Go in a Monorepo

This blog was originally written for Gopher Academy’s Advent 2015 series, but it is also mirrored here for convenience. Enjoy!

Please note that I am no longer employed by DigitalOcean.

A “monorepo” is a monolithic code repository which contains many different projects and libraries. At DigitalOcean, we have created a monorepo called cthulhu to house all of our Go code. The purpose of cthulhu is to provide a one-stop shop for our internal Go libraries, services, tools, and third party dependencies.

Bryan Liles previously wrote a blog post about cthulhu early in 2015. This post will cover many of the same topics, but will also detail some of the improvements we have made over the past year to make cthulhu even better.

Monorepo Structure

cthulhu is a single, monolithic git repository. It contains several important top-level files and directories:

├── .drone.yml
├── analyze.sh
├── docode/
│   └── src/
└── third_party/
    └── src/

By setting the $GOPATH environment variable to ${CTHULHU}/third_party:${CTHULHU}/docode, we are able to go get third party dependencies and build and test our internal code.

Internal Code

All of our internal code resides within ${CTHULHU}/docode/. We have created four partitioned areas within this directory, which each serve a different purpose:

└── src/
    ├── doge/
    ├── exp/
    ├── services/
    └── tools/

Third Party Code

Third party dependencies reside within ${CTHULHU}/third_party/. Because this directory appears in $GOPATH before ${CTHULHU}/docode/, it is possible to simply go get dependencies to add them to the repository.

To avoid dealing with the potential hassles of git submodules, we simply rename the .git directory in each dependency to .checkout_git before committing it to the repository. When we want to update a vendored dependency, we rename the directory back to .git and pull the latest changes. Perhaps there is a better approach than this technique, but because we rarely need to bump third party dependencies, this approach has worked well thus far.

Advantages of a Monorepo

When the idea of cthulhu was proposed by Antoine Grondin, I was initially very skeptical. However, after spending a couple of weeks working in a monorepo, I firmly believe that it is the best possible option when working with a team of Go developers.

Many of the excellent tools in the Go ecosystem work even better when used in a monorepo. For example, gorename is incredibly useful. Just recently, I was able to use it to rename an identifier from foo.FooDB to foo.DB in every single Go source file in the entire repository. The ability to refactor en-masse in this fashion is invaluable. Any changes made will be compiled and tested across all internal Go projects.

Recently, we have begun to experiment with adding a wide variety of automated linting checks which are run using analyze.sh during CI builds. Because all of our Go code resides in cthulhu, we can fail any builds which introduce code that does not meet the standards set by tools like gofmt and go vet. In addition, we have started working on internal linting tools using the excellent go/ast package in the standard library. One of these is a tool we call explint, which ensures that code in our experimental exp/ directory cannot be imported by code in services/ or tools/: our production code directories.

Finally, using a monorepo completely solves the issue of vendoring third party dependencies. Every project within cthulhu uses the same version of a third party library, and whenever the library is bumped, every project which makes use of it is built and tested automatically. Recently, we tried to make use of the new $GO15VENDOREXPERIMENT=1, but unfortunately, it caused issues with goimports, and we were forced to revert the change. In the future, we’d like to try it again, once goimports is updated.

Disadvantages of a Monorepo

While the benefits of using a monorepo do outweigh the costs in our case, it isn’t always the perfect approach.

As cthulhu has continued to grow, our CI test durations have grown in parallel. More projects and tests are added daily, and some of these require database integration testing. We briefly experimented with using Russ Cox’s gt utility, but as of today, it has not been adopted in cthulhu.

Because a monorepo may have many commits per day to many different projects, “feature branches” can quickly become out of date. This is why we chose to adopt the exp/ experimental directory, instead of requiring experimental code to live in feature branches. exp/ has worked well for us thus far, but we are still working out a set of guidelines which should be followed when moving a project from exp/ to a production directory like services/ or tools/.

As stated before, we decided to rename the .git directory for our third party dependencies instead of dealing with git submodules. This approach is likely not the ideal one, but because we rarely need to touch third party code, it has not been a major issue thus far.


DigitalOcean’s monorepo, cthulhu, has provided one of the most pleasant development experiences I have ever had. It works vastly better than attempting to keep dependencies up to date across multiple repositories. It enables large-scale refactoring using the excellent tools in the Go ecosystem, without fear of breaking one or more external projects.

If you have any questions or would like clarification on monorepos, feel free to contact me: “mdlayher” on Gophers Slack!