refactor: make runtime worker state authoritative

This commit is contained in:
2026-07-28 19:21:09 +09:00
parent 6114cc9018
commit 7a1b5e97c1
11 changed files with 530 additions and 1266 deletions
+33 -2
View File
@@ -193,8 +193,36 @@ pub struct WorkerController;
impl WorkerController {
pub async fn spawn<C, St>(
worker: Worker<C, St>,
runtime_base: &Path,
) -> Result<(WorkerHandle, ShutdownReceiver), std::io::Error>
where
C: LlmClient + Clone + 'static,
St: Store + WorkerMetadataStore + Clone + Send + Sync + 'static,
{
Self::spawn_inner(worker, runtime_base, false).await
}
/// Spawn a Worker owned by `worker-runtime`.
///
/// The controller still uses an ephemeral directory for Unix sockets and
/// tool spill artifacts, but does not write legacy pid/status/manifest
/// liveness projections.
pub async fn spawn_runtime_managed<C, St>(
worker: Worker<C, St>,
runtime_base: &Path,
) -> Result<(WorkerHandle, ShutdownReceiver), std::io::Error>
where
C: LlmClient + Clone + 'static,
St: Store + WorkerMetadataStore + Clone + Send + Sync + 'static,
{
Self::spawn_inner(worker, runtime_base, true).await
}
async fn spawn_inner<C, St>(
mut worker: Worker<C, St>,
runtime_base: &Path,
runtime_managed: bool,
) -> Result<(WorkerHandle, ShutdownReceiver), std::io::Error>
where
C: LlmClient + Clone + 'static,
@@ -214,8 +242,11 @@ impl WorkerController {
// the spawn-tool factories need its socket path, and before the
// initial status/history writes consume the greeting we build
// after registration is complete.
let runtime_dir =
Arc::new(RuntimeDir::create(runtime_base, &worker.manifest().worker.name).await?);
let runtime_dir = Arc::new(if runtime_managed {
RuntimeDir::create_transient(runtime_base, &worker.manifest().worker.name).await?
} else {
RuntimeDir::create(runtime_base, &worker.manifest().worker.name).await?
});
let spawner_name = worker.manifest().worker.name.clone();
let self_parent_socket = worker.callback_socket().cloned();
+41 -1
View File
@@ -41,6 +41,7 @@ pub struct SpawnedWorkerRecord {
/// The directory is removed on drop.
pub struct RuntimeDir {
path: PathBuf,
write_legacy_snapshots: bool,
}
impl RuntimeDir {
@@ -52,7 +53,23 @@ impl RuntimeDir {
let pid = std::process::id().to_string();
fs::write(path.join("pid"), pid.as_bytes()).await?;
Ok(Self { path })
Ok(Self {
path,
write_legacy_snapshots: true,
})
}
/// Create an ephemeral Runtime-owned artifact directory.
///
/// Runtime-managed Workers keep their status in the owning Runtime and do
/// not materialize legacy pid/status/manifest projections.
pub async fn create_transient(base: &Path, worker_name: &str) -> Result<Self, io::Error> {
let path = base.join(worker_name);
fs::create_dir_all(&path).await?;
Ok(Self {
path,
write_legacy_snapshots: false,
})
}
/// Create in the default base directory resolved via
@@ -64,12 +81,18 @@ impl RuntimeDir {
/// Write status.json atomically.
pub async fn write_status(&self, state: &WorkerSharedState) -> Result<(), io::Error> {
if !self.write_legacy_snapshots {
return Ok(());
}
let content = state.status_json();
atomic_write(&self.path.join("status.json"), content.as_bytes()).await
}
/// Write manifest.toml (typically once at startup).
pub async fn write_manifest(&self, toml: &str) -> Result<(), io::Error> {
if !self.write_legacy_snapshots {
return Ok(());
}
atomic_write(&self.path.join("manifest.toml"), toml.as_bytes()).await
}
@@ -200,6 +223,23 @@ mod tests {
assert_eq!(content, "[engine]\nname = \"test\"");
}
#[tokio::test]
async fn transient_directory_does_not_write_liveness_snapshots() {
let tmp = tempfile::tempdir().unwrap();
let rt = RuntimeDir::create_transient(tmp.path(), "runtime-worker")
.await
.unwrap();
rt.write_status(&test_state()).await.unwrap();
rt.write_manifest("[worker]\nname = \"runtime-worker\"")
.await
.unwrap();
assert!(!rt.path().join("pid").exists());
assert!(!rt.path().join("status.json").exists());
assert!(!rt.path().join("manifest.toml").exists());
}
#[tokio::test]
async fn write_spawned_workers_creates_file() {
use manifest::{Permission, ScopeRule};