87 lines
3.4 KiB
Go
87 lines
3.4 KiB
Go
// Package main 是 im-service 的进程入口，只做配置读取、依赖装配和生命周期托管。
|
||
package main
|
||
|
||
import (
|
||
"context"
|
||
"flag"
|
||
"log"
|
||
"os"
|
||
"os/signal"
|
||
"syscall"
|
||
"time"
|
||
|
||
"hyapp/services/im-service/internal/app"
|
||
"hyapp/services/im-service/internal/config"
|
||
)
|
||
|
||
// main 只负责装配并启动 im-service。
|
||
func main() {
|
||
// config 参数只决定当前进程读取哪份 YAML；真实服务边界由 internal/app 统一装配。
|
||
configPath := flag.String("config", "services/im-service/configs/config.yaml", "path to im-service config")
|
||
flag.Parse()
|
||
|
||
// 配置加载失败直接终止进程，避免以不完整端口、JWT secret 或 room-service 地址启动。
|
||
cfg, err := config.Load(*configPath)
|
||
if err != nil {
|
||
log.Fatal(err)
|
||
}
|
||
|
||
// App.New 会建立 room-service gRPC 连接、HTTP/WS listener、内部 gRPC listener 和健康检查状态。
|
||
application, err := app.New(cfg)
|
||
if err != nil {
|
||
log.Fatal(err)
|
||
}
|
||
|
||
// 使用带缓冲 signal channel，第二次信号可以在优雅下线未完成时触发强制收敛。
|
||
signals := make(chan os.Signal, 2)
|
||
signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
|
||
defer signal.Stop(signals)
|
||
|
||
// Run 阻塞在 HTTP 与 gRPC 两个 listener 上；这里用错误通道把运行期失败反馈到主 goroutine。
|
||
errCh := make(chan error, 1)
|
||
go func() {
|
||
errCh <- application.Run()
|
||
}()
|
||
|
||
select {
|
||
case sig := <-signals:
|
||
// 第一次信号进入 draining，ready 会立即失败，随后按配置的总预算执行优雅下线。
|
||
start := time.Now()
|
||
log.Printf("service=im-service node_id=%s signal=%s state=draining ready=false active_ws_connections=%d shutdown_deadline_ms=%d", cfg.NodeID, sig, application.ActiveWSConnections(), cfg.ShutdownTimeout.Milliseconds())
|
||
|
||
ctx, cancel := context.WithTimeout(context.Background(), cfg.ShutdownTimeout)
|
||
defer cancel()
|
||
|
||
shutdownDone := make(chan error, 1)
|
||
go func() {
|
||
shutdownDone <- application.Shutdown(ctx)
|
||
}()
|
||
|
||
select {
|
||
case err := <-shutdownDone:
|
||
if err != nil {
|
||
// 优雅下线失败也要带上耗时和结果，方便 supervisor 日志定位截断点。
|
||
log.Printf("service=im-service node_id=%s state=stopped result=shutdown_error duration_ms=%d error=%q", cfg.NodeID, time.Since(start).Milliseconds(), err.Error())
|
||
log.Fatal(err)
|
||
}
|
||
case sig := <-signals:
|
||
// 第二次信号允许加速退出，但必须先打印仍未完成的 WS 连接数量。
|
||
active := application.ActiveWSConnections()
|
||
log.Printf("service=im-service node_id=%s signal=%s state=draining result=second_signal active_ws_connections=%d", cfg.NodeID, sig, active)
|
||
forced := application.ForceClose()
|
||
log.Printf("service=im-service node_id=%s state=stopped result=forced active_ws_connections=%d duration_ms=%d", cfg.NodeID, forced, time.Since(start).Milliseconds())
|
||
case <-ctx.Done():
|
||
// shutdown_timeout 到期后强制关闭剩余连接，并记录被截断的资源数量。
|
||
active := application.ActiveWSConnections()
|
||
log.Printf("service=im-service node_id=%s state=draining result=shutdown_timeout active_ws_connections=%d", cfg.NodeID, active)
|
||
forced := application.ForceClose()
|
||
log.Printf("service=im-service node_id=%s state=stopped result=forced active_ws_connections=%d duration_ms=%d", cfg.NodeID, forced, time.Since(start).Milliseconds())
|
||
}
|
||
case err := <-errCh:
|
||
// 任一入口异常退出都视为进程级失败，避免节点处于半工作状态。
|
||
if err != nil {
|
||
log.Fatal(err)
|
||
}
|
||
}
|
||
}
|