Guest post by Kelsey Gabriel

Updating a Rails 5.2 site to 6.0 is relatively straightforward for a project with a small list of well-managed dependencies. However, though everything might be working in your development environment (or even in your rails s -e production tests), it might still fail on your production server.

If you use Puma to serve Rails in your production environment, you may have noticed rails app:update adding the following lines to your /config/puma.rb file:

# Specifies the `pidfile` that Puma will use.
pidfile ENV.fetch("PIDFILE") { "tmp/pids/server.pid" }

Puma’s “pidfile”, or process ID file, is where Puma stores the unique ID it uses while it’s running. A server admin or another process on the system may use this number to check Puma’s status or issue the process a kill command if necessary. Unfortunately, if the directory Puma wants to store this file doesn’t exist, Puma won’t start at all:

Errno::ENOENT: No such file or directory @ rb_sysopen - tmp/pids/server.pid

Creating this folder in your project repository isn’t quite enough—and in fact, in your local repository, it’s probably already there. The /tmp/ folder is added to /.gitignore by default, and for good reason: /tmp/ can be a dumping ground for temporary files and caches, and you usually don’t want to commit these files. In order to ignore the files that should be ignored and still commit the above pidfile directory to your remote repository, you’ll need to add the following lines to your .gitignore file:

# Ignore all logfiles and tempfiles.
/log/*
/tmp/*
/tmp/pids/*              # this line
!/log/.keep
!/tmp/.keep
!/tmp/pids               # this line
!/tmp/pids/.keep         # and this line

The bang (!) denotes an exception. Since /tmp/* is on the ignore list, we need to explicitly allow /tmp/pids to be tracked. But that’s not enough; git won’t commit empty directories, so we need to add a file to that directory. An empty .keep file is a common method of handling this and making sure the directory holding it can get pushed up.

The other two lines we add, /tmp/pids/* and !/tmp/pids/.keep, are a precaution. We don’t want to share any files that don’t need to be shared, and publicly broadcasting the process ID of our server instance feels sloppy. Therefore, we should exclude all the contents of /tmp/pids/ from our commits except for the .keep file itself.