mirror of
https://github.com/pjh456/OxideJS.git
synced 2026-10-24 02:31:13 +08:00
feat: server watchdog 子命令，崩溃自动重启加信号优雅停栈
This commit is contained in:
Generated
+18
-18
@@ -1160,7 +1160,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_builtins"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"chrono",
|
||||
"fast-float",
|
||||
@@ -1185,7 +1185,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_bytecode"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"num-bigint",
|
||||
"oxide_log",
|
||||
@@ -1193,7 +1193,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_cfg"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_emit",
|
||||
@@ -1204,7 +1204,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_cli"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"ansi_term",
|
||||
"bumpalo",
|
||||
@@ -1225,7 +1225,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_code_cache"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"lru",
|
||||
"oxide_bytecode",
|
||||
@@ -1234,7 +1234,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_compiler"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_cfg",
|
||||
@@ -1253,7 +1253,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_dce"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_cfg",
|
||||
@@ -1267,7 +1267,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_emit"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"num-bigint",
|
||||
"oxc_allocator",
|
||||
@@ -1282,7 +1282,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_ir"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_log",
|
||||
@@ -1291,7 +1291,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_kernel"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"bumpalo",
|
||||
"dashmap",
|
||||
@@ -1306,7 +1306,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_liveness"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_cfg",
|
||||
@@ -1318,7 +1318,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_log"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"tracing",
|
||||
"tracing-appender",
|
||||
@@ -1327,7 +1327,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_parser"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxc_allocator",
|
||||
"oxc_ast",
|
||||
@@ -1342,7 +1342,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_regalloc"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_cfg",
|
||||
@@ -1358,7 +1358,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_runtime_api"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"num-bigint",
|
||||
"num-traits",
|
||||
@@ -1370,7 +1370,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_test262"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"oxide_bytecode",
|
||||
"oxide_compiler",
|
||||
@@ -1387,7 +1387,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_types"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"bumpalo",
|
||||
"hashbrown 0.14.5",
|
||||
@@ -1399,7 +1399,7 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "oxide_vm"
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
dependencies = [
|
||||
"bumpalo",
|
||||
"chrono",
|
||||
|
||||
+1
-1
@@ -22,7 +22,7 @@ members = [
|
||||
]
|
||||
|
||||
[workspace.package]
|
||||
version = "0.8.41"
|
||||
version = "0.8.42"
|
||||
edition = "2021"
|
||||
license = "MIT"
|
||||
rust-version = "1.80"
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
#![allow(dead_code)]
|
||||
|
||||
use std::fs;
|
||||
use std::process::{Command, ExitCode, Stdio};
|
||||
use std::process::ExitCode;
|
||||
use std::sync::Arc;
|
||||
use std::time::{Duration, Instant};
|
||||
|
||||
@@ -15,6 +15,8 @@ use oxide_cli::server::log as server_log;
|
||||
use oxide_cli::server::protocol::{ForgeTarget, LookupResult, ServerRequest, ServerResponse};
|
||||
use oxide_cli::server::server::{run_server, run_server_rm, RmServerConfig, ServerConfig};
|
||||
use oxide_cli::server::sidecar;
|
||||
use oxide_cli::server::spawn;
|
||||
use oxide_cli::server::watchdog;
|
||||
use oxide_compiler::compiler::{compiled_module_hash, Compiler};
|
||||
use oxide_compiler::compiler_error;
|
||||
use oxide_kernel::kernel::{KernelConfig, KernelCore};
|
||||
@@ -169,6 +171,12 @@ enum ServerCommands {
|
||||
#[arg(long)]
|
||||
lookup: Option<String>,
|
||||
},
|
||||
/// 监控 server 存活并在崩溃后自动重启（前台进程，60 秒窗口 5 次崩溃即停）。
|
||||
Watchdog {
|
||||
/// 常驻 worker 线程数；缺省取宿主核数。
|
||||
#[arg(long)]
|
||||
workers: Option<u32>,
|
||||
},
|
||||
}
|
||||
|
||||
/// forge 查询目标（CLI 侧，映射到协议侧 `ForgeTarget`）。
|
||||
@@ -281,10 +289,8 @@ fn main() -> ExitCode {
|
||||
}
|
||||
}
|
||||
|
||||
/// server 子命令分派：start（两形态）与 cleanup 为最小实现，
|
||||
/// version/status/health/info/stop/restart 六臂为真实实现，
|
||||
/// log 为真实实现（读 well-known 日志文件，支持 --follow/--level/--lines），
|
||||
/// forge 为真实实现（查询四张共享 forge 状态，加三行动旗标）。
|
||||
/// server 子命令分派：全部十一臂为真实实现（start 两形态、cleanup、
|
||||
/// 五个控制臂、restart、log、forge、watchdog）。
|
||||
///
|
||||
/// # 步骤
|
||||
/// 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
|
||||
@@ -301,6 +307,8 @@ fn main() -> ExitCode {
|
||||
/// `--follow` 时进入阻塞跟踪循环（详见 log_command）。
|
||||
/// 6. Forge：发 forge 查询请求（目标加三行动旗标），成功打印渲染文本退 0，
|
||||
/// 失败打印错误消息退 1。
|
||||
/// 7. Watchdog：前台监控进程，拉起 server、三态裁决、崩溃后自动重启
|
||||
/// （详见 run_watchdog）。
|
||||
///
|
||||
/// # 边界与前提
|
||||
/// - `--idle-timeout` 不带 --rm 时静默忽略（语义门控归后续任务）。
|
||||
@@ -311,6 +319,7 @@ fn main() -> ExitCode {
|
||||
/// 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。
|
||||
/// - watchdog spawn 脱离的 server 子进程，崩溃后经同一入口重启。
|
||||
fn server_command(command: ServerCommands) -> ExitCode {
|
||||
match command {
|
||||
ServerCommands::Start { rm, idle_timeout, workers, foreground } => {
|
||||
@@ -411,6 +420,7 @@ fn server_command(command: ServerCommands) -> ExitCode {
|
||||
_ => None,
|
||||
})
|
||||
}
|
||||
ServerCommands::Watchdog { workers } => watchdog::run_watchdog(workers),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -484,13 +494,13 @@ fn control_arm(request: &ServerRequest, render: fn(&ServerResponse) -> Option<St
|
||||
/// - 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) {
|
||||
if let Err(err) = spawn::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)) {
|
||||
if let Err(msg) = spawn::wait_server_ready(Duration::from_secs(10)) {
|
||||
eprintln!("{}", Red.paint(format!("{msg}（可用 `oxide server cleanup` 人工检查）")));
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
@@ -499,60 +509,6 @@ fn start_daemon(workers: Option<u32>) -> ExitCode {
|
||||
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
|
||||
@@ -591,13 +547,13 @@ fn restart_command() -> ExitCode {
|
||||
}
|
||||
|
||||
// 拉起新 server：脱离进程，标准流全空，不等待。
|
||||
if let Err(err) = spawn_detached_server(None) {
|
||||
if let Err(err) = spawn::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。
|
||||
if let Err(msg) = wait_server_ready(Duration::from_secs(10)) {
|
||||
if let Err(msg) = spawn::wait_server_ready(Duration::from_secs(10)) {
|
||||
eprintln!("{}", Red.paint(format!("新 {msg}")));
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
|
||||
@@ -2,14 +2,16 @@
|
||||
//!
|
||||
//! 持久 server 是常驻进程，复用预热池、近零 spawn 成本；
|
||||
//! 通信走 Unix socket + NDJSON（换行分隔 JSON）。
|
||||
//! 模块分八层：协议数据面（帧格式与 serde 结构体）、身份注册（sidecar 文件
|
||||
//! 模块分十层：协议数据面（帧格式与 serde 结构体）、身份注册（sidecar 文件
|
||||
//! 与 liveness 检查原语）、控制客户端（读 sidecar、连接 server、发控制请求
|
||||
//! 帧、读回响应帧、三分支错误）、执行请求路径（parse → compile → spawn →
|
||||
//! run → format → drop 六段纯函数）、worker 池（固定 N 常驻 worker、各持
|
||||
//! 自有 VM 池、mpsc 轮询路由）、server 进程主体（accept 循环、控制请求直接
|
||||
//! 处理、执行请求路由、优雅退出）、日志读取（行级别解析、阈值加最后 N 行
|
||||
//! 过滤、阻塞式跟踪）、forge 状态查询（读四张共享 forge 条目数与容量、执行
|
||||
//! gc / clear-cache / lookup 三旗标）。
|
||||
//! gc / clear-cache / lookup 三旗标）、spawn 与就绪探测（start、restart、
|
||||
//! watchdog 三条路径共用的唯一共享入口）、watchdog（前台监控进程、三态裁决
|
||||
//! 加崩溃预算、崩溃后自动重启）。
|
||||
|
||||
/// 人工兜底清理命令：扫 sidecar、探活、杀 PID、删孤儿文件。
|
||||
pub mod cleanup;
|
||||
@@ -34,5 +36,9 @@ pub mod protocol;
|
||||
pub mod server;
|
||||
/// 身份注册：sidecar 文件、liveness 检查与陈旧清理。
|
||||
pub mod sidecar;
|
||||
/// spawn 与就绪探测：start、restart、watchdog 三条路径共用的唯一共享入口。
|
||||
pub mod spawn;
|
||||
/// worker 池：固定 N 常驻 worker、各持自有 VM 池、mpsc 轮询路由。
|
||||
pub mod workers;
|
||||
/// watchdog：前台监控进程，三态裁决加崩溃预算，崩溃后自动重启。
|
||||
pub mod watchdog;
|
||||
|
||||
@@ -0,0 +1,65 @@
|
||||
//! spawn 与就绪探测：拉起脱离 server 与就绪探测的唯一共享入口。
|
||||
//!
|
||||
//! start、restart、watchdog 三条路径共用本模块的同一对函数，不另起第二套
|
||||
//! spawn 逻辑。`current_exe()` 在 lib 上下文与二进制目标返回同一路径（lib
|
||||
//! 链接进同一个 oxide 二进制），spawn 行为逐字节不变。
|
||||
|
||||
use std::process::{Command, Stdio};
|
||||
use std::time::{Duration, Instant};
|
||||
|
||||
use super::client::send_control_request;
|
||||
use super::protocol::{ServerRequest, ServerResponse};
|
||||
|
||||
/// 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。
|
||||
pub 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。
|
||||
pub 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));
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,374 @@
|
||||
//! watchdog：持久 server 的前台监控进程，崩溃后自动重启。
|
||||
//!
|
||||
//! 关键约定：
|
||||
//! - release 模式 `panic=abort` 使 `catch_unwind` 失效，server 崩溃后进程直接
|
||||
//! 消亡、文件残留，watchdog 是崩溃后恢复的唯一自动手段。
|
||||
//! - 裁决信号取自既有工件加子进程收领：子进程是监控对象时，其退出状态是
|
||||
//! 死亡的权威信号（被杀的子进程是僵尸，PID 探活对僵尸误判活）；子进程在
|
||||
//! 认领阶段被拒时监控对象是既有 server，按 sidecar 存在性、PID 与启动时
|
||||
//! 刻度（liveness 原语，排 PID 复用误判）三信号裁决；日志文件（跨重启的
|
||||
//! 诊断工件）供崩溃诊断打印。
|
||||
//! - watchdog 自身不注册身份、不写任何文件；watchdog 消亡后无人自动恢复
|
||||
//! （纯标准库轮询进程，不建内核、不建池、不执行引擎代码，崩溃风险极低；
|
||||
//! server 是脱离进程，watchdog 退出或消亡都不影响 server 运行），人工重跑
|
||||
//! `oxide server watchdog` 即接管（就绪探测看到存活 server 即接管监控）。
|
||||
//! - SIGINT/SIGTERM 时先尽力向 server 发关闭请求（优雅退出、删文件）再退 0
|
||||
//! ——停 watchdog 的语义是「停掉整个监控栈」，不留下无人监控的 server。
|
||||
|
||||
use std::path::Path;
|
||||
use std::process::{ExitCode, ExitStatus};
|
||||
use std::sync::atomic::{AtomicBool, Ordering};
|
||||
use std::sync::Arc;
|
||||
use std::time::{Duration, Instant};
|
||||
|
||||
use super::client::send_control_request;
|
||||
use super::log as server_log;
|
||||
use super::protocol::ServerRequest;
|
||||
use super::sidecar::{self, ServerIdentity};
|
||||
use super::spawn::{spawn_detached_server, wait_server_ready};
|
||||
|
||||
/// 监控循环轮询间隔：500 毫秒（与日志跟踪的轮询间隔同值）。
|
||||
const POLL_INTERVAL: Duration = Duration::from_millis(500);
|
||||
|
||||
/// 崩溃重启退避时长：1 秒。
|
||||
const CRASH_BACKOFF: Duration = Duration::from_secs(1);
|
||||
|
||||
/// 崩溃预算窗口：60 秒。
|
||||
const CRASH_WINDOW: Duration = Duration::from_secs(60);
|
||||
|
||||
/// 崩溃预算上限：窗口内 5 次崩溃。
|
||||
const CRASH_LIMIT: usize = 5;
|
||||
|
||||
/// 崩溃诊断打印的日志尾部行数：20 行。
|
||||
const CRASH_LOG_LINES: usize = 20;
|
||||
|
||||
/// 就绪探测截止：10 秒（与 start / restart 路径同值）。
|
||||
const READY_DEADLINE: Duration = Duration::from_secs(10);
|
||||
|
||||
/// server 状态三态：存活、崩溃、已退出。
|
||||
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
|
||||
pub enum WatchState {
|
||||
/// 存活：PID 活且启动时刻刻度与 sidecar 记录匹配。
|
||||
Alive,
|
||||
/// 崩溃：sidecar 残留且 PID 死，或 PID 活但刻度失配（PID 已被复用）。
|
||||
Crash,
|
||||
/// 已退出：sidecar 消失（优雅退出序列的最后删除项，或外部清理）。
|
||||
Exited,
|
||||
}
|
||||
|
||||
/// 裁决表：sidecar 可读性、PID 存活、刻度匹配三信号映射三态。
|
||||
///
|
||||
/// # 边界与前提
|
||||
/// - 纯函数无 I/O，信号由调用方采集（与 cleanup 的 decide 同型）。
|
||||
pub fn decide(identity: Option<&ServerIdentity>, pid_alive: bool, starttime_match: bool) -> WatchState {
|
||||
match identity {
|
||||
None => WatchState::Exited,
|
||||
Some(_) if !pid_alive => WatchState::Crash,
|
||||
Some(_) if starttime_match => WatchState::Alive,
|
||||
Some(_) => WatchState::Crash,
|
||||
}
|
||||
}
|
||||
|
||||
/// 崩溃预算：`window` 滑动窗口内至多 `limit` 次崩溃，超出即放弃自动重启。
|
||||
///
|
||||
/// 窗口语义：窗口内崩溃次数超过上限即判 server 不稳定，继续重启只会空转
|
||||
/// （典型场景：二进制损坏导致每次启动即崩溃，但每次都通过了就绪探测）。
|
||||
#[derive(Debug)]
|
||||
pub struct CrashBudget {
|
||||
window: Duration,
|
||||
limit: usize,
|
||||
times: Vec<Instant>,
|
||||
}
|
||||
|
||||
impl CrashBudget {
|
||||
/// 默认预算：60 秒窗口、5 次崩溃。
|
||||
pub fn new() -> Self {
|
||||
Self::with(CRASH_WINDOW, CRASH_LIMIT)
|
||||
}
|
||||
|
||||
/// 指定窗口与上限构造预算（生产取默认值，测试注入短窗口）。
|
||||
pub fn with(window: Duration, limit: usize) -> Self {
|
||||
CrashBudget {
|
||||
window,
|
||||
limit,
|
||||
times: Vec::new(),
|
||||
}
|
||||
}
|
||||
|
||||
/// 记录一次崩溃：清掉窗口外条目后追加，返回预算是否耗尽。
|
||||
///
|
||||
/// # 边界与前提
|
||||
/// - 耗尽是幂等的：耗尽后继续记录仍返回耗尽。
|
||||
pub fn record(&mut self, now: Instant) -> bool {
|
||||
self.times.retain(|t| now.duration_since(*t) <= self.window);
|
||||
self.times.push(now);
|
||||
self.times.len() > self.limit
|
||||
}
|
||||
}
|
||||
|
||||
impl Default for CrashBudget {
|
||||
fn default() -> Self {
|
||||
Self::new()
|
||||
}
|
||||
}
|
||||
|
||||
/// watchdog 主循环：拉起 server、就绪探测、监控循环、崩溃重启、信号退出。
|
||||
///
|
||||
/// # 步骤
|
||||
/// 1. 注册 ctrlc 处理器置位停止标志（SIGINT/SIGTERM，termination 特性）。
|
||||
/// 须在初始 spawn 之前注册：sidecar 出现早于 socket 绑定，外部信号可能
|
||||
/// 在就绪探测期间到达，处理器未就位时走默认终止行为。
|
||||
/// 2. 初始 spawn（共享入口）加就绪探测（10 秒），失败打印退 1；成功打印
|
||||
/// 监控开始标记行（外部崩溃模拟与信号在此之后才安全）。
|
||||
/// 3. 监控循环（500 毫秒间隔）：停止标志置位即转优雅关闭；收领子进程
|
||||
/// （被杀的子进程是僵尸，PID 探活对僵尸误判活，子进程退出状态是死亡的
|
||||
/// 权威信号）；子进程即监控对象时按其存活判 Alive、已退出按 sidecar
|
||||
/// 残留判 Crash；子进程在认领阶段被拒时监控对象是既有 server，按 decide
|
||||
/// 三信号裁决；sidecar 连续两次（1 秒）判缺失即 server 已正常退出、退 0。
|
||||
/// 4. 崩溃处理：打印崩溃提示（含进程号）、读日志文件最后 20 行打印、记预算
|
||||
/// （耗尽则打印放弃提示退 1）、退避 1 秒、经同一 spawn 入口重启加就绪
|
||||
/// 探测（失败打印退 1）、继续循环。
|
||||
/// 5. 优雅关闭：尽力发关闭请求（忽略错误，server 可能已死），打印退 0。
|
||||
///
|
||||
/// # 边界与前提
|
||||
/// - sidecar 缺失须连续两次轮询（1 秒）确认才判 Exited：新 server 的认领
|
||||
/// 路径先删旧 sidecar 再重写，删除与重写之间是微秒级窗口，单次轮询落入
|
||||
/// 窗口会误判「server 已正常退出」而静默退出、留下无人监控的 server；
|
||||
/// 两次确认把该窗口压到实际不可达。
|
||||
/// - 已有存活 server 时再跑 watchdog：spawn 的子进程在认领阶段被拒（同版本
|
||||
/// 判 `AlreadyRunning`，退 1、不触碰文件），就绪探测看到第一个 server 即
|
||||
/// 通过，watchdog 监控的是第一个 server（与双启动幂等同一语义，天然安全）。
|
||||
/// - 新 server 启动即崩溃（绑定前消亡、sidecar 从未出现）：就绪探测 10 秒
|
||||
/// 超时后 watchdog 退 1（自限，不形成无限循环），消息带 `oxide server
|
||||
/// cleanup` 人工检查提示。
|
||||
///
|
||||
/// # 副作用
|
||||
/// - spawn 脱离的 server 子进程；崩溃后经同一入口重启。
|
||||
/// - SIGINT/SIGTERM 时尽力向 server 发关闭请求。
|
||||
pub fn run_watchdog(workers: Option<u32>) -> ExitCode {
|
||||
// 先注册信号处理器：信号到达时置位停止标志（处理器体只做一次原子写）。
|
||||
// 须在初始 spawn 之前注册：sidecar 出现早于 socket 绑定，外部信号可能
|
||||
// 在就绪探测期间到达，处理器未就位时 SIGINT 走默认终止行为。
|
||||
let stop = Arc::new(AtomicBool::new(false));
|
||||
let flag = Arc::clone(&stop);
|
||||
if let Err(err) = ctrlc::set_handler(move || {
|
||||
flag.store(true, Ordering::SeqCst);
|
||||
}) {
|
||||
eprintln!("注册信号处理器失败：{err}");
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
|
||||
// 初始 spawn 与就绪探测：失败即退 1，不进入监控循环。
|
||||
let mut child = match spawn_detached_server(workers) {
|
||||
Ok(child) => child,
|
||||
Err(err) => {
|
||||
eprintln!("spawn server 进程失败：{err}");
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
};
|
||||
if let Err(msg) = wait_server_ready(READY_DEADLINE) {
|
||||
eprintln!("{msg}（可用 `oxide server cleanup` 人工检查）");
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
// 初始就绪探测通过：外部信号（崩溃模拟、SIGINT）在此之后才安全，
|
||||
// 该行同时是监控循环开始的标记。
|
||||
println!("server 已就绪，watchdog 开始监控");
|
||||
|
||||
let mut budget = CrashBudget::new();
|
||||
let sidecar_path = sidecar::well_known_sidecar_path();
|
||||
let log_path = sidecar::well_known_log_path();
|
||||
let mut missing_streak = 0usize;
|
||||
// 子进程退出状态：收领后只记一次（try_wait 收领后不再重复出状态）。
|
||||
let mut child_exited: Option<ExitStatus> = None;
|
||||
|
||||
loop {
|
||||
// 停止标志置位：转优雅关闭（尽力停 server，退 0）。
|
||||
if stop.load(Ordering::SeqCst) {
|
||||
return graceful_stop();
|
||||
}
|
||||
std::thread::sleep(POLL_INTERVAL);
|
||||
|
||||
// 收领子进程：子进程是监控对象时，其退出状态是死亡的权威信号
|
||||
// （被杀的子进程是僵尸，kill -0 对僵尸返回活，PID 探活会误判）。
|
||||
if child_exited.is_none() {
|
||||
if let Some(status) = child.try_wait().ok().flatten() {
|
||||
child_exited = Some(status);
|
||||
}
|
||||
}
|
||||
|
||||
// 读 sidecar 采集信号：缺失要连续两次确认（认领路径的删旧写新窗口
|
||||
// 是微秒级，单次轮询落入窗口会误判正常退出）。
|
||||
let identity = sidecar::read_identity(&sidecar_path);
|
||||
let state = match &identity {
|
||||
None => {
|
||||
missing_streak += 1;
|
||||
if missing_streak >= 2 {
|
||||
println!("server 已正常退出，watchdog 退出");
|
||||
return ExitCode::SUCCESS;
|
||||
}
|
||||
continue;
|
||||
}
|
||||
Some(id) => {
|
||||
missing_streak = 0;
|
||||
if id.pid == child.id() {
|
||||
// 子进程即监控对象：存活判 Alive，已退出按 sidecar 残留
|
||||
// 区分崩溃与正常退出。
|
||||
match child_exited {
|
||||
Some(_) => WatchState::Crash,
|
||||
None => WatchState::Alive,
|
||||
}
|
||||
} else {
|
||||
// 子进程在认领阶段被拒（已有存活 server）：监控对象是
|
||||
// 既有 server，按三信号裁决。
|
||||
let pid_alive = sidecar::pid_alive(id.pid);
|
||||
let starttime_match = sidecar::process_starttime(id.pid) == Some(id.starttime);
|
||||
decide(Some(id), pid_alive, starttime_match)
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
match state {
|
||||
WatchState::Alive => {}
|
||||
WatchState::Crash => {
|
||||
// 崩溃处理：诊断打印、记预算、退避、经同一入口重启。
|
||||
// 崩溃态下 sidecar 必可读（Crash 只从 Some 分支产生），进程号
|
||||
// 取 sidecar 记录；读不到时以 0 占位（防御性兜底，不 panic）。
|
||||
let pid = identity.as_ref().map(|id| id.pid).unwrap_or(0);
|
||||
println!("server 已崩溃（进程号 {pid}），准备重启");
|
||||
print_crash_log_tail(&log_path);
|
||||
if budget.record(Instant::now()) {
|
||||
eprintln!(
|
||||
"崩溃预算耗尽（{} 秒窗口内 {} 次崩溃），放弃自动重启，可用 `oxide server cleanup` 检查",
|
||||
CRASH_WINDOW.as_secs(),
|
||||
CRASH_LIMIT
|
||||
);
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
std::thread::sleep(CRASH_BACKOFF);
|
||||
match spawn_detached_server(workers) {
|
||||
Ok(new_child) => {
|
||||
child = new_child;
|
||||
child_exited = None;
|
||||
}
|
||||
Err(err) => {
|
||||
eprintln!("重启 server 失败：{err}");
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
}
|
||||
if let Err(msg) = wait_server_ready(READY_DEADLINE) {
|
||||
eprintln!("新 {msg}（可用 `oxide server cleanup` 人工检查）");
|
||||
return ExitCode::FAILURE;
|
||||
}
|
||||
// 重启就绪探测通过：新 server 已可服务，外部操作（关闭请求）
|
||||
// 在此之后才安全；该行与初始标记同文，是监控循环恢复的标记。
|
||||
println!("server 已就绪，watchdog 开始监控");
|
||||
}
|
||||
WatchState::Exited => {
|
||||
// 防御性兜底：sidecar 缺失分支已两次确认后返回，到达此处即
|
||||
// 信号采集与裁决不一致。
|
||||
println!("server 已正常退出，watchdog 退出");
|
||||
return ExitCode::SUCCESS;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// 打印日志文件最后 20 行（崩溃诊断）；文件缺失或为空不视为错误。
|
||||
///
|
||||
/// # 边界与前提
|
||||
/// - server 可能在写日志前就崩了，空日志与缺失文件都只跳过打印。
|
||||
fn print_crash_log_tail(log_path: &Path) {
|
||||
let lines = server_log::read_log(log_path, None, Some(CRASH_LOG_LINES)).unwrap_or_default();
|
||||
if lines.is_empty() {
|
||||
return;
|
||||
}
|
||||
println!("日志尾部（最后 {CRASH_LOG_LINES} 行）：");
|
||||
for line in lines {
|
||||
println!("{line}");
|
||||
}
|
||||
}
|
||||
|
||||
/// 优雅关闭：尽力向 server 发关闭请求（忽略错误，server 可能已死），打印退 0。
|
||||
fn graceful_stop() -> ExitCode {
|
||||
let _ = send_control_request(&ServerRequest::Shutdown);
|
||||
println!("server 已停止，watchdog 退出");
|
||||
ExitCode::SUCCESS
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
|
||||
/// 测试身份：字段为占位值（decide 只看 Option 与两个布尔信号）。
|
||||
fn test_identity() -> ServerIdentity {
|
||||
ServerIdentity {
|
||||
pid: 1,
|
||||
socket_path: "/tmp/oxide-test.sock".into(),
|
||||
version: "0.0.0".into(),
|
||||
started_at: 0,
|
||||
starttime: 0,
|
||||
}
|
||||
}
|
||||
|
||||
/// decide 四行裁决表：sidecar 缺失判 Exited、PID 死判 Crash、PID 活刻度
|
||||
/// 匹配判 Alive、PID 活刻度失配判 Crash。
|
||||
#[test]
|
||||
fn decide_four_row_table() {
|
||||
let id = test_identity();
|
||||
assert_eq!(decide(None, false, false), WatchState::Exited, "sidecar 缺失应判 Exited");
|
||||
assert_eq!(decide(Some(&id), false, false), WatchState::Crash, "PID 死应判 Crash");
|
||||
assert_eq!(decide(Some(&id), true, true), WatchState::Alive, "PID 活刻度匹配应判 Alive");
|
||||
assert_eq!(decide(Some(&id), true, false), WatchState::Crash, "PID 活刻度失配应判 Crash");
|
||||
}
|
||||
|
||||
/// 预算窗口语义：上限内未耗尽，窗口外条目被清后不再算入。
|
||||
#[test]
|
||||
fn budget_window_expiry() {
|
||||
let mut budget = CrashBudget::with(Duration::from_millis(100), 2);
|
||||
let t0 = Instant::now();
|
||||
assert!(!budget.record(t0), "首次记录不应耗尽");
|
||||
assert!(!budget.record(t0), "窗口内未超上限不应耗尽");
|
||||
// 窗口过期后旧条目被清：新记录不再与旧条目叠加。
|
||||
std::thread::sleep(Duration::from_millis(150));
|
||||
let t1 = t0 + Duration::from_millis(150);
|
||||
assert!(!budget.record(t1), "窗口外条目应被清，不应耗尽");
|
||||
}
|
||||
|
||||
/// 第 6 次崩溃判耗尽（默认窗口 5 次上限）。
|
||||
#[test]
|
||||
fn budget_exhausted_on_sixth() {
|
||||
let mut budget = CrashBudget::new();
|
||||
let start = Instant::now();
|
||||
for _ in 0..CRASH_LIMIT {
|
||||
assert!(!budget.record(start), "上限内不应耗尽");
|
||||
}
|
||||
assert!(budget.record(start), "第 6 次应判耗尽");
|
||||
}
|
||||
|
||||
/// 预算耗尽后的 record 仍返回耗尽（幂等）。
|
||||
#[test]
|
||||
fn budget_exhausted_is_idempotent() {
|
||||
let mut budget = CrashBudget::with(Duration::from_secs(60), 2);
|
||||
let start = Instant::now();
|
||||
budget.record(start);
|
||||
budget.record(start);
|
||||
assert!(budget.record(start), "第 3 次应判耗尽");
|
||||
assert!(budget.record(start), "耗尽后 record 应仍返回耗尽");
|
||||
}
|
||||
|
||||
/// 记录时刻单调推进时窗口滑动：仅窗口内条目计入。
|
||||
#[test]
|
||||
fn budget_sliding_window_counts_only_in_window() {
|
||||
let mut budget = CrashBudget::with(Duration::from_millis(100), 3);
|
||||
let t0 = Instant::now();
|
||||
budget.record(t0);
|
||||
budget.record(t0);
|
||||
std::thread::sleep(Duration::from_millis(150));
|
||||
let t1 = t0 + Duration::from_millis(150);
|
||||
// 窗口外旧条目被清：窗口内 3 条恰达上限不耗尽，第 4 条耗尽。
|
||||
for _ in 0..3 {
|
||||
assert!(!budget.record(t1), "窗口内未超上限不应耗尽");
|
||||
}
|
||||
assert!(budget.record(t1), "窗口内第 4 条应判耗尽");
|
||||
}
|
||||
}
|
||||
@@ -1,16 +1,18 @@
|
||||
//! server CLI 集成测试：真实二进制的分派面（help 列子命令、无 server 时
|
||||
//! 控制臂与 forge 退 1、cleanup 幂等、start 与 restart 端到端、forge 端到端）。
|
||||
//! 控制臂与 forge 退 1、cleanup 幂等、start 与 restart 端到端、forge 端到端、
|
||||
//! watchdog 崩溃自动重启与 SIGINT 优雅停 server）。
|
||||
//!
|
||||
//! start / restart 端到端与 cleanup 走 well-known 全局路径（每用户单例）：
|
||||
//! 测试先探活，socket 存活即 panic 不抢占存活 server；三枚触碰全局路径的
|
||||
//! 测试经同一把锁串行，避免相互干扰。
|
||||
//! start / restart / watchdog 端到端与 cleanup 走 well-known 全局路径
|
||||
//! （每用户单例）：测试先探活，socket 存活即 panic 不抢占存活 server；
|
||||
//! 触碰全局路径的测试经同一把锁串行，避免相互干扰。
|
||||
|
||||
use std::io::{BufReader, Write};
|
||||
use std::io::{BufRead, BufReader, Write};
|
||||
use std::os::unix::net::UnixStream;
|
||||
use std::process::{Command, Output};
|
||||
use std::sync::Mutex;
|
||||
use std::time::{Duration, Instant};
|
||||
|
||||
use oxide_cli::server::client::send_control_request;
|
||||
use oxide_cli::server::protocol::{self, FrameReader, ServerRequest, ServerResponse};
|
||||
use oxide_cli::server::sidecar;
|
||||
|
||||
@@ -25,6 +27,31 @@ fn oxide(args: &[&str]) -> Output {
|
||||
.expect("failed to run oxide")
|
||||
}
|
||||
|
||||
/// 向进程发信号（与 sidecar liveness 探活同一 `kill` 二进制路径）。
|
||||
fn kill_process(pid: u32, signal: &str) {
|
||||
Command::new("kill")
|
||||
.args([signal, &pid.to_string()])
|
||||
.status()
|
||||
.expect("发送信号应成功");
|
||||
}
|
||||
|
||||
/// 轮询健康请求直至得 healthy（100 毫秒间隔、给定截止）。
|
||||
///
|
||||
/// sidecar 出现早于 socket 绑定与池预热，不能直接作为就绪信号；watchdog
|
||||
/// 的初始就绪探测与测试的崩溃模拟都以健康请求通过为界。
|
||||
fn wait_for_healthy(deadline: Duration) {
|
||||
let end = Instant::now() + deadline;
|
||||
loop {
|
||||
let ready =
|
||||
matches!(send_control_request(&ServerRequest::Health), Ok(ServerResponse::Health { healthy: true }));
|
||||
if ready {
|
||||
return;
|
||||
}
|
||||
assert!(Instant::now() < end, "server 应在有界时间内就绪");
|
||||
std::thread::sleep(Duration::from_millis(100));
|
||||
}
|
||||
}
|
||||
|
||||
/// 发送请求帧并读回一帧响应。
|
||||
fn send_and_recv(stream: &mut UnixStream, request: &ServerRequest) -> ServerResponse {
|
||||
stream
|
||||
@@ -35,7 +62,7 @@ fn send_and_recv(stream: &mut UnixStream, request: &ServerRequest) -> ServerResp
|
||||
protocol::parse_response(&frame).expect("响应帧解析应成功")
|
||||
}
|
||||
|
||||
/// help 列全部十个子命令。
|
||||
/// help 列全部十一个子命令。
|
||||
#[test]
|
||||
fn server_help_lists_subcommands() {
|
||||
let output = oxide(&["server", "--help"]);
|
||||
@@ -43,6 +70,7 @@ fn server_help_lists_subcommands() {
|
||||
let stdout = String::from_utf8_lossy(&output.stdout);
|
||||
for name in [
|
||||
"start", "stop", "status", "health", "info", "version", "cleanup", "restart", "log", "forge",
|
||||
"watchdog",
|
||||
] {
|
||||
assert!(stdout.contains(name), "help 应列出 {name}：{stdout}");
|
||||
}
|
||||
@@ -432,3 +460,152 @@ fn server_forge_e2e() {
|
||||
assert!(!socket.exists(), "socket 文件应被删除");
|
||||
assert!(!sidecar_path.exists(), "sidecar 文件应被删除");
|
||||
}
|
||||
|
||||
/// watchdog 端到端：拉起 watchdog 前台进程（其拉起 server），`kill -9` 模拟
|
||||
/// 崩溃，断言 watchdog 自动重启（新 sidecar 进程号不同、新 server 健康），
|
||||
/// 关闭后 watchdog 退 0 且 stdout 含崩溃提示与日志尾部。
|
||||
#[test]
|
||||
fn server_watchdog_e2e() {
|
||||
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");
|
||||
|
||||
// 拉起 watchdog 前台进程（stdout 管道捕获崩溃诊断输出）。
|
||||
let mut child = Command::new(env!("CARGO_BIN_EXE_oxide"))
|
||||
.args(["server", "watchdog", "--workers", "2"])
|
||||
.stdout(std::process::Stdio::piped())
|
||||
.spawn()
|
||||
.expect("oxide server watchdog 应可启动");
|
||||
|
||||
// 后台线程消费 stdout 收集全部行，见到「开始监控」标记行即通知：
|
||||
// 该标记行蕴含 watchdog 已越过初始就绪探测，崩溃模拟在此之后进行
|
||||
// 才不与初始探测窗口竞争（预热期间杀 server 会让探测超时退 1、不重启）。
|
||||
let stdout_pipe = child.stdout.take().expect("stdout 管道应可取");
|
||||
let (monitoring_tx, monitoring_rx) = std::sync::mpsc::channel::<()>();
|
||||
let out_handle = std::thread::spawn(move || {
|
||||
let mut reader = BufReader::new(stdout_pipe);
|
||||
let mut lines = Vec::new();
|
||||
let mut line = String::new();
|
||||
loop {
|
||||
line.clear();
|
||||
if reader.read_line(&mut line).unwrap_or(0) == 0 {
|
||||
break;
|
||||
}
|
||||
lines.push(line.clone());
|
||||
if line.contains("watchdog 开始监控") {
|
||||
let _ = monitoring_tx.send(());
|
||||
}
|
||||
}
|
||||
lines
|
||||
});
|
||||
|
||||
// 轮询 sidecar 出现（10 秒截止）判 server 已启动。
|
||||
let deadline = Instant::now() + Duration::from_secs(10);
|
||||
while sidecar::read_identity(&sidecar_path).is_none() {
|
||||
assert!(Instant::now() < deadline, "sidecar 未出现");
|
||||
std::thread::sleep(Duration::from_millis(50));
|
||||
}
|
||||
let first_id = sidecar::read_identity(&sidecar_path).expect("读 sidecar 应成功");
|
||||
|
||||
// 等 watchdog 越过初始就绪探测进入监控循环（30 秒截止）。
|
||||
monitoring_rx
|
||||
.recv_timeout(Duration::from_secs(30))
|
||||
.expect("watchdog 应在有界时间内进入监控循环");
|
||||
|
||||
// 模拟崩溃：kill -9（文件残留、无退出序列，与 panic=abort 消亡等价）。
|
||||
kill_process(first_id.pid, "-9");
|
||||
|
||||
// 轮询新 sidecar 出现且进程号不同（30 秒截止、50 毫秒间隔）判 watchdog
|
||||
// 已重启。
|
||||
let deadline = Instant::now() + Duration::from_secs(30);
|
||||
let second_id = loop {
|
||||
if let Some(id) = sidecar::read_identity(&sidecar_path) {
|
||||
if id.pid != first_id.pid {
|
||||
break id;
|
||||
}
|
||||
}
|
||||
assert!(Instant::now() < deadline, "watchdog 应在有界时间内重启 server");
|
||||
std::thread::sleep(Duration::from_millis(50));
|
||||
};
|
||||
|
||||
// 等 watchdog 越过重启就绪探测（30 秒截止）：关闭请求发给探测完成前的
|
||||
// 新 server 会让它优雅退出、删 sidecar，与 watchdog 的重启探测竞争
|
||||
// （探测超时退 1、不重启）。
|
||||
monitoring_rx
|
||||
.recv_timeout(Duration::from_secs(30))
|
||||
.expect("watchdog 应在有界时间内完成重启探测");
|
||||
|
||||
// 连接新 server：健康请求得 healthy。
|
||||
let mut stream = UnixStream::connect(&socket).expect("连接新 server 应成功");
|
||||
match send_and_recv(&mut stream, &ServerRequest::Health) {
|
||||
ServerResponse::Health { healthy } => assert!(healthy, "新 server 应健康"),
|
||||
other => panic!("应得 Health 帧，实得 {other:?}"),
|
||||
}
|
||||
|
||||
// 发关闭请求得确认帧，轮询 sidecar 消失（watchdog 判 Exited 退 0）。
|
||||
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));
|
||||
}
|
||||
|
||||
// 等 watchdog 子进程退出断言退出码 0，收集 stdout 断言含崩溃提示与日志尾部。
|
||||
let status = child.wait().expect("等待 watchdog 应成功");
|
||||
assert_eq!(status.code(), Some(0), "watchdog 应以退出码 0 退出：{status:?}");
|
||||
let lines = out_handle.join().expect("stdout 读取线程应可汇合");
|
||||
let stdout = lines.join("");
|
||||
assert!(stdout.contains("server 已崩溃"), "watchdog 应打印崩溃提示：{stdout}");
|
||||
assert!(stdout.contains("日志尾部"), "watchdog 应打印日志尾部：{stdout}");
|
||||
assert_ne!(second_id.pid, first_id.pid, "重启后的 server 进程号应不同");
|
||||
}
|
||||
|
||||
/// watchdog SIGINT：拉起 watchdog 前台进程后向其发 SIGINT，断言 watchdog 退 0、
|
||||
/// sidecar 消失、server 进程已退出（停 watchdog 即优雅停 server）。
|
||||
#[test]
|
||||
fn server_watchdog_sigint_stops_server() {
|
||||
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");
|
||||
|
||||
// 拉起 watchdog 前台进程。
|
||||
let mut child = Command::new(env!("CARGO_BIN_EXE_oxide"))
|
||||
.args(["server", "watchdog", "--workers", "2"])
|
||||
.spawn()
|
||||
.expect("oxide server watchdog 应可启动");
|
||||
|
||||
// 轮询 sidecar 出现（10 秒截止）判 server 已启动。
|
||||
let deadline = Instant::now() + Duration::from_secs(10);
|
||||
while sidecar::read_identity(&sidecar_path).is_none() {
|
||||
assert!(Instant::now() < deadline, "sidecar 未出现");
|
||||
std::thread::sleep(Duration::from_millis(50));
|
||||
}
|
||||
let id = sidecar::read_identity(&sidecar_path).expect("读 sidecar 应成功");
|
||||
|
||||
// 等健康请求通过判 server 就绪：watchdog 已越过初始就绪探测进入监控
|
||||
// 循环，SIGINT 的优雅关闭请求发给已就绪的 server 即被处理。
|
||||
wait_for_healthy(Duration::from_secs(60));
|
||||
|
||||
// 向 watchdog 发 SIGINT。
|
||||
kill_process(child.id(), "-INT");
|
||||
|
||||
// 等 watchdog 退出断言退出码 0。
|
||||
let status = child.wait().expect("等待 watchdog 应成功");
|
||||
assert_eq!(status.code(), Some(0), "watchdog 应以退出码 0 退出：{status:?}");
|
||||
|
||||
// server 被优雅停止：sidecar 消失、进程已退出。
|
||||
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!(!sidecar::pid_alive(id.pid), "server 进程应已退出");
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user