feat: server start 改守护形态,spawn 前台子进程加就绪探测

This commit is contained in:
2026-10-06 18:07:35 +00:00
parent 7eab58a133
commit 176c7dffe1
4 changed files with 181 additions and 77 deletions
Generated
+18 -18
View File
@@ -1160,7 +1160,7 @@ dependencies = [
[[package]]
name = "oxide_builtins"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"chrono",
"fast-float",
@@ -1185,7 +1185,7 @@ dependencies = [
[[package]]
name = "oxide_bytecode"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"num-bigint",
"oxide_log",
@@ -1193,7 +1193,7 @@ dependencies = [
[[package]]
name = "oxide_cfg"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_emit",
@@ -1204,7 +1204,7 @@ dependencies = [
[[package]]
name = "oxide_cli"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"ansi_term",
"bumpalo",
@@ -1225,7 +1225,7 @@ dependencies = [
[[package]]
name = "oxide_code_cache"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"lru",
"oxide_bytecode",
@@ -1234,7 +1234,7 @@ dependencies = [
[[package]]
name = "oxide_compiler"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1253,7 +1253,7 @@ dependencies = [
[[package]]
name = "oxide_dce"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1267,7 +1267,7 @@ dependencies = [
[[package]]
name = "oxide_emit"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"num-bigint",
"oxc_allocator",
@@ -1282,7 +1282,7 @@ dependencies = [
[[package]]
name = "oxide_ir"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_log",
@@ -1291,7 +1291,7 @@ dependencies = [
[[package]]
name = "oxide_kernel"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"bumpalo",
"dashmap",
@@ -1306,7 +1306,7 @@ dependencies = [
[[package]]
name = "oxide_liveness"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1318,7 +1318,7 @@ dependencies = [
[[package]]
name = "oxide_log"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"tracing",
"tracing-appender",
@@ -1327,7 +1327,7 @@ dependencies = [
[[package]]
name = "oxide_parser"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxc_allocator",
"oxc_ast",
@@ -1342,7 +1342,7 @@ dependencies = [
[[package]]
name = "oxide_regalloc"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_cfg",
@@ -1358,7 +1358,7 @@ dependencies = [
[[package]]
name = "oxide_runtime_api"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"num-bigint",
"num-traits",
@@ -1370,7 +1370,7 @@ dependencies = [
[[package]]
name = "oxide_test262"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"oxide_bytecode",
"oxide_compiler",
@@ -1387,7 +1387,7 @@ dependencies = [
[[package]]
name = "oxide_types"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"bumpalo",
"hashbrown 0.14.5",
@@ -1399,7 +1399,7 @@ dependencies = [
[[package]]
name = "oxide_vm"
version = "0.8.38"
version = "0.8.39"
dependencies = [
"bumpalo",
"chrono",
+1 -1
View File
@@ -22,7 +22,7 @@ members = [
]
[workspace.package]
version = "0.8.38"
version = "0.8.39"
edition = "2021"
license = "MIT"
rust-version = "1.80"
+108 -32
View File
@@ -110,7 +110,7 @@ enum Commands {
/// server 子命令:持久 server 进程管理与 --rm 独立模式。
#[derive(Subcommand)]
enum ServerCommands {
/// 启动 server。持久形态前台常驻(阻塞至关闭请求、信号或 yield 触发退出);
/// 启动 server。持久形态为守护形态(spawn 脱离的后台进程,就绪探测通过后返回);
/// --rm 为独立形态(进程唯一 socket、不注册 sidecar、空闲超时或断开即退出)。
Start {
/// 独立模式:进程唯一 socket 路径,不注册 sidecar,空闲超时或断开即自动退出。
@@ -122,6 +122,10 @@ enum ServerCommands {
/// 常驻 worker 线程数;缺省取宿主核数。
#[arg(long)]
workers: Option<u32>,
/// 前台形态:当前进程直接运行持久 server(守护形态 spawn 的子进程用,
/// 不向用户暴露)。
#[arg(long, hide = true)]
foreground: bool,
},
/// 向持久 server 发送关闭请求。
Stop,
@@ -212,8 +216,10 @@ fn main() -> ExitCode {
/// 其余两臂为 not_implemented 占位。
///
/// # 步骤
/// 1. Start:rm 为假调持久 server 入口(well-known 路径),rm 为真调独立
/// 模式入口(进程唯一路径);`--workers` 为 Some 时覆盖 worker 数。
/// 1. Start:rm 为真调独立模式入口(进程唯一路径);foreground 为真当前进程
/// 直接运行持久 server(守护形态 spawn 的子进程用);否则 spawn 脱离的
/// server 子进程并就绪探测通过后返回(守护形态,well-known 路径);
/// `--workers` 为 Some 时覆盖 worker 数。
/// 2. Cleanup:调 well-known 路径清理入口,四态结果映射退出码(无残留与
/// 已清理退 0,拒绝与杀进程失败退 1)并打印结果。
/// 3. 五个控制臂(version/status/health/info/stop):发对应控制请求,
@@ -227,12 +233,13 @@ fn main() -> ExitCode {
/// - 全局旗标 -v / -q / --profile 在 server 臂为空操作。
///
/// # 副作用
/// - start 阻塞当前进程至 server 退出(持久形态为前台常驻)。
/// - start 持久形态 spawn 脱离进程,就绪探测通过后返回(server 由 init
/// 收领继续运行)。
/// - cleanup 可能向 sidecar 记录的进程发 SIGTERM 并删除残留文件。
/// - 控制臂向 server 建立并关闭一条 Unix socket 连接。
fn server_command(command: ServerCommands) -> ExitCode {
match command {
ServerCommands::Start { rm, idle_timeout, workers } => {
ServerCommands::Start { rm, idle_timeout, workers, foreground } => {
if rm {
let mut config = RmServerConfig {
idle_timeout: Duration::from_secs(idle_timeout),
@@ -242,12 +249,14 @@ fn server_command(command: ServerCommands) -> ExitCode {
config.worker_count = n as usize;
}
start_result(run_server_rm(&config))
} else {
} else if foreground {
let mut config = ServerConfig::well_known();
if let Some(n) = workers {
config.worker_count = n as usize;
}
start_result(run_server(&config))
} else {
start_daemon(workers)
}
}
ServerCommands::Cleanup => match well_known_cleanup() {
@@ -322,6 +331,95 @@ fn control_arm(request: &ServerRequest, render: fn(&ServerResponse) -> Option<St
}
}
/// start 持久形态:spawn 脱离的前台 server 子进程(`server start
/// --foreground`),就绪探测通过后打印返回。
///
/// # 步骤
/// 1. spawn 脱离的前台 server 子进程(标准流全空、不等待);spawn 失败以
/// 红色打印退 1。
/// 2. 就绪探测:轮询健康请求直至得 healthy(100 毫秒间隔、10 秒截止);
/// 超时以红色打印退 1。
/// 3. 打印「server 已启动」退 0。
///
/// # 边界与前提
/// - sidecar 文件由被拉起的 server 进程自己写(先认领再绑定 socket 的启动
/// 顺序契约在 server 进程内执行),本命令不写 sidecar。
/// - 已有存活 server 时再跑本命令:第二个 spawn 的子进程在认领阶段被拒
/// (同版本不触碰文件),探测看到第一个 server 即通过,命令退 0(天然幂等)。
///
/// # 副作用
/// - server 为脱离进程,CLI 退出后由 init 收领继续运行。
fn start_daemon(workers: Option<u32>) -> ExitCode {
// spawn:脱离子进程,标准流全空,不等待。
if let Err(err) = spawn_detached_server(workers) {
eprintln!("{}", Red.paint(format!("spawn server 进程失败:{err}")));
return ExitCode::FAILURE;
}
// 就绪探测:单循环全覆盖「sidecar 出现 + 健康请求通过」。
if let Err(msg) = wait_server_ready(Duration::from_secs(10)) {
eprintln!("{}", Red.paint(format!("{msg}(可用 `oxide server cleanup` 人工检查)")));
return ExitCode::FAILURE;
}
println!("server 已启动");
ExitCode::SUCCESS
}
/// spawn 脱离的前台 server 子进程(`server start --foreground`):标准流全空,不等待。
///
/// # 步骤
/// 1. 取当前可执行文件路径。
/// 2. 构造命令 `server start --foreground` 加可选 `--workers N`。
/// 3. 三个标准流置空并 spawn。
///
/// # 边界与前提
/// - 必须带 `--foreground`:守护形态 spawn 的是前台 server 入口,不带该旗标会
/// 再次进入守护形态形成递归 spawn。
///
/// # 副作用
/// - 创建脱离的子进程,调用方退出后由 init 收领继续运行。
fn spawn_detached_server(workers: Option<u32>) -> std::io::Result<std::process::Child> {
let exe = std::env::current_exe()?;
let mut cmd = Command::new(exe);
cmd.args(["server", "start", "--foreground"]);
if let Some(n) = workers {
cmd.args(["--workers", &n.to_string()]);
}
cmd.stdin(Stdio::null())
.stdout(Stdio::null())
.stderr(Stdio::null())
.spawn()
}
/// 就绪探测:轮询健康请求直至得 healthy。
///
/// # 步骤
/// 1. 循环发健康请求,得 healthy 即返回成功。
/// 2. 超过截止返回含超时秒数的消息。
///
/// # 边界与前提
/// - 控制客户端发请求前先读 sidecar 取 socket 路径,sidecar 缺失时自然走
/// 连接失败分支,「sidecar 出现」与「健康请求通过」一个循环全覆盖
/// (socket 文件出现蕴含监听器已绑定,健康请求通过蕴含 accept 循环在运行)。
///
/// # 副作用
/// - 每轮建立并关闭一条 Unix socket 连接。
fn wait_server_ready(deadline: Duration) -> Result<(), String> {
let end = Instant::now() + deadline;
loop {
let ready =
matches!(send_control_request(&ServerRequest::Health), Ok(ServerResponse::Health { healthy: true }));
if ready {
return Ok(());
}
if Instant::now() >= end {
return Err(format!("server 未在 {} 秒内就绪", deadline.as_secs()));
}
std::thread::sleep(Duration::from_millis(100));
}
}
/// restart 子命令:向存活 server 发关闭请求,等其退出后拉起新 server 进程。
///
/// # 步骤
@@ -360,37 +458,15 @@ fn restart_command() -> ExitCode {
}
// 拉起新 server:脱离进程,标准流全空,不等待。
let exe = match std::env::current_exe() {
Ok(path) => path,
Err(err) => {
eprintln!("{}", Red.paint(format!("获取当前可执行文件路径失败:{err}")));
return ExitCode::FAILURE;
}
};
if let Err(err) = Command::new(exe)
.args(["server", "start"])
.stdin(Stdio::null())
.stdout(Stdio::null())
.stderr(Stdio::null())
.spawn()
{
if let Err(err) = spawn_detached_server(None) {
eprintln!("{}", Red.paint(format!("拉起新 server 失败:{err}")));
return ExitCode::FAILURE;
}
// 就绪探测:健康请求通过即新 server 的监听与 accept 循环在运行。
let deadline = Instant::now() + Duration::from_secs(10);
loop {
let ready =
matches!(send_control_request(&ServerRequest::Health), Ok(ServerResponse::Health { healthy: true }));
if ready {
break;
}
if Instant::now() >= deadline {
eprintln!("{}", Red.paint("新 server 未在 10 秒内就绪"));
return ExitCode::FAILURE;
}
std::thread::sleep(Duration::from_millis(100));
if let Err(msg) = wait_server_ready(Duration::from_secs(10)) {
eprintln!("{}", Red.paint(format!("新 {msg}")));
return ExitCode::FAILURE;
}
println!("已重启");
+54 -26
View File
@@ -99,8 +99,9 @@ fn server_cleanup_exit_0() {
assert!(!sidecar::well_known_sidecar_path().exists(), "sidecar 文件应不存在");
}
/// start 端到端:起真实子进程,well-known 路径连接,版本帧断言构建期版本,
/// 关闭帧后子进程退 0,socket 与 sidecar 文件被删。
/// start 端到端:CLI 即刻返回(server 是脱离的孙进程),断言退 0 且打印
/// 「已启动」,well-known 路径连接,版本帧断言构建期版本,关闭帧后 server
/// 进程退出,socket 与 sidecar 文件被删。
#[test]
fn server_start_e2e() {
let _guard = WELL_KNOWN_LOCK.lock().expect("全局路径锁不应中毒");
@@ -110,18 +111,15 @@ fn server_start_e2e() {
// 探活先行:存活 server 占用全局路径时 panic,不抢占。
assert!(!sidecar::is_server_alive(&socket), "存活 server 占用 well-known 路径,测试不抢占");
// worker 数压到 2:避免按宿主核数预热的成本。
let mut child = Command::new(env!("CARGO_BIN_EXE_oxide"))
.args(["server", "start", "--workers", "2"])
.spawn()
.expect("oxide server start 应可启动");
// start 为守护形态:CLI 即刻返回,server 是脱离的孙进程;worker 数压到 2
// 避免按宿主核数预热的成本。
let output = oxide(&["server", "start", "--workers", "2"]);
assert_eq!(output.status.code(), Some(0), "start 应退 0");
let stdout = String::from_utf8_lossy(&output.stdout);
assert!(stdout.contains("已启动"), "start 应打印「已启动」:{stdout}");
// 轮询 socket 文件出现(10 秒截止)判就绪。
let deadline = Instant::now() + Duration::from_secs(10);
while !socket.exists() {
assert!(Instant::now() < deadline, "socket 文件未出现");
std::thread::sleep(Duration::from_millis(10));
}
// 读 sidecar 取 server 进程号。
let id = sidecar::read_identity(&sidecar_path).expect("读 sidecar 应成功");
// 版本请求帧:版本应等于构建期版本。
let mut stream = UnixStream::connect(&socket).expect("连接 server 应成功");
@@ -138,24 +136,54 @@ fn server_start_e2e() {
assert!(matches!(response, ServerResponse::Shutdown), "应得关闭确认帧:{response:?}");
drop(stream);
// 等子进程以退出码 0 退出(30 秒截止)。
// 轮询 sidecar 文件消失(30 秒截止、50 毫秒间隔)判 server 进程已退出。
let deadline = Instant::now() + Duration::from_secs(30);
loop {
match child.try_wait() {
Ok(Some(status)) => {
assert_eq!(status.code(), Some(0), "server 应以退出码 0 退出:{status:?}");
break;
}
Ok(None) => {
assert!(Instant::now() < deadline, "server 应在有界时间内退出");
std::thread::sleep(Duration::from_millis(50));
}
Err(err) => panic!("等待子进程失败:{err}"),
}
while sidecar_path.exists() {
assert!(Instant::now() < deadline, "server 应在有界时间内退出");
std::thread::sleep(Duration::from_millis(50));
}
assert!(!socket.exists(), "socket 文件应被删除");
assert!(!sidecar_path.exists(), "sidecar 文件应被删除");
assert!(!sidecar::pid_alive(id.pid), "server 进程应已退出");
}
/// start 双启动幂等:已有存活 server 时再跑 start,断言第二次退 0 且打印
/// 「已启动」(第二个 spawn 的子进程认领被拒退 1 不触碰文件,探测看到第一个
/// server 即通过)。
#[test]
fn server_start_when_running() {
let _guard = WELL_KNOWN_LOCK.lock().expect("全局路径锁不应中毒");
let socket = sidecar::well_known_socket_path();
let sidecar_path = sidecar::well_known_sidecar_path();
// 探活先行:存活 server 占用全局路径时 panic,不抢占。
assert!(!sidecar::is_server_alive(&socket), "存活 server 占用 well-known 路径,测试不抢占");
// 首次 start:CLI 即刻返回,server 是脱离的孙进程。
let output = oxide(&["server", "start", "--workers", "2"]);
assert_eq!(output.status.code(), Some(0), "首次 start 应退 0");
let stdout = String::from_utf8_lossy(&output.stdout);
assert!(stdout.contains("已启动"), "首次 start 应打印「已启动」:{stdout}");
// 二次 start:第二个 spawn 的子进程认领被拒,探测看到第一个 server 即通过。
let output = oxide(&["server", "start", "--workers", "2"]);
assert_eq!(output.status.code(), Some(0), "二次 start 应退 0");
let stdout = String::from_utf8_lossy(&output.stdout);
assert!(stdout.contains("已启动"), "二次 start 应打印「已启动」:{stdout}");
// 收尾:发关闭请求,等 sidecar 消失。
let mut stream = UnixStream::connect(&socket).expect("连接 server 应成功");
let response = send_and_recv(&mut stream, &ServerRequest::Shutdown);
assert!(matches!(response, ServerResponse::Shutdown), "应得关闭确认帧:{response:?}");
drop(stream);
let deadline = Instant::now() + Duration::from_secs(30);
while sidecar_path.exists() {
assert!(Instant::now() < deadline, "server 应在有界时间内退出");
std::thread::sleep(Duration::from_millis(50));
}
assert!(!socket.exists(), "socket 文件应被删除");
assert!(!sidecar_path.exists(), "sidecar 文件应被删除");
}
/// restart 端到端:起真实子进程 A,跑 `oxide server restart`(发关闭、等退出、