ruby-on-rails - 随机 unicorn 超时和拍打

Question

我在 Unicorn 和 Nginx(多个 Web 节点)上运行了一个非常高流量的 Rails 3.2 应用程序，但每隔一段时间我就会看到 Unicorn 工作人员开始超时并在所有节点上被 Unicorn 主控程序 sigkill。当然，当 Unicorn worker 被 Unicorn master sigkill 时，一个新的 worker 会被 fork 到它的位置，但它也会挂起 60 秒然后超时并被杀死。这基本上会反复发生，直到我硬杀死所有 unicorn 主人和 worker 。
Unicorn log:

E, [2013-04-18T12:57:50.007623 #14002] ERROR -- : worker=8 PID:14968 timeout (62s > 60s), killing
E, [2013-04-18T12:57:50.108364 #14002] ERROR -- : reaped #<Process::Status: pid 14968 SIGKILL (signal 9)> worker=8
I, [2013-04-18T12:57:50.489505 #15726]  INFO -- : worker=8 ready
E, [2013-04-18T12:57:52.175842 #14002] ERROR -- : worker=5 PID:15033 timeout (61s > 60s), killing
E, [2013-04-18T12:57:52.276586 #14002] ERROR -- : reaped #<Process::Status: pid 15033 SIGKILL (signal 9)> worker=5
I, [2013-04-18T12:57:52.653069 #15782]  INFO -- : worker=5 ready
E, [2013-04-18T12:57:56.340290 #14002] ERROR -- : worker=3 PID:15074 timeout (61s > 60s), killing
E, [2013-04-18T12:57:56.440993 #14002] ERROR -- : reaped #<Process::Status: pid 15074 SIGKILL (signal 9)> worker=3
I, [2013-04-18T12:57:56.809730 #15832]  INFO -- : worker=3 ready
E, [2013-04-18T12:57:57.504142 #14002] ERROR -- : worker=7 PID:15087 timeout (61s > 60s), killing
E, [2013-04-18T12:57:57.604886 #14002] ERROR -- : reaped #<Process::Status: pid 15087 SIGKILL (signal 9)> worker=7
I, [2013-04-18T12:57:57.983581 #15845]  INFO -- : worker=7 ready
E, [2013-04-18T12:57:59.669664 #14002] ERROR -- : worker=4 PID:15108 timeout (61s > 60s), killing
E, [2013-04-18T12:57:59.770427 #14002] ERROR -- : reaped #<Process::Status: pid 15108 SIGKILL (signal 9)> worker=4
I, [2013-04-18T12:58:00.155461 #15879]  INFO -- : worker=4 ready
E, [2013-04-18T12:58:06.839906 #14002] ERROR -- : worker=9 PID:15192 timeout (61s > 60s), killing
E, [2013-04-18T12:58:06.940829 #14002] ERROR -- : reaped #<Process::Status: pid 15192 SIGKILL (signal 9)> worker=9
I, [2013-04-18T12:58:07.302766 #15956]  INFO -- : worker=9 ready
E, [2013-04-18T12:58:08.003330 #14002] ERROR -- : worker=6 PID:15213 timeout (61s > 60s), killing
E, [2013-04-18T12:58:08.104006 #14002] ERROR -- : reaped #<Process::Status: pid 15213 SIGKILL (signal 9)> worker=6
I, [2013-04-18T12:58:08.466790 #15973]  INFO -- : worker=6 ready

监控系统显示外部服务(Postgres 数据库、Memcached、Redis)都在正确响应并且没有延迟问题。

以下是一些可能有值(value)的输出:

在这些中断期间，我注意到尝试连接到 Unicorn 套接字的大量积压。当站点未关闭时，通常以下命令仅返回一两行。
netstat | grep unic

....
unix  2      [ ]         STREAM     CONNECTING    0        /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTING    0        /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTING    0        /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTING    0        /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTING    0        /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7768134  /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7767311  /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7766999  /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7767309  /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7766941  /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7767287  /tmp/unicorn.sock
unix  2      [ ]         STREAM     CONNECTED     7766225  /tmp/unicorn.sock

任何人都知道可能导致这种情况的原因是什么？这会在多个服务器上同时发生。

Answer 1

DDOS？在探索应用程序级别之前，我会确保您知道网络级别会发生什么。它工作时有多少个连接？没有的时候有多少？来自哪些IP？等等。