Lowering Capistrano concurrency March 5th

If you happen to run into problems when deploying code from your GitHub repository to a larger number of machines in parallel (the dreaded ssh_exchange_identification: Connection closed by remote host error) you may want to look into Capistrano’s :max_hosts option to limit the operation at hand to a given number of machines at once.

This is all nice and dandy, unless you find out that there is no way to change that option for an existing task, for example from Capistrano’s default library of recipes. To solve this (and make my deploys to the 12 machines in a cluster I manage more reliable) I simply duplicated the update_code task to my application’s local config/deploy.rb:

namespace :deploy do
  task :update_code, :except => { :no_release => true }, :max_hosts => 4 do
    on_rollback { run "rm -rf #{release_path}; true" }
    strategy.deploy!
    finalize_update
  end
end

That way, the update_code task is split into 3 batches of 4 hosts and checkouts/code updates from the GitHub repository (which is probably just rate limited to not be slammed with requests all at once) work nicely again.

Since this task is fairly compact, I didn’t do any further research. Is there a better way? Let me know in the comments.