Migrate test plugins to Rust and async-ify syscall modules

- Replace Python test plugins (echo, ai_test, claudecode) with Rust
  implementations driven by the plugin SDK
- Convert fs/network/process/timer syscall modules to async/await
- Change Network.OpenConn to return stream ConnEvent (conn_id handed
  to client in first "open" frame for Send/Close routing)
- Bind timer callbacks to plugin identity; stream_call subscriptions
  are identity-checked
- Update docs and regenerate Python SDK protobuf stubs

Verified: cargo build/test/clippy clean; echo + ai_test full syscall
suites pass; hermes & claudecode register and heartbeat; cross-plugin
bridge.call routing and auth denial verified end-to-end.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
2026-06-10 14:27:46 +08:00
parent d533d0a30e
commit 3299468df6
26 changed files with 843 additions and 814 deletions

View File

@@ -2,7 +2,8 @@ use std::collections::HashMap;
use std::sync::Arc;
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::process::Command;
use tokio::sync::{mpsc, Mutex, RwLock};
use tokio::sync::{mpsc, watch, Mutex, RwLock};
use tokio::time::Duration;
/// Per-process managed state, decoupled from the Child handle.
struct ManagedProcess {
@@ -10,8 +11,8 @@ struct ManagedProcess {
status: Arc<Mutex<String>>,
stdin_tx: Option<mpsc::Sender<Vec<u8>>>,
stdout_rx: Arc<Mutex<mpsc::Receiver<Vec<u8>>>>,
/// Notified when the process exits. Clone the receiver to wait.
exit_notify: Arc<tokio::sync::Notify>,
/// Watch receiver: becomes `true` when process exits. Clone to wait.
exit_rx: watch::Receiver<bool>,
/// Handle to abort background tasks on kill.
_tasks: Vec<tokio::task::JoinHandle<()>>,
}
@@ -19,13 +20,13 @@ struct ManagedProcess {
/// Process syscall: spawn and manage subprocesses.
/// Uses the real OS PID as the identifier.
pub struct ProcessSyscall {
procs: RwLock<HashMap<u64, Arc<ManagedProcess>>>,
procs: Arc<RwLock<HashMap<u64, Arc<ManagedProcess>>>>,
}
impl ProcessSyscall {
pub fn new() -> Self {
Self {
procs: RwLock::new(HashMap::new()),
procs: Arc::new(RwLock::new(HashMap::new())),
}
}
@@ -61,7 +62,7 @@ impl ProcessSyscall {
let child_stderr = child.stderr.take();
let status = Arc::new(Mutex::new("running".to_string()));
let exit_notify = Arc::new(tokio::sync::Notify::new());
let (exit_tx, exit_rx) = watch::channel(false);
// Stdin writer task
let (stdin_tx, mut stdin_rx) = mpsc::channel::<Vec<u8>>(64);
@@ -111,14 +112,17 @@ impl ProcessSyscall {
}
});
// Reaper task: waits for exit and updates status
// Reaper task: waits for exit, updates status, signals watchers, delayed cleanup
let status_clone = status.clone();
let exit_notify_clone = exit_notify.clone();
let procs_clone = Arc::clone(&self.procs);
let reaper_task = tokio::spawn(async move {
let exit_status = child.wait().await;
let code = exit_status.map(|s| s.code().unwrap_or(-1)).unwrap_or(-1);
*status_clone.lock().await = format!("exited({})", code);
exit_notify_clone.notify_waiters();
let _ = exit_tx.send(true);
// Delayed cleanup: give stdout readers time to drain buffered data
tokio::time::sleep(Duration::from_secs(5)).await;
procs_clone.write().await.remove(&pid);
});
let managed = Arc::new(ManagedProcess {
@@ -126,7 +130,7 @@ impl ProcessSyscall {
status,
stdin_tx: Some(stdin_tx),
stdout_rx: Arc::new(Mutex::new(stdout_rx)),
exit_notify,
exit_rx,
_tasks: vec![stdin_task, stdout_task, stderr_task, reaper_task],
});
@@ -136,19 +140,25 @@ impl ProcessSyscall {
}
pub async fn kill(&self, pid: u64) -> Result<(), String> {
let procs = self.procs.read().await;
let proc = procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone();
drop(procs);
let proc = {
let procs = self.procs.read().await;
procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone()
};
// Set status to killed first
*proc.status.lock().await = "killed".to_string();
// Abort all background tasks (stdin writer, stdout reader, stderr drain, reaper)
// Aborting reaper drops exit_tx, so any wait() blocked on changed() will wake with Err.
for task in proc._tasks.iter() {
task.abort();
}
*proc.status.lock().await = "killed".to_string();
proc.exit_notify.notify_waiters();
// Remove from map
self.procs.write().await.remove(&pid);
// Also send OS kill
#[cfg(unix)]
@@ -161,46 +171,41 @@ impl ProcessSyscall {
// On Windows, aborting the reaper task triggers kill_on_drop on the Child
}
self.procs.write().await.remove(&pid);
tracing::info!("process killed: pid={}", pid);
Ok(())
}
pub async fn wait(&self, pid: u64) -> Result<i32, String> {
let procs = self.procs.read().await;
let proc = procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone();
drop(procs);
let proc = {
let procs = self.procs.read().await;
procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone()
};
// Check if already exited before waiting
{
let status = proc.status.lock().await;
if status.starts_with("exited") || *status == "killed" {
let s = status.clone();
drop(status);
self.procs.write().await.remove(&pid);
return parse_exit_code(&s);
// Use watch channel: no race between checking status and waiting.
// watch retains the latest value, so even if send(true) happened before
// we subscribe, borrow_and_update() will see `true` immediately.
let mut rx = proc.exit_rx.clone();
while !*rx.borrow_and_update() {
if rx.changed().await.is_err() {
break; // sender dropped (e.g. kill aborted reaper)
}
}
// Wait for exit notification
proc.exit_notify.notified().await;
let status = proc.status.lock().await.clone();
// Clean up from map
self.procs.write().await.remove(&pid);
parse_exit_code(&status)
}
pub async fn write_stdin(&self, pid: u64, data: &[u8]) -> Result<(), String> {
let procs = self.procs.read().await;
let proc = procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone();
drop(procs);
let proc = {
let procs = self.procs.read().await;
procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone()
};
if let Some(ref tx) = proc.stdin_tx {
tx.send(data.to_vec())
@@ -212,12 +217,13 @@ impl ProcessSyscall {
}
pub async fn read_stdout(&self, pid: u64, buf: &mut [u8]) -> Result<usize, String> {
let procs = self.procs.read().await;
let proc = procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone();
drop(procs);
let proc = {
let procs = self.procs.read().await;
procs
.get(&pid)
.ok_or_else(|| format!("pid {} not found", pid))?
.clone()
};
let mut rx = proc.stdout_rx.lock().await;
match rx.recv().await {
@@ -268,9 +274,13 @@ impl ProcessSyscall {
}
pub async fn list(&self) -> Vec<(u64, String, String)> {
let procs = self.procs.read().await;
let mut result = Vec::with_capacity(procs.len());
for (pid, mp) in procs.iter() {
let snapshot: Vec<(u64, Arc<ManagedProcess>)> = {
let procs = self.procs.read().await;
procs.iter().map(|(pid, mp)| (*pid, mp.clone())).collect()
};
// procs read lock is dropped here
let mut result = Vec::with_capacity(snapshot.len());
for (pid, mp) in &snapshot {
let status = mp.status.lock().await.clone();
result.push((*pid, mp.cmd.clone(), status));
}
@@ -288,3 +298,4 @@ fn parse_exit_code(status: &str) -> Result<i32, String> {
Ok(-1)
}
}