Oxide Control Plane

Table of Contents

• Documentation
• Build and run
  • Run tests with nextest
  • rustfmt and clippy
• Working in Omicron
  • Key tips
  • Adding a new system library dependency
  • Rust packages in Omicron
  • Workspace management
  • Why am I getting compile errors after I thought I’d already built everything?
  • Generated Service Clients and Updating
  • Resolving merge conflicts in Cargo.lock
  • Configuring ClickHouse

This repo houses the work-in-progress Oxide Rack control plane.

Omicron is open-source. But we’re pretty focused on our own goals for the foreseeable future and not able to help external contributors. Please see CONTRIBUTING.md for more information.

Documentation

Docs are automatically generated for the public (externally-facing) API based on the OpenAPI spec that itself is automatically generated from the server implementation. You can generate your own docs for either the public API or any of the internal APIs by feeding the corresponding OpenAPI specs (in ./openapi) into an OpenAPI doc generator.

There are some internal design docs in the ./docs directory.

For more design documentation and internal Rust API docs, see the generated Rust documentation. You can generate this yourself with:

$ cargo doc --document-private-items

Note that --document-private-items is configured by default, so you can actually just use cargo doc.

Folks with access to Oxide RFDs may find RFD 48 ("Control Plane Requirements") and other control plane RFDs relevant. These are not currently publicly available.

Build and run

Omicron has two modes of operation: "simulated" and "non-simulated".

The simulated version of Omicron allows the high-level control plane logic to run without actually managing any sled-local resources. This version can be executed on Linux, Mac, and illumos. This mode of operation is provided for development and testing only.

To build and run the simulated version of Omicron, see: docs/how-to-run-simulated.adoc.

The non-simulated version of Omicron actually manages sled-local resources, and may only be executed on hosts running Helios. This mode of operation will be used in production.

To build and run the non-simulated version of Omicron, see: docs/how-to-run.adoc.

Run tests with nextest

The supported way to run tests is via cargo-nextest.

Note	cargo test will not work for many of our tests, since they rely on nextest-specific features.

If you don’t already have nextest installed, get started by downloading a pre-built binary or installing nextest via your package manager. Nextest has pre-built binaries for Linux, macOS and illumos.

Then, run tests with:

$ cargo nextest run

Nextest does not support doctests. Run doctests separately with cargo test --doc.

Similarly, you can run tests inside a Falcon based VM. This is described in the test-utils README.

rustfmt and clippy

You can format the code using cargo fmt. Make sure to run this before pushing changes. The CI checks that the code is correctly formatted.

You can run the Clippy linter using cargo xtask clippy. CI checks that code is clippy-clean.

Working in Omicron

Omicron is a pretty large repo containing a bunch of related components. (Why? See docs/repo.adoc.) If you just build the whole thing with cargo build or cargo nextest run, it can take a while, even for incremental builds. Since most people are only working on a few of these components at a time, it’s helpful to be know about Cargo’s tools for working with individual packages in a workspace.

Note	This section assumes you’re already familiar with the prerequisites and environment setup needed to do any work on Omicron. See docs/how-to-run-simulated.adoc or docs/how-to-run.adoc for more on that.

Key tips

• Use cargo check when you just want to know if your code compiles. It’s much faster than cargo build or cargo nextest run.

• When using Cargo’s check/build/test/clippy commands, you can use the -p PACKAGE flag to only operate on a specific package. This often saves a lot of time for incremental builds.

• When using Cargo’s check/build/clippy commands, use --all-targets to make sure you’re checking or building the test code, too.

These are explained a bit more below, along with some common pitfalls.

Here’s an example workflow. Suppose you’re working on some changes to the Nexus database model (nexus-db-model package, located at nexus/db-model from the root). While you’re actively writing and checking code, you might run:

cargo check --all-targets

without any -p flag. Running this incrementally is pretty fast even on the whole workspace. This also uncovers places where your changes have broken code that uses this package. (If you’re making big changes, you might not want that right away. In that case, you might choose to use -p nexus-db-model here.)

When you’re ready to test the changes you’ve made, start with building and running tests for the most specific package you’ve changed:

cargo nextest run -p nexus-db-model

Once that works, check the tests for the next package up:

cargo nextest run -p omicron-nexus

When you’re happy with things and want to make sure you haven’t missed something, test everything:

cargo nextest run

Adding a new system library dependency

We check that certain system library dependencies are not leaked outside of their intended binaries via cargo xtask verify-libraries in CI. If you are adding a new dependency on a illumos/helios library it is recommended that you update .cargo/xtask.toml with an allow list of where you expect the dependency to show up. For example some libraries such as libnvme.so.1 are only available in the global zone and therefore will not be present in any other zone. This check is here to help us catch any leakage before we go to deploy on a rack. You can inspect a compiled binary in the target directory for what it requires by using elfedit - for example elfedit -r -e 'dyn:tag NEEDED' /path/to/omicron/target/debug/sled-agent.

Rust packages in Omicron

Note

The term "package" is overloaded: most programming languages and operating systems have their own definitions of a package. On top of that, Omicron bundles up components into our own kind of "package" that gets delivered via the install and update systems. These are described in the package-manifest.toml file in the root of the repo. In this section, we’re just concerned with Rust packages.

Note

There’s also confusion in the Rust world about the terms "packages" and "crates". Packages are the things that have a Cargo.toml file. (Workspaces like Omicron itself have Cargo.toml files, too.) Packages are also the things that you publish to crates.io (confusingly). One package might have a library, a standalone executable binary, several examples, integration tests, etc. that are all compiled individually and produce separate artifacts. These are what Rust calls crates. We’re generally just concerned with packages here, not crates.

Here are some of the big components in the control plane that live in this repo:

Main rust package	Component	Description
omicron-nexus	Nexus	Service responsible for handling external API requests and orchestrating the rest of the control plane.
omicron-sled-agent	Sled Agent	Service that runs on each compute sled (server) to manage resources on that Sled
dns-server	Internal DNS server, External DNS server	DNS server component used for both internal service discovery and external DNS
omicron-gateway	Management Gateway Service	Connects Nexus (and other control plane services) to services on the rack management network (e.g., service processors)
oximeter/oximeter	Oximeter	Collects telemetry from other services and stores it into Clickhouse
wicket/wicketd	Wicket	CLI interface made available to operators on the rack technician port for rack setup and recovery

For those with access to Oxide RFDs, RFD 61 discusses the organization principles and key components in more detail.

Many of these components themselves are made up of other packages (e.g., nexus-db-model is under omicron-nexus). There are also many more top-level packages than what’s mentioned above. These are used for common code, clients, tools, etc. For more, see the Rustdoc for each module. (Where docs are missing or incomplete, please contribute!)

Use Cargo’s -p PACKAGE to check/build/test only the package you’re working on. Since people are usually only working on one or two components at a time, you can usually iterate faster this way.

Workspace management

Omicron uses cargo-hakari to ensure that all workspace dependencies enable the same set of features. This dramatically improves compilation time when switching between different subsets of packages (e.g. -p wicket or -p nexus-db-model), because the sets of enabled features remain consistent.

cargo hakari status is checked in CI; if the CI check fails, then update the configuration locally with

cargo install cargo-hakari --locked # only needed on the first run
cargo hakari generate
cargo hakari manage-deps

Why am I getting compile errors after I thought I’d already built everything?

Say you’re iterating on code, running cargo build -p nexus-db-model to build just that package. You work through lots of compiler errors until finally it works. Now you run tests: cargo nextest run -p nexus-db-model. Now you see a bunch of compiler errors again! What gives?

By default, Cargo does not operate on the tests. Cargo’s check/build/clippy commands ignore them. This is another reason we suggest using --all-targets most of the time.

Generated Service Clients and Updating

Each service is a Dropshot server that presents an HTTP API. The description of that API is serialized as an OpenAPI document which we store in omicron/openapi and check in to this repo. In order to ensure that changes to those APIs are made intentionally, each service contains a test that validates that the current API matches. This allows us 1. to catch accidental changes as test failures and 2. to explicitly observe API changes during code review (and in the git history).

We also use these OpenAPI documents as the source for the clients we generate using Progenitor. Clients are automatically updated when the coresponding OpenAPI document is modified.

Note that Omicron contains a nominally circular dependency:

• Nexus depends on the Sled Agent client

• The Sled Agent client is derived from the OpenAPI document emitted by Sled Agent

• Sled Agent depends on the Nexus client

• The Nexus client is derived from the OpenAPI document emitted by Nexus

We effectively "break" this circular dependency by virtue of the OpenAPI documents being checked in.

In general, changes any service API require the following set of build steps:

1. Make changes to the service API.

2. Update the OpenAPI document by running the relevant test with overwrite set: EXPECTORATE=overwrite cargo nextest run -p <package> — test_nexus_openapi_internal (changing the package name and test name as necessary). It’s important to do this before the next step.

3. This will cause the generated client to be updated which may break the build for dependent consumers.

4. Modify any dependent services to fix calls to the generated client.

Note that if you make changes to both Nexus and Sled Agent simultaneously, you may end up in a spot where neither can build and therefore neither OpenAPI document can be generated. In this case, revert or comment out changes in one so that the OpenAPI document can be generated.

This is a particular problem if you find yourself resolving merge conflicts in the generated files. You have basically two options for this:

• Resolve the merge conflicts by hand. This is usually not too bad in practice.

• Take the upstream copy of the file, back out your client side changes (git stash and its -p option can be helpful for this), follow the steps above to regenerate the file using the automated test, and finally re-apply your changes to the client side. This is essentially getting yourself back to step 1 above and then following the procedure above.

Resolving merge conflicts in Cargo.lock

When pulling in new changes from upstream "main", you may find conflicts in Cargo.lock. The easiest way to deal with these is usually to take the upstream changes as-is, then trigger any Cargo operation that updates the lockfile. cargo metadata is a quick one. Here’s an example:

# Pull in changes from upstream "main"
$ git fetch
$ git merge origin/main

# Oh no!  We've got conflicts in Cargo.lock.  First, let's just take what's upstream:
$ git show origin/main:Cargo.lock > Cargo.lock

# Now, run any command that causes Cargo to update the lock file as needed.
$ cargo metadata > /dev/null

When you do this, Cargo makes only changes to Cargo.lock that are necessary based on the various Cargo.toml files in the workspace and dependencies.

Here are things you don’t want to do to resolve this conflict:

• Run cargo generate-lockfile to generate a new lock file from scratch.

• Remove Cargo.lock and let Cargo regenerate it from scratch.

Both of these will cause Cargo to make many more changes (relative to "main") than necessary because it’s choosing the latest version of all dependencies in the whole tree. You’ll be inadvertently updating all of Omicron’s transitive dependencies. (You might conceivably want that. But usually we update dependencies either as-needed for a particular change or via individual PRs via dependabot, not all at once because someone had to merge Cargo.lock.)

You can also resolve conflicts by hand. It’s tedious and error-prone.

Configuring ClickHouse

The ClickHouse binary uses several sources for its configuration. The binary expects an XML config file, usually named config.xml to be available, or one may be specified with the -C command-line flag. The binary also includes a minimal configuration embedded within it, which will be used if no configuration file is given or present in the current directory. The server also accepts command-line flags for overriding the values of the configuration parameters.

The packages downloaded by ci_download_clickhouse include a config.xml file with them. You should probably run ClickHouse via the omicron-dev tool, but if you decide to run it manually, you can start the server with:

$ /path/to/clickhouse server --config-file /path/to/config.xml

The configuration file contains a large number of parameters, but most of them are described with comments in the included config.xml, or you may learn more about them here and here. Parameters may be updated in the config.xml, and the server will automatically reload them. You may also specify many of them on the command-line with:

$ /path/to/clickhouse server --config-file /path/to/config.xml -- --param_name param_value ...