This file should serve as an entrypoint for learning about and contributing to Rill Developer.
If you're a Rill team member, you can run rill devtool start
from the project root to start a full local development environment. If you select the cloud preset, you can fill it with seed data using rill devtool seed cloud
. See cli/cmd/devtool/README.md
for more details.
This is a full list of development dependencies:
- Docker
- Node.js 18 (we recommend installing it with nvm)
- Go 1.23 (on macOS, install with
brew install go
) - Buf (Protocol Buffers) (on macOS, install with
brew install bufbuild/buf/buf
)
- Integrate
golangci-lint
(instructions)
Running make
will build a production-ready binary and output it to ./rill
(see cli/README.md
for details).
For detailed instructions on how to run or test the application in development, see the README.md
file in the individual components' directories (e.g. web-local/README.md
for the local web app).
To release a new version of Rill, first create a release branch named release-<minor version>
:
git checkout -b release-0.47
git push
This will trigger a rollout of the release branch to the staging environment, which can be used to QA the release. Any subsequent fixes should be contributed to main
and merged or cherry-picked into the release branch.
When ready to release, create and push a Git tag on the release branch with the new version number:
git checkout release-0.47
git tag -a v0.47.0 -m "v0.47.0 release"
git push origin v0.47.0
This will trigger the cli-release.yml
Github Action, which will:
- Build binaries for macOS (arm64, amd64) and Linux (amd64)
- Upload the binaries to
https://cdn.rilldata.com/rill/$VERSION/$NAME
- Upload the newest version of the install script (in
scripts/install.sh
) tohttps://cdn.rilldata.com/install.sh
- Create a Github release containing an auto-generated changelog and the new binaries
- Publish the new version to our brew tap
rilldata/tap/rill
You can follow the progress of the release action from the "Actions" tab. It usually completes in about 10 minutes. See our internal release run book for more details.
Check out the current release branch and cherry pick the fixes to include in the patch release:
git checkout release-0.47
git cherry-pick <commit>
Then when ready to release, create and push a Git tag on the release branch with the new version number:
git checkout release-0.47
git tag -a v0.47.1 -m "v0.47.1 release"
git push origin v0.47.1
Here's a high-level overview of the technologies we use for different parts of the project:
- Typescript and SvelteKit for all frontend code
- Go for the CLI, control plane and runtime (backend)
- DuckDB for OLAP on small data
- ClickHouse and Apache Druid for OLAP on big data
- Postgres for handling metadata in hosted deployments
- OpenAPI and/or gRPC for most APIs
- Docker for running dependencies like Postgres in local development
Rill uses a monorepo and you can expect to find all its code in this repository. This allows us to move faster as we can coordinate changes across multiple components in a single PR. It also gives people a single place to learn about the project and follow its development.
We want the codebase to be easy to understand and contribute to. To achieve that, every directory that contains code of non-trivial complexity should include a README.md
file that provides details about the module, such as its purpose, how to run and test it, links to relevant tutorials or docs, etc. Only the root README.md
file should be considered user-facing.
The project uses NPM for Node.js (specifically, NPM workspaces), Go modules for Go, and Maven for Java. They function well alongside each other and we have not yet found a need for a cross-language build tool.
Here's a guide to the top-level structure of the repository:
.github
contains CI/CD workflows.admin
contains the backend control plane for the managed, multi-user version of Rill.cli
contains the CLI and a server for the local frontend (used only in production).docs
contains the user-facing documentation that we deploy to docs.rilldata.com.proto
contains protocol buffer definitions for all Rill components, which notably includes our API interfaces.runtime
contains the engine (data plane) responsible for orchestrating and serving data.scripts
contains various scripts and other resources used in development.web-admin
contains the frontend control plane for the managed, multi-user version of Rill.web-auth
contains the frontend code forauth.rilldata.com
(managed with Auth0).web-common
contains common functionality shared across the local and admin frontend applications.web-local
contains the local Rill application, notably the data modeller.
Rill is comprised of multiple services that we currently support running in two configurations, local and cloud.
When running rill start
locally, the same version of the relevant services are started simultaneously in a single process. However, in cloud deployments, the relevant services are deployed individually for better isolation and scalability (for example, runtimes are provisioned dynamically when new projects are deployed).
This means that during rollout of a release, a newer version of one service may be communicating with an older version of another service, necessitating backwards compatibility. The backwards compatibility requirements are tied to the release rollout sequence, which is:
- The
admin
service is upgraded first. This means theadmin
service must be backwards compatible with both older runtime and UI versions. - Then the
runtime
s are upgraded. This means theruntime
service must be backwards compatible only with older UI versions. - Lastly the UI (
web-admin
) is upgraded. This means the UI does not need to worry about backwards compatibility.