olai-stack-runtime

Runs trestle environments on Docker Compose. olai-stack-topology plans and renders an environment; this crate starts, stops and watches it. The trestle CLI and the desktop app share it.

• ComposeEngine: up, down, ls, ps, logs, events and network removal, as an async trait.

  • DockerCli runs the docker CLI and tells "Docker is not available" apart from "nothing is running".
  • FakeEngine (feature testing) is an in-memory stand-in for tests.

• Typed status: ProjectStatus, ServiceStatus with State and Health, EngineEvent and LogLine, parsed from the CLI's JSON output. The test fixtures are captured from real engines.

• probe_host_ports: whether a plan's host ports are free before starting it.

• managed_projects, remove_mesh_if_idle: match what Docker runs against what the caller manages.

• Instances: runs the revisions in an olai-stack-state store, each in its own compose project, slot and host label, so two revisions of one environment run side by side.

  • It starts, stops and removes instances, reports their health and events, and reconciles with Docker after a restart.
  • The rendered tree is a cache under the state directory and holds no credential.

• EffectiveEnv: an instance's variables from every layer (catalog defaults, revision, placement, client env, workspace/environment/revision vars, source overlays, instance overrides), with provenance and a masked view.

• SecretProvider: EnvProvider, plus KeychainProvider behind feature keychain. A reference to a provider that isn't configured is an error, never a fallback.

• Source: overlays of variables, endpoints and expiring credentials from outside the store. UserVars is built in.

Credentials go to Compose as process environment (UpOptions::env), never through a file.

use olai_stack_runtime::{ComposeEngine, DockerCli, UpOptions};

# async fn run() -> olai_stack_runtime::Result<()> {
let docker = DockerCli::new();
let opts = UpOptions { wait: true, ..UpOptions::default() };
docker.up(std::path::Path::new("demo/dev"), &opts).await?;
for project in docker.ls().await? {
    println!("{}: {}", project.name, project.status);
}
# Ok(())
# }

See docs/runtime.md for the captured engine behaviour and how credentials are handled.