A buildpack to run mulitple processes within a dyno via runit.
Imagine you want to run nginx in your web dynos, binding to $PORT, and fronting unicorn. You want them to run individually as if they were specified directly in the Procfile (including output to the log stream) without worrying about daemonizing.
You'll probably use this in conjunction with multiple buildpacks.
For every process type (eg web) you want to run mulitple processes in, add an entry like this to your Procfile:
web: bin/runsvdir-dyno
Then, create a Procfile.web file (where web matches the process type in Procfile). It might look like this:
nginx: bin/nginx-runner -p $PORT ...
unicorn: bundle exec unicorn -c config/unicorn.rb ...
Now, when running a web dyno, both the nginx and unicorn processes will run under runit.
If any process specified in Procfile.web crashes, the entire dyno will be restarted.
The runit package is cached in Heroku's build cache and checked against its pinned
SHA-256 before every use. Missing or invalid cache entries are downloaded from
mirrors.edge.kernel.org, falling back to archive.ubuntu.com if the download or
checksum verification fails. Each source has a 10-second connection timeout, a
30-second transfer timeout, and one retry. Only verified downloads enter the cache;
the build fails if neither source provides a valid package.
Run the compile and supervisor lifecycle regression tests directly:
test/compile
test/runsvdir-dynoThe compile test uses stub downloads and extraction with real checksum checks.
The supervisor test uses a temporary Procfile and a stub runsvdir. Neither needs
network access, a Heroku app, or an installed copy of runit.
GitHub Actions runs both tests and shellcheck bin/* test/* on pushes and pull
requests.