feat: server start 接通端口旗标与 well-known TCP 端点

This commit is contained in:
2026-10-11 13:34:45 +00:00
parent b40628fed7
commit be565bb945
12 changed files with 408 additions and 265 deletions
+51 -36
View File
@@ -44,16 +44,16 @@ pub enum LivenessOutcome {
KillFailed,
}
/// 启动时 liveness 扫描入口:扫 sidecar 与 socket,清理陈旧残留,僵尸态自动恢复。
/// 启动时 liveness 扫描入口:扫 sidecar 与 endpoint,清理陈旧残留,僵尸态自动恢复。
///
/// # 步骤
/// 1. 采集五个探活信号(sidecar 可解析性、socket 存活、socket 文件存在、
/// 1. 采集五个探活信号(sidecar 可解析性、endpoint 存活、sidecar 文件存在、
/// PID 存活、刻度匹配)。
/// 2. socket 存活守卫:存活即无操作(自动扫描绝不杀存活 server)。
/// 2. endpoint 存活守卫:存活即无操作(自动扫描绝不杀存活 server)。
/// 3. 按 `decide` 裁决表分派:无操作 / 只删文件 / 僵尸分支 / 拒绝。
///
/// # 边界与前提
/// - 只删 `config` 注入的 sidecar 与 socket 两条路径;日志文件不碰。
/// - 只删 `config` 注入的 sidecar 与 socket 两条路径(Unix 臂);日志文件不碰。
/// - 杀进程等待取缺省 10 秒(长于 server 的在途排水安全超时 5 秒)。
///
/// # 副作用
@@ -61,20 +61,21 @@ pub enum LivenessOutcome {
///
/// # 注意事项
/// - 须在 `claim_sidecar` 之前调用(认领裁决前清掉陈旧状态)。
pub fn liveness_scan(sidecar_path: &Path, socket_path: &Path) -> LivenessOutcome {
liveness_scan_with_waits(sidecar_path, socket_path, ZOMBIE_EXIT_WINDOW, cleanup::DEFAULT_KILL_WAIT)
pub fn liveness_scan(sidecar_path: &Path, endpoint: &Endpoint) -> LivenessOutcome {
liveness_scan_with_waits(sidecar_path, endpoint, ZOMBIE_EXIT_WINDOW, cleanup::DEFAULT_KILL_WAIT)
}
/// 启动时 liveness 扫描(可注入两个超时,测试注入短窗口)。
///
/// # 步骤
/// 1. 采集信号:`read_identity`(缺失与损坏同归 `None`)、socket 文件存在、
/// `is_server_alive`(仅文件存在时探活)、`pid_alive`、`process_starttime`
/// 与 sidecar 记录值比对(PID 死时匹配位恒假)。
/// 2. socket 存活守卫:存活即无操作。
/// 1. 采集信号:`read_identity`(缺失与损坏同归 `None`)、sidecar 文件存在、
/// `is_server_alive`(探 sidecar 记录的 endpoint,不可读时探调用方
/// endpoint 兜底)、`pid_alive`、`process_starttime` 与 sidecar 记录值比对
/// (PID 死时匹配位恒假)。
/// 2. endpoint 存活守卫:存活即无操作。
/// 3. 按 `decide` 裁决表分派:
/// - `None` 无操作。
/// - `DeleteOnly` 删文件(socket 先、sidecar 后)。
/// - `None` 无操作(Unix 臂 sidecar 缺失而 socket 文件存在时按孤儿删除)。
/// - `DeleteOnly` 删文件(socket 先、sidecar 后,Unix 臂)。
/// - `Kill` 僵尸分支:有界等待 sidecar 消失,仍在则杀进程后删文件。
/// - `Refuse` 无操作(文件不碰,归 `claim_sidecar` 裁决)。
///
@@ -88,13 +89,18 @@ pub fn liveness_scan(sidecar_path: &Path, socket_path: &Path) -> LivenessOutcome
/// - 可能向 sidecar 记录的进程发 SIGTERM;可能删除 socket 与 sidecar 文件。
/// - 杀进程决策与失败记入跨重启保留的日志文件(守护形态 stderr 是 null)。
pub fn liveness_scan_with_waits(
sidecar_path: &Path, socket_path: &Path, zombie_exit_window: Duration, kill_wait: Duration,
sidecar_path: &Path, endpoint: &Endpoint, zombie_exit_window: Duration, kill_wait: Duration,
) -> LivenessOutcome {
// 采集信号:sidecar 可解析性(缺失与损坏同归 None)、socket 存活、
// PID 存活与刻度匹配(PID 死时匹配位恒假)。
// 采集信号:sidecar 可解析性(缺失与损坏同归 None)、endpoint 存活、
// PID 存活与刻度匹配(PID 死时匹配位恒假)。endpoint 探活目标取 sidecar
// 记录的 endpoint(正确性红线),不可读时取调用方 endpoint(防御性兜底)。
let identity = sidecar::read_identity(sidecar_path);
let socket_exists = socket_path.exists();
let socket_alive = socket_exists && sidecar::is_server_alive(&Endpoint::Unix(socket_path.to_path_buf()));
let registry_exists = sidecar_path.exists();
let probe = match &identity {
Some(id) => sidecar::parse_endpoint(&id.socket_path),
None => endpoint.clone(),
};
let endpoint_alive = sidecar::is_server_alive(&probe);
let (pid_alive, starttime_match) = match &identity {
Some(id) => {
let alive = sidecar::pid_alive(id.pid);
@@ -103,19 +109,28 @@ pub fn liveness_scan_with_waits(
None => (false, false),
};
// socket 存活守卫:存活 server 的处置归 `claim_sidecar` 裁决(同版本
// endpoint 存活守卫:存活 server 的处置归 `claim_sidecar` 裁决(同版本
// AlreadyRunning 拒绝、异版本 VersionMismatch 走交接)。自动扫描绝不杀
// 存活 server:`decide` 表的 Kill 行人工清理语义,杀存活 server 等于抢占,
// 违反单例交接协议。
if socket_alive {
if endpoint_alive {
return LivenessOutcome::Noop;
}
// 按裁决表分派。
match cleanup::decide(identity.as_ref(), socket_exists, socket_alive, pid_alive, starttime_match) {
Decision::None => LivenessOutcome::Noop,
match cleanup::decide(identity.as_ref(), registry_exists, endpoint_alive, pid_alive, starttime_match) {
Decision::None => {
// 遗留信号:Unix 臂 sidecar 缺失而 socket 文件存在,按孤儿删除。
if let Some(path) = endpoint.socket_path() {
if !registry_exists && path.exists() {
let _ = std::fs::remove_file(path);
return LivenessOutcome::Cleaned;
}
}
LivenessOutcome::Noop
}
Decision::DeleteOnly => {
cleanup::remove_files(sidecar_path, socket_path);
cleanup::remove_files(sidecar_path, endpoint);
LivenessOutcome::Cleaned
}
Decision::Kill => {
@@ -124,7 +139,7 @@ pub fn liveness_scan_with_waits(
let Some(id) = identity else {
return LivenessOutcome::Noop;
};
kill_zombie(&id, sidecar_path, socket_path, zombie_exit_window, kill_wait)
kill_zombie(&id, sidecar_path, endpoint, zombie_exit_window, kill_wait)
}
Decision::Refuse => LivenessOutcome::Noop,
}
@@ -148,7 +163,7 @@ pub fn liveness_scan_with_waits(
/// - 可能向 sidecar 记录的进程发 SIGTERM;可能删除 socket 与 sidecar 文件。
/// - 杀进程决策与失败记入跨重启保留的日志文件。
fn kill_zombie(
id: &sidecar::ServerIdentity, sidecar_path: &Path, socket_path: &Path, zombie_exit_window: Duration,
id: &sidecar::ServerIdentity, sidecar_path: &Path, endpoint: &Endpoint, zombie_exit_window: Duration,
kill_wait: Duration,
) -> LivenessOutcome {
// 有界等待 sidecar 消失:覆盖旧 server 退出序列窗口(排水 5 秒,sidecar
@@ -184,7 +199,7 @@ fn kill_zombie(
}
// 进程已退出:删文件(尽力而为;旧 server 的退出序列自己删文件,此处兜底)。
cleanup::remove_files(sidecar_path, socket_path);
cleanup::remove_files(sidecar_path, endpoint);
oxide_log::__log_event!(
"oxide::kernel",
oxide_log::tracing::Level::INFO,
@@ -250,7 +265,7 @@ mod tests {
#[test]
fn no_residue_noop() {
let dir = TestDir::new();
let outcome = liveness_scan(&dir.path("sidecar.json"), &dir.path("server.sock"));
let outcome = liveness_scan(&dir.path("sidecar.json"), &Endpoint::Unix(dir.path("server.sock")));
assert_eq!(outcome, LivenessOutcome::Noop, "无文件应判无操作");
}
@@ -261,7 +276,7 @@ mod tests {
let sidecar = dir.path("sidecar.json");
let socket = dir.path("server.sock");
fs::write(&socket, b"stale").expect("写陈旧 socket 文件应成功");
let outcome = liveness_scan(&sidecar, &socket);
let outcome = liveness_scan(&sidecar, &Endpoint::Unix(socket.clone()));
assert_eq!(outcome, LivenessOutcome::Cleaned, "孤儿 socket 文件应判已清理");
assert!(!socket.exists(), "socket 文件应被删除");
assert!(!sidecar.exists(), "sidecar 不应被创建");
@@ -274,7 +289,7 @@ mod tests {
let sidecar = dir.path("sidecar.json");
let socket = dir.path("server.sock");
sidecar::write_exclusive(&test_identity(dead_pid(), 0, &socket), &sidecar).expect("写 sidecar 应成功");
let outcome = liveness_scan(&sidecar, &socket);
let outcome = liveness_scan(&sidecar, &Endpoint::Unix(socket.clone()));
assert_eq!(outcome, LivenessOutcome::Cleaned, "死 PID 残留应判已清理");
assert!(!sidecar.exists(), "sidecar 应被删除");
assert!(!socket.exists(), "socket 文件应保持不存在");
@@ -288,7 +303,7 @@ mod tests {
let sidecar = dir.path("sidecar.json");
let socket = dir.path("server.sock");
sidecar::write_exclusive(&test_identity(std::process::id(), 0, &socket), &sidecar).expect("写 sidecar 应成功");
let outcome = liveness_scan(&sidecar, &socket);
let outcome = liveness_scan(&sidecar, &Endpoint::Unix(socket.clone()));
assert_eq!(outcome, LivenessOutcome::Cleaned, "刻度失配残留应判已清理");
assert!(!sidecar.exists(), "sidecar 应被删除");
assert!(sidecar::pid_alive(std::process::id()), "本进程应不受影响");
@@ -302,7 +317,7 @@ mod tests {
let socket = dir.path("server.sock");
fs::write(&sidecar, "not json at all").expect("写损坏文件应成功");
fs::write(&socket, b"stale").expect("写陈旧 socket 文件应成功");
let outcome = liveness_scan(&sidecar, &socket);
let outcome = liveness_scan(&sidecar, &Endpoint::Unix(socket.clone()));
assert_eq!(outcome, LivenessOutcome::Cleaned, "损坏 sidecar 加陈旧 socket 应判已清理");
assert!(!sidecar.exists(), "sidecar 应被删除");
assert!(!socket.exists(), "socket 文件应被删除");
@@ -316,9 +331,9 @@ mod tests {
let sidecar = dir.path("sidecar.json");
let socket = dir.path("server.sock");
let listener = UnixListener::bind(&socket).expect("绑定监听者应成功");
sidecar::write_exclusive(&sidecar::ServerIdentity::new(socket.to_str().unwrap(), "0.0.0"), &sidecar)
sidecar::write_exclusive(&sidecar::ServerIdentity::new(&Endpoint::Unix(socket.clone()), "0.0.0"), &sidecar)
.expect("写 sidecar 应成功");
let outcome = liveness_scan(&sidecar, &socket);
let outcome = liveness_scan(&sidecar, &Endpoint::Unix(socket.clone()));
assert_eq!(outcome, LivenessOutcome::Noop, "存活 socket 应判无操作");
assert!(sidecar.exists(), "sidecar 应原样");
assert!(socket.exists(), "socket 文件应原样");
@@ -333,13 +348,13 @@ mod tests {
let sidecar = dir.path("sidecar.json");
let socket = dir.path("server.sock");
// sidecar 记本进程 PID 加真实刻度(进程活且刻度匹配)、无 socket。
let id = sidecar::ServerIdentity::new(socket.to_str().unwrap(), "0.0.0");
let id = sidecar::ServerIdentity::new(&Endpoint::Unix(socket.clone()), "0.0.0");
sidecar::write_exclusive(&id, &sidecar).expect("写 sidecar 应成功");
// 扫描线程:sidecar 存在时进入僵尸分支,有界等待 sidecar 消失。
let sidecar_scan = sidecar.clone();
let socket_scan = socket.clone();
let scan_handle = std::thread::spawn(move || {
liveness_scan_with_waits(&sidecar_scan, &socket_scan, Duration::from_secs(5), Duration::from_secs(1))
liveness_scan_with_waits(&sidecar_scan, &Endpoint::Unix(socket_scan), Duration::from_secs(5), Duration::from_secs(1))
});
// 主线程在窗口内删 sidecar(模拟旧 server 退出序列),扫描应判无操作。
std::thread::sleep(Duration::from_millis(200));
@@ -366,7 +381,7 @@ mod tests {
sidecar::write_exclusive(&id, &sidecar).expect("写 sidecar 应成功");
// 注入短僵尸退出窗口(500 毫秒)加短杀进程等待(3 秒):子进程存活,
// sidecar 不消失,扫描等窗口后杀进程。
let outcome = liveness_scan_with_waits(&sidecar, &socket, Duration::from_millis(500), Duration::from_secs(3));
let outcome = liveness_scan_with_waits(&sidecar, &Endpoint::Unix(socket.clone()), Duration::from_millis(500), Duration::from_secs(3));
assert_eq!(outcome, LivenessOutcome::Cleaned, "刻度匹配的存活进程应被杀且文件被删");
// 有界等待回收线程:子进程被杀后收尸即退出,不无限阻塞。
let deadline = Instant::now() + Duration::from_secs(5);
@@ -396,7 +411,7 @@ mod tests {
// 注入短僵尸退出窗口(500 毫秒)加短杀进程等待(300 毫秒):子进程忽略
// SIGTERM,扫描判杀进程失败,文件原样。
let outcome =
liveness_scan_with_waits(&sidecar, &socket, Duration::from_millis(500), Duration::from_millis(300));
liveness_scan_with_waits(&sidecar, &Endpoint::Unix(socket.clone()), Duration::from_millis(500), Duration::from_millis(300));
assert_eq!(outcome, LivenessOutcome::KillFailed, "忽略 SIGTERM 应判杀进程失败");
assert!(sidecar.exists(), "sidecar 应原样");
assert!(!socket.exists(), "socket 文件应保持不存在");