开启Prefork的Go-Fiber应用RabbitMQ连接共享空指针异常排查

我有一个基于Go-Fiber的应用，尝试向RabbitMQ发布消息，简化后的代码如下：

main.go #

// main.go
type custom_handler struct {
    RMQConn *rabbitmq.Connection
}

func newHandler(conn *rabbitmq.Connection) custom_handler {
    return custom_handler{
        RMQConn: conn,
    }
}

func main() {
    var hndlr custom_handler
    if !fiber.IsChild() {
        messageq.ConnectMQ()
        hndlr = newHandler(messageq.RMQConn)
        log.Println(messageq.RMQConn) // not nil
    }

    engine := html.New("./static", ".html")
    app := fiber.New(fiber.Config{
        Views:     engine,
        Prefork:   true,  // ATTENTION!!!
    })

    app.Use(cors.New())
    app.Get("/", func(c *fiber.Ctx) error {
        return c.Render("index", fiber.Map{})
    })

    app.Post("/", hndlr.handleFileupload)

    log.Fatal(app.Listen(":8080"))
}

func (hndlr custom_handler) handleFileupload(c *fiber.Ctx) error {
    log.Println(hndlr.RMQConn) // nil
    RMQChan, err := hndlr.RMQConn.Channel()
    if err != nil {
        return err
    }
    defer RMQChan.Close()
}

messageq包代码 #

// messageq.go
var RMQConn *rabbitmq.Connection

func ConnectMQ() {
    var err error
    amqpServerURL := os.Getenv("RABBITMQ_SERVER_URL")
    RMQConn, err = rabbitmq.Dial(amqpServerURL)
    if err != nil {
        panic(err)
    }
}

运行应用时出现段错误，错误日志如下：

producer_1  | panic: runtime error: invalid memory address or nil pointer dereference
producer_1  | [signal SIGSEGV: segmentation violation code=0x1 addr=0x14 pc=0x8ef6db]
imgress-producer_1  |
producer_1  | goroutine 7 [running]:
producer_1  | github.com/rabbitmq/amqp091-go.(*Connection).allocateChannel(0x0)
producer_1  |   /go/pkg/mod/github.com/rabbitmq/amqp091-go@v1.5.0/connection.go:679 +0x5b
producer_1  | github.com/rabbitmq/amqp091-go.(*Connection).openChannel(0x4f46c8?)
producer_1  |   /go/pkg/mod/github.com/rabbitmq/amqp091-go@v1.5.0/connection.go:709 +0x25
producer_1  | github.com/rabbitmq/amqp091-go.(*Connection).Channel(...)
producer_1  |   /go/pkg/mod/github.com/rabbitmq/amqp091-go@v1.5.0/connection.go:736
producer_1  | main.ValidateAndPublish({0xc0000120c8, 0x1, 0xb7aeff?}, 0x1?, {0xc0000e7f38, 0x8}, 0xffffffffffffffff?)
producer_1  |   /app/validate.go:14 +0x8a
producer_1  | main.custom_handler.handleFileupload({0x0?, 0x0?}, 0xc00015f080?)
producer_1  |   /app/main.go:94 +0x4bb
producer_1  | github.com/gofiber/fiber/v2.(*App).next(0xc000221b80, 0xc00009e000)
producer_1  |   /go/pkg/mod/github.com/gofiber/fiber/v2@v2.39.0/router.go:132 +0x1c2
producer_1  | github.com/gofiber/fiber/v2.(*Ctx).Next(0xc0001bcf18?)
producer_1  |   /go/pkg/mod/github.com/gofiber/fiber/v2@v2.39.0/ctx.go:945 +0x53
producer_1  | github.com/gofiber/fiber/v2/middleware/cors.New.func1(0xc00009e000)
producer_1  |   /go/pkg/mod/github.com/gofiber/fiber/v2@v2.39.0/middleware/cors/cors.go:141 +0x286
producer_1  | github.com/gofiber/fiber/v2.(*App).next(0xc000221b80, 0xc00009e000)
producer_1  |   /go/pkg/mod/github.com/gofiber/fiber/v2@v2.39.0/router.go:132 +0x1c2
producer_1  | github.com/gofiber/fiber/v2.(*App).handler(0xc000221b80, 0x485df7?)
producer_1  |   /go/pkg/mod/github.com/gofiber/fiber/v2@v2.39.0/router.go:159 +0x45
producer_1  | github.com/valyala/fasthttp.(*Server).serveConn(0xc0001478c0, {0xc8dd90?, 0xc000012070})
producer_1  |   /go/pkg/mod/github.com/valyala/fasthttp@v1.40.0/server.go:2311 +0x1268
producer_1  | github.com/valyala/fasthttp.(*workerPool).workerFunc(0xc0000000a0, 0xc0000400a0)
producer_1  |   /go/pkg/mod/github.com/valyala/fasthttp@v1.40.0/workerpool.go:224 +0xa9
producer_1  | github.com/valyala/fasthttp.(*workerPool).getCh.func1()
producer_1  |   /go/pkg/mod/github.com/valyala/fasthttp@v1.40.0/workerpool.go:196 +0x38
producer_1  | created by github.com/valyala/fasthttp.(*workerPool).getCh
producer_1  |   /go/pkg/mod/github.com/valyala/fasthttp@v1.40.0/workerpool.go:195 +0x1b0

我推测是Prefork: true导致每个请求由独立goroutine处理，handler无法访问连接，但不确定具体忽略了什么，请问如何让RabbitMQ连接对handler可用？

你的核心问题是对Fiber的Prefork模式理解有误：Prefork是多进程模式，不是多goroutine模式。当开启Prefork: true时，Fiber会先启动一个父进程，然后fork出多个子进程来处理请求，每个子进程都是独立的操作系统进程，进程间的内存空间完全隔离，父进程中初始化的RabbitMQ连接无法被子进程共享。

你当前的代码只在父进程（!fiber.IsChild()）中初始化了hndlr和RabbitMQ连接，但子进程启动时不会执行这段初始化逻辑，导致子进程里的hndlr.RMQConn是nil，调用Channel()时触发空指针引用错误。

有两种常见的解决方式，根据你的业务需求选择：

方案1：关闭Prefork模式 #

如果你的应用不需要利用多核CPU的极致性能，或者可以通过Fiber的默认goroutine池处理并发，可以直接关闭Prefork：

app := fiber.New(fiber.Config{
    Views:     engine,
    Prefork:   false, // 关闭Prefork
})

此时应用运行在单进程多goroutine模式下，RabbitMQ连接可以安全地在goroutine间共享（RabbitMQ的Connection本身是线程安全的，多个goroutine可以共用一个Connection创建Channel）。

方案2：在每个子进程中初始化RabbitMQ连接 #

如果必须保留Prefork模式，需要让每个子进程都建立自己的RabbitMQ连接，修改初始化逻辑：

func main() {
    // 去掉fiber.IsChild()判断，让所有进程（父+子）都初始化连接
    messageq.ConnectMQ()
    hndlr := newHandler(messageq.RMQConn)
    log.Println(messageq.RMQConn)

    engine := html.New("./static", ".html")
    app := fiber.New(fiber.Config{
        Views:     engine,
        Prefork:   true,
    })

    // ... 后续路由注册逻辑不变
}

注意：每个子进程都会创建独立的RabbitMQ连接，这种方式会消耗更多的RabbitMQ连接资源，建议配合RabbitMQ的连接配置（如最大连接数）进行调整，同时确保进程退出时正确关闭连接。

另外，也可以优化messageq包，增加连接关闭的逻辑：

// messageq.go
var RMQConn *rabbitmq.Connection

func ConnectMQ() {
    var err error
    amqpServerURL := os.Getenv("RABBITMQ_SERVER_URL")
    RMQConn, err = rabbitmq.Dial(amqpServerURL)
    if err != nil {
        panic(err)
    }
}

func CloseMQ() {
    if RMQConn != nil {
        _ = RMQConn.Close()
    }
}

然后在main函数中添加退出回调：

func main() {
    messageq.ConnectMQ()
    defer messageq.CloseMQ() // 进程退出时关闭连接
    // ... 其他逻辑
}

内容的提问来源于stack exchange，提问作者Eziz Durdyyev