fix: 修正Core故障退避与熔断计数

frontend/src-tauri/src/core.rs

@@ -208,6 +208,25 @@ pub struct RequestSession {
 }
 
 impl CoreSupervisor {
+    fn record_failure(&mut self, now: Instant) {
+        self.attempts
+            .retain(|time| now.duration_since(*time) < Duration::from_secs(300));
+        self.attempts.push_back(now);
+        self.next_attempt = Some(now + Duration::from_secs(1 << (self.attempts.len() - 1).min(4)));
+    }
+
+    fn reap_failed_session(&mut self) -> bool {
+        let failed = self
+            .session
+            .as_mut()
+            .is_some_and(|session| !matches!(session.child.try_wait(), Ok(None)));
+        if failed {
+            self.session.take();
+            self.record_failure(Instant::now());
+        }
+        failed
+    }
+
     pub fn new(
         executable: PathBuf,
         arguments: Vec<String>,
@@ -238,9 +257,17 @@ impl CoreSupervisor {
     }
 
     pub fn available(&mut self) -> bool {
-        self.session
+        self.reap_failed_session();
-            .as_mut()
+        self.session.is_some()
-            .is_some_and(|s| matches!(s.child.try_wait(), Ok(None)))
+    }
+
+    /// 返回受 Host 管理的 Core 启动进程 ID，用于诊断和故障注入。
+    pub fn process_id(&mut self) -> Option<u32> {
+        if self.available() {
+            self.session.as_ref().map(|session| session.child.id())
+        } else {
+            None
+        }
     }
 
     pub fn request_session(&mut self, path: &str) -> Result<RequestSession> {
@@ -260,18 +287,15 @@ impl CoreSupervisor {
         if self.available() {
             return Ok(());
         }
-        self.session.take();
         let now = Instant::now();
         self.attempts
             .retain(|t| now.duration_since(*t) < Duration::from_secs(300));
-        if self.attempts.len() >= 5 {
+        if self.attempts.len() > 5 {
             return Err("CORE_RESTART_LIMIT".into());
         }
         if self.next_attempt.is_some_and(|t| now < t) {
             return Err("CORE_RESTART_BACKOFF".into());
         }
-        self.attempts.push_back(now);
-        self.next_attempt = Some(now + Duration::from_secs(1 << (self.attempts.len() - 1)));
         if let Some(manifest) = &self.bundle_manifest {
             verify_bundle(&self.working_dir, manifest)?;
         }
@@ -281,8 +305,10 @@ impl CoreSupervisor {
             &self.working_dir,
             &self.data_dir,
             self.broker.clone(),
-        )?;
+        )
+        .inspect_err(|_| self.record_failure(Instant::now()))?;
         self.session = Some(session);
+        self.next_attempt = None;
         Ok(())
     }
 

frontend/src-tauri/tests/core_process.rs

@@ -2,9 +2,29 @@
 use notesagent_host::core::CoreSupervisor;
 use notesagent_host::credentials::{CredentialBroker, CredentialId, Scope};
 use std::path::Path;
+#[cfg(windows)]
+use std::process::Command;
 use std::sync::{Arc, Mutex};
+#[cfg(windows)]
+use std::time::{Duration, Instant};
 use zeroize::Zeroizing;
 
+#[cfg(windows)]
+fn file_sha256(path: &Path) -> String {
+    use sha2::{Digest, Sha256};
+    let bytes = std::fs::read(path).unwrap();
+    format!("{:x}", Sha256::digest(bytes))
+}
+
+#[cfg(windows)]
+fn terminate_process_tree(pid: u32) {
+    let status = Command::new("taskkill")
+        .args(["/PID", &pid.to_string(), "/T", "/F"])
+        .status()
+        .unwrap();
+    assert!(status.success(), "无法终止 Core 进程树 {pid}");
+}
+
 #[test]
 fn real_python_core_authenticates_and_rotates_generation() {
     let backend = Path::new(env!("CARGO_MANIFEST_DIR"))
@@ -48,6 +68,64 @@ fn real_python_core_authenticates_and_rotates_generation() {
     assert!(std::net::TcpStream::connect(endpoint).is_err());
 }
 
+#[cfg(windows)]
+#[test]
+fn six_real_core_crashes_back_off_then_open_the_circuit_without_touching_local_edits() {
+    let backend = Path::new(env!("CARGO_MANIFEST_DIR"))
+        .join("../../backend")
+        .canonicalize()
+        .unwrap();
+    let python = backend.join(".venv/Scripts/python.exe");
+    assert!(python.is_file(), "需要已锁定的后端虚拟环境");
+    let temp = tempfile::tempdir().unwrap();
+    let note = temp.path().join("local-edit.md");
+    std::fs::write(&note, "Core 故障期间仍由 Host 保存的本地修改。\n").unwrap();
+    let expected_hash = file_sha256(&note);
+    let mut core = CoreSupervisor::new(
+        python,
+        vec!["-m".into(), "app.sidecar".into()],
+        backend,
+        temp.path().join("core"),
+    );
+
+    let first = core.request_session("/health").unwrap();
+    let first_endpoint = first
+        .url
+        .trim_start_matches("http://")
+        .trim_end_matches("/health")
+        .to_string();
+    for crash in 0..6 {
+        terminate_process_tree(core.process_id().unwrap());
+        let stopped = Instant::now();
+        while core.available() {
+            assert!(stopped.elapsed() < Duration::from_secs(2));
+            std::thread::sleep(Duration::from_millis(10));
+        }
+        assert_eq!(file_sha256(&note), expected_hash);
+        if crash == 5 {
+            assert_eq!(
+                core.request_session("/health").err().unwrap(),
+                "CORE_RESTART_LIMIT"
+            );
+            break;
+        }
+        let expected_delay = Duration::from_secs(1 << crash);
+        assert_eq!(
+            core.request_session("/health").err().unwrap(),
+            "CORE_RESTART_BACKOFF"
+        );
+        while stopped.elapsed() < expected_delay {
+            std::thread::sleep(Duration::from_millis(20));
+        }
+        core.request_session("/health").unwrap();
+    }
+
+    let shutdown = Instant::now();
+    drop(core);
+    assert!(shutdown.elapsed() < Duration::from_secs(10));
+    assert!(std::net::TcpStream::connect(first_endpoint).is_err());
+}
+
 #[test]
 fn real_core_credential_api_uses_host_stronghold_without_plaintext_response() {
     use std::io::{Read, Write};