golang 執行時死鎖排查和檢測

Berkaroad發表於2024-03-20
Golang

當執行的系統發生goroutine等待獲取鎖時間超過預期時，判定為發生了死鎖。因目前程式碼中使用了一些公開的鎖例項，呼叫鏈也比較長，對問題排查帶來了很大困擾。為了便於問題排查，需要藉助工具來實現。

1. 發生死鎖的判定依據和原因

1.1 判定依據

如下為使用Mutex鎖產生的鎖等待，並持續了222分鐘。這類不合理現象，就判定為發生了死鎖。（以下為cpu-pprof匯出的stack資訊）

goroutine 61508 [semacquire, 222 minutes]:
sync.runtime_SemacquireMutex(0xc00048f564, 0x4f4300, 0x1)
    /opt/go/src/runtime/sema.go:71 +0x47
sync.(*Mutex).lockSlow(0xc00048f560)
    /opt/go/src/sync/mutex.go:138 +0x105
sync.(*Mutex).Lock(...)
    /opt/go/src/sync/mutex.go:81
......

1.2 產生死鎖的原因

  • 鎖重入

  • 多把鎖加鎖，併發訪問時加鎖順序不一致

  • 鎖未釋放

2. 死鎖問題診斷

常用的標準庫鎖： sync.Mutex, sync.RWMutex，目標暫定為以檢測這兩類鎖的死鎖問題。

檢測死鎖的方法：分為靜態程式碼分析、執行期分析。

2.1 靜態程式碼分析

  • 優勢：

    可以做到場景全覆蓋。

  • 劣勢：

    好像沒有劣勢，但開發難度較高，目前發現的一個開源工具 go-tools，驗證下來無法檢測互鎖問題，但可以檢測鎖未釋放的問題。

2.2 執行期分析

  • 優勢：

    可以快速定位問題，檢測相對容易。

  • 劣勢：

    對程式碼有侵入性，需要封裝標準庫的鎖，對效能會有一定影響，應採取條件編譯方式，生產環境勿啟用。

3. 執行期分析工具開發

3.1 目標

  • 支援 Mutex 和 RWMutex

  • 支援鎖重入、互鎖問題的檢測

3.2 用法

import (
    "github.com/berkaroad/detectlock-go"
)

// 應用啟動時，設定啟用除錯
detectlock.EnableDebug()

// 宣告 sync.Mutex、sync.RWMutex 替換為 detectlock.Mutex、detectlock.RWMutex
var locker1 *detectlock.Mutex = &detectlock.Mutex{}
var locker2 *detectlock.RWMutex = &detectlock.RWMutex{}

// 非同步檢測死鎖
items := detectlock.Items()
fmt.Println(detectlock.DetectAcquired(items)) // 檢測獲得鎖的goroutine列表
fmt.Println(detectlock.DetectReentry(items)) // 檢測鎖重入的goroutine列表
fmt.Println(detectlock.DetectLockedEachOther(items)) // 檢測互鎖的goroutine列表

// 關閉除錯，並清理鎖使用資訊
detectlock.DisableDebug()

3.3 死鎖示例資料

  • 檢測 sync.Mutex 多把鎖順序不一致導致的互鎖
--- DetectAcquired ---
goroutine 29: [(0xc0000b4008, acquired, main.B (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:30)), (0xc0000b4000, wait, main.B (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:33))]
goroutine 30: [(0xc0000b4000, acquired, main.A (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:20)), (0xc0000b4008, wait, main.A (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:23))]

--- DetectLockedEachOther ---
goroutine 29: [(0xc0000b4008, acquired, main.B (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:30)), (0xc0000b4000, wait, main.B (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:33))]
goroutine 30: [(0xc0000b4000, acquired, main.A (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:20)), (0xc0000b4008, wait, main.A (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex01/main.go:23))]
  • 檢測 sync.Mutex 鎖重入
--- DetectAcquired ---
goroutine 53: [(0xc0000160b8, acquired, main.C (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex02/main.go:19)), (0xc0000160b8, wait, main.C (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex02/main.go:22))]

--- DetectReentry ---
goroutine 53: [(0xc0000160b8, acquired, main.C (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex02/main.go:19)), (0xc0000160b8, wait, main.C (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex02/main.go:22))]
  • 檢測 sync.Mutex、sync.RWMutex 多把鎖順序不一致導致的互鎖
--- DetectAcquired ---
goroutine 8: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 10: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 13: [(0xc0000160b8, acquired, main.E (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:30)), (0xc0000180c0, wait, main.E (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:33))]
goroutine 14: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 16: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 50: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 52: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 54: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 56: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 58: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]

--- DetectLockedEachOther ---
goroutine 8: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 10: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 13: [(0xc0000160b8, acquired, main.E (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:30)), (0xc0000180c0, wait, main.E (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:33))]
goroutine 14: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 16: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 50: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 52: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 54: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 56: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]
goroutine 58: [(0xc0000180c0, r-acquired, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:21)), (0xc0000160b8, wait, main.D (file: /home/berkaroad/github/berkaroad/detectlock-go/examples/mutex03/main.go:24))]

---------------------------------分割線-------------------------------------------------------------

我的detectlock-go專案: https://github.com/berkaroad/detectlock-go，已經掛在github上，有興趣的可以去了解下。

使用中有何問題，歡迎在github上給我提issue，謝謝！

相關文章