I have a Ruby web app that is hosted on a Digital Ocean VPS. Under the current setup, I have a bash script that lives on the VPS that does the following upon execution:
Obviously, this approach isn't scalable (or even good). So I'm trying to setup a deployment pipeline using Travis CI + Capistrano that automatically installs, builds, tests, and deploys this web app upon git push.
So far, we have Travis CI performing the install, build, and testing of our code upon every git push, but we are stumbling on the best way to accomplish the deployment step. Because we have JS and SASS files that need to be built using gulp, we are unable to just pull straight from git onto the server and run. And since Travis CI is building these files already, we are curious if it is appropriate to leverage the built files from Travis CI and send them directly from Travis CI to our servers.
Some options we've considered:
SCP + Direct Transfer
Like the custom deployment FTP example on Travis CI, we could SCP the built files to our server and call a bash script that runs migrations and restarts the app server.
Capistrano + Direct Transfer
We could install Capistrano on Travis CI and use it to directly transfer files that were built on Travis CI to our servers. Before packaging and sending, we would need to clean up any files/directories that would not need to be transferred (node_modules, bower, etc.). After transferring the package, we could again use Capistrano to extract it, run database migrations, and restart the app server.
For this, how do we minimize deployment downtime? And how do we manage errors and rollbacks?
Capistrano + Git
We could push the built Travis CI files to a Git tag on github and use Capistrano to pull the Git tag on the server, run database migrations, and restart the app server.
For this, it seems that git releases should only be for prod, so how do we manage the different environments (dev, stage, and prod)?
Anyway, we've researched quite a bit online, and haven't been able to find a solution for our needs that lays out a best-standard approach to deploying from Travis CI to a non-supported deployment provider.
Given the situation above, what is the best way to deploy from Travis CI to a VPS?
I ended up using a Capistrano task to package the release from Travis CI and upload it to the appropriate server.
I had to create a new SCM in Capistrano (which I called Travis) that overwrote the release creation tasks of the default SCM.
I've posted it to this github thread: https://github.com/capistrano/capistrano/issues/722#issuecomment-54653745
And for completion, the custom Capistrano code is also below.
set :scm, :git
namespace :travis do
desc 'Check that travis is reachable'
task :check do
exit 1 unless true
end
desc 'Package to release'
task :create_release do
run_locally do
execute :mkdir, '-p', :'tmp'
execute "tar -cz --exclude tests --exclude vendor --exclude .git --exclude node_modules --exclude tmp/#{fetch(:release_timestamp)}.tar.gz -f tmp/#{fetch(:release_timestamp)}.tar.gz ."
end
on release_roles :all do
execute :mkdir, '-p', release_path
upload! "tmp/#{fetch(:release_timestamp)}.tar.gz", "#{release_path}/#{fetch(:release_timestamp)}.tar.gz"
execute "tar -xvf #{release_path}/#{fetch(:release_timestamp)}.tar.gz --directory #{release_path}"
execute "rm #{release_path}/#{fetch(:release_timestamp)}.tar.gz"
end
run_locally do
execute "rm -rf tmp"
end
end
desc 'Determine the revision that will be deployed'
task :set_current_revision do
run_locally do
set :current_revision, capture(:git, "rev-parse --short #{fetch(:branch)}")
end
end
end
namespace :deploy do
desc 'Use Travis'
task :use_travis do
set :scm, :travis
end
before :starting, :use_travis
end
Collected from the Internet
Please contact [email protected] to delete if infringement.
Comments