txt", :host => s.host
    run %Q{rvmsudo /etc/init.d/nginx-passenger restart > /dev/null}, :host => s.host
    sleep 60
    run "cd #{deploy_to}/current && echo 'ok' > public/lb.txt", :host => s.host
  end
end

This almost works, however, during the deploy it seemed to run the loop through the servers once per servers listed in the :app role. We currently have 6 app servers, so the loop runs 6 times, restarting nginx-passenger 6 times per server.

I just need this loop to run through one time.

I know it seems that eventually passenger will get rolling restarts, but they do not seem to exist yet.

If it helps, we are using Capistrano 2.x and Rails 3

Any help would be great.

Thanks.

---

run "cd #{deploy_to}/current && echo 'ok' > public/lb.txt", :host => s.host

should actually be:

run "cd #{deploy_to}/current && echo 'ok' > public/lb.txt", :hosts => s.host

---

I ran across this gem capify-ec2 which has a rolling restart feature. capify-ec2 on github.

I am about to install it and try it out.

Here's the description from the readme that describes what their rolling restart feature does:

"This feature allows you to deploy your code to instances one at a time, rather than simultaneously. This becomes useful for more complex applications that may take longer to startup after a deployment. Capistrano will perform a full deploy (including any custom hooks) against a single instance, optionally perform a HTTP healthcheck against the instance, then proceed to the next instance if deployment was successful.

After deployment a status report is displayed, indicating on which instances deployment succeeded, failed or did not begin. With some failures, further action may need to be taken manually; for example if an instance is removed from an ELB (see the 'Usage with Elastic Load Balancers' section below) and the deployment fails, the instance will not be reregistered with the ELB, for safety reasons."