Google Cloud Kuberbetes 失控 systemd CPU 使用率 100%

Google Cloud Kuberbetes 失控 systemd CPU 使用率 100%

上周,在将我们的 GKE 集群升级到 Kubernetes 1.13.6-gke.13 后,集群中的所有节点都因 CPU 使用率过高而开始出现故障。节点本身上的 Kubernetes 软件占用了所有 CPU,而不是 pod。

top当我们通过 SSH 进入节点时显示的内容:

top - 10:11:27 up 5 days, 22 min,  1 user,  load average: 23.71, 21.90, 20.32
Tasks: 858 total,   3 running, 854 sleeping,   0 stopped,   1 zombie
%Cpu(s): 33.5 us, 30.9 sy,  0.0 ni, 35.5 id,  0.0 wa,  0.0 hi,  0.1 si,  0.0 st
MiB Mem :  30157.0 total,  14125.6 free,   4771.2 used,  11260.1 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.  24762.7 avail Mem

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
      1 root      20   0  114636  22564   4928 R  65.2   0.1   5254:53 systemd
    356 message+  20   0  110104  98784   2916 S  33.8   0.3   2705:49 dbus-daemon
   1236 root      20   0 3174876 147328  57464 S  22.0   0.5   1541:29 kubelet
    366 root      20   0   12020   3864   3396 S  21.6   0.0   1705:21 systemd-logind
   1161 root      20   0 5371428 586832  26756 S  18.7   1.9   1848:18 dockerd
    432 root      20   0 5585144  49040  13420 S  11.1   0.2 848:54.06 containerd
  23797 root      20   0  994620   8784   6088 S   3.0   0.0  96:58.79 containerd-shim
  45359 root      20   0  994620   8968   5600 S   3.0   0.0 100:28.46 containerd-shim
  35913 root      20   0 1068352   8192   5600 S   2.3   0.0 104:54.12 containerd-shim
  10806 root      20   0  994620   8908   5596 S   2.0   0.0 102:57.45 containerd-shim
  15378 root      20   0  994620   9084   5600 S   2.0   0.0 102:24.08 containerd-shim
  33141 root      20   0  994620   8856   5848 S   2.0   0.0  95:26.89 containerd-shim
  34299 root      20   0  994620   8824   5848 S   2.0   0.0  90:55.28 containerd-shim
  48411 root      20   0  994620   9488   6216 S   2.0   0.0  95:38.56 containerd-shim
1824641 root      20   0 1068352   6836   5716 S   2.0   0.0  65:45.81 containerd-shim
  10257 root      20   0  994620   9404   5596 S   1.6   0.0 101:10.45 containerd-shim
  15400 root      20   0 1068352   8916   6216 S   1.6   0.0  97:47.99 containerd-shim
  22895 root      20   0 1068352   8408   5472 S   1.6   0.0 102:55.97 containerd-shim
  29969 root      20   0  994620   9124   5600 S   1.6   0.0  98:32.32 containerd-shim
  34720 root      20   0  994620   8444   5472 S   1.6   0.0  97:23.98 containerd-shim
  10073 root      20   0 1068352   9796   6152 S   1.3   0.0 100:54.30 containerd-shim

在此之后,这些节点上的 pod 会出现 CPU 匮乏、运行不佳的情况,并由于活性检查不佳而被杀死。

为了尝试解决该问题,我们重新创建了所有节点。我们创建了一个具有同等资源的新池,并通过将旧池缩减为 0 个节点来迁移所有 pod。(这很困难,因为至少有 2 个以前的节点在很长时间后仍无法关闭。最后,我们不得不关闭底层虚拟机来杀死这些节点。)这没有帮助。

我们还尝试将节点从容器操作系统切换到常规 Ubuntu。这也没有帮助。

接下来,我们又创建了另一个池,这次 CPU 数量是原来的两倍。这有所帮助,但并没有解决根本问题。一些节点的 systemd、dbus-daemon、kubelet 等 CPU 使用率仍然非常高。

现在我们每个节点都运行着大量额外的 CPU,这掩盖了问题。除了有问题的系统服务外,还有足够的 CPU 来运行实际的 pod。

但整个系统仍然在浪费大量资源(和金钱),所以我们真的需要弄清楚这里到底发生了什么。什么事情systemd如此耗费 CPU?dbus-daemon在做什么?我们检查了各种日志(见下文),但没有明显的错误。

我们如何才能找出这里到底出了什么问题?


摘录自journalctl -u kubelet

Jul 04 05:49:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 05:49:34.849808    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 05:54:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 05:54:34.850598    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 05:59:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 05:59:34.851797    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:04:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:04:34.858344    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:09:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:09:34.859626    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:14:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:14:34.861142    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:19:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:19:34.862185    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:24:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:24:34.863160    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:29:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:29:34.864156    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:34:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:34:34.865041    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:39:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:39:34.866044    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:44:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:44:34.866969    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:49:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:49:34.867979    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:54:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:54:34.869429    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 06:59:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 06:59:34.870359    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:04:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:04:34.871190    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:09:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:09:34.872063    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:14:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:14:34.873240    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:19:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:19:34.874123    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:24:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:24:34.875010    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:29:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:29:34.876612    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:34:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:34:34.877420    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:39:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:39:34.878368    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:44:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:44:34.879221    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:49:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:49:34.880239    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:54:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:54:34.881172    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 07:59:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 07:59:34.881868    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 08:04:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 08:04:34.882653    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 08:09:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 08:09:34.883432    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 08:14:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 08:14:34.884175    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 08:19:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 08:19:34.885043    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 08:24:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 08:24:34.885845    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service
Jul 04 08:29:34 gke-cluster0-pool-1-9c29a343-vqtq kubelet[1236]: I0704 08:29:34.886690    1236 container_manager_linux.go:434] [ContainerManager]: Discovered runtime cgroups name: /system.slice/docker.service

输出自strace -c -p `pidof systemd`

strace: Process 1 attached
strace: Process 1 detached
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 24.20    0.109101           5     22827           lstat
 17.45    0.078654           9      8610           sendmsg
 15.51    0.069914           7     10549           read
 10.96    0.049406           3     17310           getpid
  6.45    0.029075           8      3806           openat
  6.08    0.027385           7      3783           readlinkat
  5.76    0.025945           3      7579           fstat
  4.47    0.020167           4      5700           getrandom
  3.62    0.016301           4      3806           close
  3.14    0.014133           7      1892           access
  2.28    0.010278           5      1924        11 stat
  0.03    0.000145           4        33           epoll_wait
  0.02    0.000089           4        22           readlink
  0.01    0.000029           3        11           prctl
  0.01    0.000029           3        11           clock_gettime
  0.01    0.000027           2        11           getgid
  0.01    0.000026           2        11           geteuid
  0.00    0.000020           2        11           getuid
  0.00    0.000020           2        11           getegid
------ ----------- ----------- --------- --------- ----------------
100.00    0.450744                 87907        11 total

strace lstat 是什么?下面是我从完整的 strace 输出中随机截取的一些相关代码片段:

...
stat("/dev/sds", {st_mode=S_IFBLK|0660, st_rdev=makedev(65, 32), ...}) = 0
readlinkat(AT_FDCWD, "/sys/dev/block/65:32", "../../devices/pci0000:00/0000:00"..., 99) = 83
lstat("/sys", {st_mode=S_IFDIR|0555, st_size=0, ...}) = 0
lstat("/sys/dev", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/dev/block", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:19", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:19/0:0:19:0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:19/0:0:19:0/block", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:19/0:0:19:0/block/sds", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
access("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:19/0:0:19:0/block/sds/uevent", F_OK) = 0
openat(AT_FDCWD, "/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:19/0:0:19:0/block/sds/uevent", O_RDONLY|O_CLOEXEC) = 16

...

lstat("/sys", {st_mode=S_IFDIR|0555, st_size=0, ...}) = 0
lstat("/sys/dev", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/dev/block", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:18", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:18/0:0:18:0", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:18/0:0:18:0/block", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
lstat("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:18/0:0:18:0/block/sdr", {st_mode=S_IFDIR|0755, st_size=0, ...}) = 0
access("/sys/devices/pci0000:00/0000:00:03.0/virtio0/host0/target0:0:18/0:0:18:0/block/sdr/uevent", F_OK) = 0

dbus-monitor --system --profile | head -n 20

dbus-monitor: unable to enable new-style monitoring: org.freedesktop.DBus.Error.AccessDenied: "Rejected send message, 1 matched rules; type="method_call", sender=":1.165" (uid=5004 pid=769854 comm="dbus-monitor --system --profile ") interface="org.freedesktop.DBus.Monitoring" member="BecomeMonitor" error name="(unset)" requested_reply="0" destination="org.freedesktop.DBus" (bus)". Falling back to eavesdropping.
#type    timestamp    serial    sender    destination    path    interface    member
#                    in_reply_to
sig    1562263380.765023    2    org.freedesktop.DBus    :1.165    /org/freedesktop/DBus    org.freedesktop.DBus    NameAcquired
sig    1562263380.953812    132870362    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d5d1be6e08bfe7552d6f9ee50a943eca88dd0dd749ec248594aa0be91879a2cdb_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.957890    132870363    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d5d1be6e08bfe7552d6f9ee50a943eca88dd0dd749ec248594aa0be91879a2cdb_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.957912    132870364    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d539cd8b8367913182e438aa1b3b05714c8f3f131e20bcadabdeb850c375a8125_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.957918    132870365    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d539cd8b8367913182e438aa1b3b05714c8f3f131e20bcadabdeb850c375a8125_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.957923    132870366    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2dcontainers_2d4c04ee8d1bf693ff2c9300b198b2b47bbf2c240265af5b9edc1f07b6cbd0c1ce_2dmounts_2dshm_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958014    132870367    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2dcontainers_2d4c04ee8d1bf693ff2c9300b198b2b47bbf2c240265af5b9edc1f07b6cbd0c1ce_2dmounts_2dshm_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958020    132870368    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d872dd2c0a63a9f3b5c9c5e4972e06fcf89d28b4c7f59aea7ea4d38f5a6bf0db6_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958025    132870369    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d872dd2c0a63a9f3b5c9c5e4972e06fcf89d28b4c7f59aea7ea4d38f5a6bf0db6_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958029    132870370    :1.157    <none>    /org/freedesktop/systemd1/unit/home_2dkubernetes_2dcontainerized_5fmounter_2drootfs_2dvar_2dlib_2dkubelet_2dpods_2d2f4e6eae_5cx2d9e51_5cx2d11e9_5cx2db4ee_5cx2d42010a80000f_2dvolumes_2dkubernetes_2eio_5cx7esecret_2dfluentd_5cx2dgcp_5cx2dtoken_5cx2dzfrkb_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958111    132870371    :1.157    <none>    /org/freedesktop/systemd1/unit/home_2dkubernetes_2dcontainerized_5fmounter_2drootfs_2dvar_2dlib_2dkubelet_2dpods_2d2f4e6eae_5cx2d9e51_5cx2d11e9_5cx2db4ee_5cx2d42010a80000f_2dvolumes_2dkubernetes_2eio_5cx7esecret_2dfluentd_5cx2dgcp_5cx2dtoken_5cx2dzfrkb_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958117    132870372    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2dkubelet_2dpods_2d2f4e6eae_5cx2d9e51_5cx2d11e9_5cx2db4ee_5cx2d42010a80000f_2dvolumes_2dkubernetes_2eio_5cx7esecret_2dfluentd_5cx2dgcp_5cx2dtoken_5cx2dzfrkb_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958121    132870373    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2dkubelet_2dpods_2d2f4e6eae_5cx2d9e51_5cx2d11e9_5cx2db4ee_5cx2d42010a80000f_2dvolumes_2dkubernetes_2eio_5cx7esecret_2dfluentd_5cx2dgcp_5cx2dtoken_5cx2dzfrkb_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958126    132870374    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d667d713fe45ea0af609c85fbfd3aafbca9494574fe10509bda8cd3d13d1e6b66_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958223    132870375    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d667d713fe45ea0af609c85fbfd3aafbca9494574fe10509bda8cd3d13d1e6b66_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958229    132870376    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d0a9f899f23c0965b43cae81dd04f46a78e4b1b063fa5679323146b06932474a9_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958233    132870377    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d0a9f899f23c0965b43cae81dd04f46a78e4b1b063fa5679323146b06932474a9_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged
sig    1562263380.958238    132870378    :1.157    <none>    /org/freedesktop/systemd1/unit/var_2dlib_2ddocker_2doverlay2_2d875734c124694cd54a7c26516f740cedcb853a18ff5805a89669747c1188b65d_2dmerged_2emount    org.freedesktop.DBus.Properties    PropertiesChanged

答案1

与 Google 合作后,我们终于解决了这个问题。简而言之,我们必须创建一个运行 GKE 1.12.8-gke.10 的新节点池,并使用持久卷将所有 pod 转移到该池。

长版本

这是我希望在调试时立即想到的事情,但我必须感谢谷歌支持团队的 Francisco 和 Lynn。

故障排除过程中的一个关键时刻是我们对负载进行分段。有一次,我们为每种类型的 Pod 指定了一个特定的节点池。这使我们能够回答这个问题:问题是否特定于某种类型的 Pod?我们怀疑安装活动与该问题有关,因此特别感兴趣的是查看具有持久卷的 Pod 是否与节点性能下降相关。

事实证明确实如此。只有运行带有卷的 pod 的节点才会发热。

根本原因

Google GKE 团队的假设是,Docker 在 17.03 和 18.09 版本之间的变化导致每当execpod 中有东西被“添加”时,就会产生大量的 systemd 活动。具体来说,他们的一名工程师发现了一个变化,runc这会导致所有挂载单元作为 exec 活跃度探测的一部分被重新加载。

因此,解决这个问题的秘诀有三个:

  1. Docker 18(GKE 13 使用)
  2. 每个节点安装多个卷
  3. 用于exec活性探测(或者其他我认为的)

我们每个节点运行大约 40 个 Redis pod。按照惯例,我们的 Redis pod 使用基于 exec 的存活探测器。此外,每个 pod 都有一个挂载卷用于永久存储。存活探测器每分钟运行一次。因此,我们每分钟有 40 个卷被重新加载 40 次。

解决问题

为了解决这个问题,我们使用节点污点隔离了所有 Redis 负载。我们创建了一个专用于 Redis 的新节点池,运行 GKE 1.12.8-gke.10。虽然 GKE 不允许您降级集群,但它允许您创建运行旧版本的新节点池。当然,我们禁用了这个新池的自动升级。

然后我们强制将所有 Redis pod 迁移到这个新池。

结果是立竿见影的:我们现在可以用 6 个 CPU 核心运行与之前用 24 个 CPU 核心一样多的负载,而且情况看起来很稳定。

答案2

这里需要考虑两件事:

  1. 要查看集群中是否发生任何问题,您需要检查日志

系统组件有两种类型:在容器中运行的组件和不在容器中运行的组件。例如:

Kubernetes 调度程序和 kube-proxy 在容器中运行。kubelet 和容器运行时(例如 Docker)不在容器中运行。在具有 systemd 的机器上,kubelet 和容器运行时会写入 journald。如果不存在 systemd,它们会写入 /var/log 目录中的 .log 文件。容器内的系统组件始终会写入 /var/log 目录,从而绕过默认日志记录机制。

例如,用于journalctl -u kubelet分析 kubelet 日志。您还可以检查pod 日志

  1. 如果没有错误,但您可能需要使用的资源仍然存在问题自动缩放
    • HPA 根据 CPU/内存使用情况扩展 pod 副本的数量。
    • 垂直 Pod 自动扩缩器 (VPA) 为现有 Pod 分配更多(或更少)的 CPU 或内存。
    • 集群自动扩缩器 (CA) 根据待处理的 Pod 扩缩集群节点。

使用上述一种或多种方法的组合来动态扩展您的集群,而无需手动设置。

如果有帮助的话请告诉我。

相关内容