运行命令变成僵尸后，Docker容器拒绝被杀死

Z时代
2024-01-10
分类：问答

首先是第一件事。我的系统信息和版本：

$ lsb_release -a No LSB modules are available. Distributor ID: Ubuntu Description: Ubuntu 13.04 Release: 13.04 Codename: raring $ sudo docker version Client version: 0.9.0 Go version (client): go1.2.1 Git commit (client): 2b3fdf2 Server version: 0.9.0 Git commit (server): 2b3fdf2 Go version (server): go1.2.1 $ lxc-version lxc version: 0.9.0 $ uname -a Linux ip-10-0-2-86 3.8.0-19-generic #29-Ubuntu SMP Wed Apr 17 18:16:28 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

容器中的进程变成僵尸之后，我无法停止容器。升级到docker 0.9.0后，我在服务器上看到大量僵尸。例：

$ ps axo stat,ppid,pid,comm | grep -w defunct
Zl   25327 25332 node <defunct>
$ pstree -p
init(1)─┬
        ├─sh(819)───docker(831)─┬
                                ├─lxc-start(25327)───node(25332)───{node}(25378)

我可以看到 lxc-

start(25327)没有在节点进程25332上调用wait（）来保持僵尸存活。所以我检查了strace在做什么，它似乎卡在了epoll_wait。stract实际上一开始就卡住了，只是显示了这一点：

$sudo strace -ir -ttt -T -v -p 25327
Process 25327 attached - interrupt to quit (when asked to kill)
     0.000103 [    7fe59b9d34b3] epoll_wait(8,

但是在我运行sudo docker kill 3da5764b7bc9358之后，我得到了更多输出：

 0.000103 [    7fe59b9d34b3] epoll_wait(8, {{EPOLLIN, {u32=21673408, u64=21673408}}}, 10, 4294967295) = 1 <8.935002>
 8.935097 [    7fe59bcaff60] accept(4, 0, NULL) = 9 <0.000035>
 0.000095 [    7fe59bcafeb3] fcntl(9, F_SETFD, FD_CLOEXEC) = 0 <0.000027>
 0.000083 [    7fe59b9d401a] setsockopt(9, SOL_SOCKET, SO_PASSCRED, [1], 4) = 0 <0.000027>
 0.000089 [    7fe59b9d347a] epoll_ctl(8, EPOLL_CTL_ADD, 9, {EPOLLIN, {u32=21673472, u64=21673472}}) = 0 <0.000023>
 0.000087 [    7fe59b9d34b3] epoll_wait(8, {{EPOLLIN, {u32=21673472, u64=21673472}}}, 10, 4294967295) = 1 <0.000026>
 0.000090 [    7fe59bcb0130] recvmsg(9, {msg_name(0)=NULL, msg_iov(1)=[{"\3\0\0\0\0\0\0\0", 8}], msg_controllen=32, {cmsg_len=28, cmsg_level=SOL_SOCKET, cmsg_type=SCM_CREDENTIALS{pid=773, uid=0, gid=0}}, msg_flags=0}, 0) = 8 <0.000034>
 0.000128 [    7fe59bcb019d] sendto(9, "\0\0\0\0\0\0\0\0\364b\0\0\0\0\0\0\0\0\0\0\0\0\0\0", 24, 0, NULL, 0) = 24 <0.000029>
 0.000090 [    7fe59b9d34b3] epoll_wait(8, {{EPOLLIN|EPOLLHUP, {u32=21673472, u64=21673472}}}, 10, 4294967295) = 1 <0.000018>
 0.000091 [    7fe59bcb0130] recvmsg(9, {msg_name(0)=NULL, msg_iov(1)=[{"\3\0\0\0\0\0\0\0", 8}], msg_controllen=32, {cmsg_len=28, cmsg_level=SOL_SOCKET, cmsg_type=SCM_CREDENTIALS{pid=0, uid=0, gid=0}}, msg_flags=0}, 0) = 0 <0.000026>
 0.000122 [    7fe59b9d347a] epoll_ctl(8, EPOLL_CTL_DEL, 9, NULL) = 0 <0.000037>
 0.000084 [    7fe59bcafd00] close(9) = 0 <0.000048>
 0.000103 [    7fe59b9d34b3] epoll_wait(8, {{EPOLLIN, {u32=21673408, u64=21673408}}}, 10, 4294967295) = 1 <1.091839>
 1.091916 [    7fe59bcaff60] accept(4, 0, NULL) = 9 <0.000035>
 0.000093 [    7fe59bcafeb3] fcntl(9, F_SETFD, FD_CLOEXEC) = 0 <0.000027>
 0.000083 [    7fe59b9d401a] setsockopt(9, SOL_SOCKET, SO_PASSCRED, [1], 4) = 0 <0.000026>
 0.000090 [    7fe59b9d347a] epoll_ctl(8, EPOLL_CTL_ADD, 9, {EPOLLIN, {u32=21673504, u64=21673504}}) = 0 <0.000032>
 0.000100 [    7fe59b9d34b3] epoll_wait(8, {{EPOLLIN, {u32=21673504, u64=21673504}}}, 10, 4294967295) = 1 <0.000028>
 0.000088 [    7fe59bcb0130] recvmsg(9, {msg_name(0)=NULL, msg_iov(1)=[{"\3\0\0\0\0\0\0\0", 8}], msg_controllen=32, {cmsg_len=28, cmsg_level=SOL_SOCKET, cmsg_type=SCM_CREDENTIALS{pid=774, uid=0, gid=0}}, msg_flags=0}, 0) = 8 <0.000030>
 0.000125 [    7fe59bcb019d] sendto(9, "\0\0\0\0\0\0\0\0\364b\0\0\0\0\0\0\0\0\0\0\0\0\0\0", 24, 0, NULL, 0) = 24 <0.000032>
 0.000119 [    7fe59b9d34b3] epoll_wait(8, {{EPOLLIN|EPOLLHUP, {u32=21673504, u64=21673504}}}, 10, 4294967295) = 1 <0.000071>
 0.000139 [    7fe59bcb0130] recvmsg(9, {msg_name(0)=NULL, msg_iov(1)=[{"\3\0\0\0\0\0\0\0", 8}], msg_controllen=32, {cmsg_len=28, cmsg_level=SOL_SOCKET, cmsg_type=SCM_CREDENTIALS{pid=0, uid=0, gid=0}}, msg_flags=0}, 0) = 0 <0.000018>
 0.000112 [    7fe59b9d347a] epoll_ctl(8, EPOLL_CTL_DEL, 9, NULL) = 0 <0.000028>
 0.000076 [    7fe59bcafd00] close(9) = 0 <0.000027>
 0.000096 [    7fe59b9d34b3] epoll_wait(8,

然后我查看了epoll_wait正在等待什么，看起来像文件8（我猜这是从epoll_wait(8, {{EPOLLIN, {u32=21673408,

u64=21673408}}}, 10, 4294967295) = 1 <8.935002>形式上int epoll_wait(int epfd,

struct epoll_event *events, int maxevents, int timeout);

$ cat /proc/25327/fdinfo/8 pos: 0 flags: 02000002 tfd: 7 events: 19 data: 14ab830 tfd: 4 events: 19 data: 14ab5c0

还根据上面的tfd添加了7和4（不确定tfd的真正含义）

$ cat /proc/25327/fdinfo/4 pos: 0 flags: 02000002 $ cat /proc/25327/fdinfo/7 pos: 0 flags: 02000002 sigmask: fffffffe7ffbfab7 $ cd /proc/25327/fd $ ls -al lr-x------ 1 root root 64 Mar 13 22:28 0 -> /dev/null lrwx------ 1 root root 64 Mar 13 22:28 1 -> /dev/pts/17 lrwx------ 1 root root 64 Mar 13 22:28 2 -> /dev/pts/17 l-wx------ 1 root root 64 Mar 13 22:28 3 -> /var/log/lxc/3da5764b7bc935896a72abc9371ce68d4d658d8c70b56e1090aacb631080ec0e.log lrwx------ 1 root root 64 Mar 13 22:28 4 -> socket:[48415] lrwx------ 1 root root 64 Mar 14 00:03 5 -> /dev/ptmx lrwx------ 1 root root 64 Mar 14 00:03 6 -> /dev/pts/18 lrwx------ 1 root root 64 Mar 14 00:03 7 -> anon_inode:[signalfd] lrwx------ 1 root root 64 Mar 14 00:03 8 -> anon_inode:[eventpoll]

有关套接字的信息：

$ sudo netstat -anp | grep 48415 Proto RefCnt Flags Type State I-Node PID/Program name Path unix 2 [ ACC ] STREAM LISTENING 48415 25327/lxc-start @/var/lib/lxc/3da5764b7bc935896a72abc9371ce68d4d658d8c70b56e1090aacb631080ec0e/command

在docker.log中似乎确实存在一种通用模式，所有不停止的容器都具有以下签名：

2014/03/16 16:33:15 Container beb71548b3b23ba3337ca30c6c2efcbfcaf19d4638cf3d5ec5b8a3e4c5f1059a failed to exit within 0 seconds of SIGTERM - using the force 2014/03/16 16:33:25 Container SIGKILL failed to exit within 10 seconds of lxc-kill beb71548b3b2 - trying direct SIGKILL

在这一点上，我不知道下一步该怎么做。关于如何找出导致这些容器无法退出的原因的任何建议？我还应该收集其他数据吗？我也将SIGCHLD发送给该过程，但无济于事。

更多数据：将日志添加到节点进程的末尾，我们在容器中使用start命令开始：

Mon Mar 17 2014 20:52:52 GMT+0000 (UTC) process: main process = exit code: 0

这是来自docker的日志：

2014/03/17 20:52:52 Container f8a3d55e0f... failed to exit within 0 seconds of SIGTERM - using the force 2014/03/17 20:53:02 Container SIGKILL failed to exit within 10 seconds of lxc-kill f8a3d55e0fd8 - trying direct SIGKILL

时间戳显示进程已退出@ 20:52:52

使用本地和lxc泊坞窗驱动程序都会发生这种情况。

编辑：复制步骤！

将其变成bash脚本，然后运行并观看几乎50％的容器变成僵尸！

CNT=0
while true
do 
  echo $CNT
  DOCK=$(sudo docker run -d -t anandkumarpatel/zombie_bug ./node index.js)
  sleep 60 && sudo docker stop $DOCK > out.log &
  sleep 1
  CNT=$(($CNT+1))
  if [[ "$CNT" == "50" ]]; then
    exit
  fi
done

回答：

更改为最新内核可解决此问题

发现确切的内核差异：

REPRO：linux-image-3.8.0-31-generic

没有REPRO：linux-image-3.8.0-32-generic

我认为这是解决方法：

+++ linux-3.8.0/kernel/pid_namespace.c
@@ -181,6 +181,7 @@
    int nr;
    int rc;
    struct task_struct *task, *me = current;
+   int init_pids = thread_group_leader(me) ? 1 : 2;
    /* Don't allow any more processes into the pid namespace */
    disable_pid_allocation(pid_ns);
@@ -230,7 +231,7 @@
     */
    for (;;) {
        set_current_state(TASK_UNINTERRUPTIBLE);
-       if (pid_ns->nr_hashed == 1)
+       if (pid_ns->nr_hashed == init_pids)
            break;
        schedule();
    }