103 lines
8.1 KiB
Markdown
103 lines
8.1 KiB
Markdown
# cgroup 泄露
|
||
|
||
## 现象
|
||
|
||
创建 Pod 失败，运行时报错 `no space left on device`:
|
||
|
||
```txt
|
||
Dec 24 11:54:31 VM_16_11_centos dockerd[11419]: time="2018-12-24T11:54:31.195900301+08:00" level=error msg="Handler for POST /v1.31/containers/b98d4aea818bf9d1d1aa84079e1688cd9b4218e008c58a8ef6d6c3c106403e7b/start returned error: OCI runtime create failed: container_linux.go:348: starting container process caused \"process_linux.go:279: applying cgroup configuration for process caused \\\"mkdir /sys/fs/cgroup/memory/kubepods/burstable/pod79fe803c-072f-11e9-90ca-525400090c71/b98d4aea818bf9d1d1aa84079e1688cd9b4218e008c58a8ef6d6c3c106403e7b: no space left on device\\\"\": unknown"
|
||
```
|
||
|
||
## 内核 Bug
|
||
|
||
`memcg` 是 Linux 内核中用于管理 cgroup 内存的模块，整个生命周期应该是跟随 cgroup 的，但是在低版本内核中\(已知3.10\)，一旦给某个 memory cgroup 开启 kmem accounting 中的 `memory.kmem.limit_in_bytes` 就可能会导致不能彻底删除 memcg 和对应的 cssid，也就是说应用即使已经删除了 cgroup \(`/sys/fs/cgroup/memory` 下对应的 cgroup 目录已经删除\), 但在内核中没有释放 cssid，导致内核认为的 cgroup 的数量实际数量不一致，我们也无法得知内核认为的 cgroup 数量是多少。
|
||
|
||
关于 cgroup kernel memory，在 [kernel.org](https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v1/memory.html#kernel-memory-extension-config-memcg-kmem) 中有如下描述：
|
||
|
||
```
|
||
2.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM)
|
||
-----------------------------------------------
|
||
|
||
With the Kernel memory extension, the Memory Controller is able to limit
|
||
the amount of kernel memory used by the system. Kernel memory is fundamentally
|
||
different than user memory, since it can't be swapped out, which makes it
|
||
possible to DoS the system by consuming too much of this precious resource.
|
||
|
||
Kernel memory accounting is enabled for all memory cgroups by default. But
|
||
it can be disabled system-wide by passing cgroup.memory=nokmem to the kernel
|
||
at boot time. In this case, kernel memory will not be accounted at all.
|
||
|
||
Kernel memory limits are not imposed for the root cgroup. Usage for the root
|
||
cgroup may or may not be accounted. The memory used is accumulated into
|
||
memory.kmem.usage_in_bytes, or in a separate counter when it makes sense.
|
||
(currently only for tcp).
|
||
|
||
The main "kmem" counter is fed into the main counter, so kmem charges will
|
||
also be visible from the user counter.
|
||
|
||
Currently no soft limit is implemented for kernel memory. It is future work
|
||
to trigger slab reclaim when those limits are reached.
|
||
```
|
||
|
||
这是一个 cgroup memory 的扩展，用于限制对 kernel memory 的使用，但该特性在老于 4.0 版本中是个实验特性，存在泄露问题，在 4.x 较低的版本也还有泄露问题，应该是造成泄露的代码路径没有完全修复，推荐 4.3 以上的内核。
|
||
|
||
## 造成容器创建失败
|
||
|
||
这个问题可能会导致创建容器失败，因为创建容器为其需要创建 cgroup 来做隔离，而低版本内核有个限制：允许创建的 cgroup 最大数量写死为 65535 \([点我跳转到 commit](https://github.com/torvalds/linux/commit/38460b48d06440de46b34cb778bd6c4855030754#diff-c04090c51d3c6700c7128e84c58b1291R3384)\)，如果节点上经常创建和销毁大量容器导致创建很多 cgroup，删除容器但没有彻底删除 cgroup 造成泄露\(真实数量我们无法得知\)，到达 65535 后再创建容器就会报创建 cgroup 失败并报错 `no space left on device`，使用 kubernetes 最直观的感受就是 pod 创建之后无法启动成功。
|
||
|
||
pod 启动失败，报 event 示例:
|
||
|
||
``` bash
|
||
Events:
|
||
Type Reason Age From Message
|
||
---- ------ ---- ---- -------
|
||
Normal Scheduled 15m default-scheduler Successfully assigned jenkins/jenkins-7845b9b665-nrvks to 10.10.252.4
|
||
Warning FailedCreatePodContainer 25s (x70 over 15m) kubelet, 10.10.252.4 unable to ensure pod container exists: failed to create container for [kubepods besteffort podc6eeec88-8664-11e9-9524-5254007057ba] : mkdir /sys/fs/cgroup/memory/kubepods/besteffort/podc6eeec88-8664-11e9-9524-5254007057ba: no space left on device
|
||
```
|
||
|
||
dockerd 日志报错示例:
|
||
|
||
``` bash
|
||
Dec 24 11:54:31 VM_16_11_centos dockerd[11419]: time="2018-12-24T11:54:31.195900301+08:00" level=error msg="Handler for POST /v1.31/containers/b98d4aea818bf9d1d1aa84079e1688cd9b4218e008c58a8ef6d6c3c106403e7b/start returned error: OCI runtime create failed: container_linux.go:348: starting container process caused \"process_linux.go:279: applying cgroup configuration for process caused \\\"mkdir /sys/fs/cgroup/memory/kubepods/burstable/pod79fe803c-072f-11e9-90ca-525400090c71/b98d4aea818bf9d1d1aa84079e1688cd9b4218e008c58a8ef6d6c3c106403e7b: no space left on device\\\"\": unknown"
|
||
```
|
||
|
||
kubelet 日志报错示例:
|
||
|
||
``` bash
|
||
Sep 09 18:09:09 VM-0-39-ubuntu kubelet[18902]: I0909 18:09:09.449722 18902 remote_runtime.go:92] RunPodSandbox from runtime service failed: rpc error: code = Unknown desc = failed to start sandbox container for pod "osp-xxx-com-ljqm19-54bf7678b8-bvz9s": Error response from daemon: oci runtime error: container_linux.go:247: starting container process caused "process_linux.go:258: applying cgroup configuration for process caused \"mkdir /sys/fs/cgroup/memory/kubepods/burstable/podf1bd9e87-1ef2-11e8-afd3-fa163ecf2dce/8710c146b3c8b52f5da62e222273703b1e3d54a6a6270a0ea7ce1b194f1b5053: no space left on device\""
|
||
```
|
||
|
||
新版的内核限制为 `2^31` \(可以看成几乎不限制，[点我跳转到代码](https://github.com/torvalds/linux/blob/3120b9a6a3f7487f96af7bd634ec49c87ef712ab/kernel/cgroup/cgroup.c#L5233)\): `cgroup_idr_alloc()` 传入 end 为 0 到 `idr_alloc()`， 再传给 `idr_alloc_u32()`, end 的值最终被三元运算符 `end>0 ? end-1 : INT_MAX` 转成了 `INT_MAX` 常量，即 `2^31`。所以如果新版内核有泄露问题会更难定位，表现形式会是内存消耗严重，幸运的是新版内核已经修复，推荐 4.3 以上。
|
||
|
||
### 规避方案
|
||
|
||
如果你用的低版本内核\(比如 CentOS 7 v3.10 的内核\)并且不方便升级内核，可以通过不开启 kmem accounting 来实现规避，但会比较麻烦。
|
||
|
||
kubelet 和 runc 都会给 memory cgroup 开启 kmem accounting，所以要规避这个问题，就要保证kubelet 和 runc 都别开启 kmem accounting，下面分别进行说明:
|
||
|
||
#### runc
|
||
|
||
runc 在合并 [这个PR](https://github.com/opencontainers/runc/pull/1350/files) \(2017-02-27\) 之后创建的容器都默认开启了 kmem accounting，后来社区也注意到这个问题，并做了比较灵活的修复， [PR 1921](https://github.com/opencontainers/runc/pull/1921) 给 runc 增加了 "nokmem" 编译选项，缺省的 release 版本没有使用这个选项， 自己使用 nokmem 选项编译 runc 的方法:
|
||
|
||
``` bash
|
||
cd $GO_PATH/src/github.com/opencontainers/runc/
|
||
make BUILDTAGS="seccomp nokmem"
|
||
```
|
||
|
||
docker-ce v18.09.1 之后的 runc 默认关闭了 kmem accounting，所以也可以直接升级 docker 到这个版本之后。
|
||
|
||
#### kubelet
|
||
|
||
如果是 1.14 版本及其以上，可以在编译的时候通过 build tag 来关闭 kmem accounting:
|
||
|
||
``` bash
|
||
KUBE_GIT_VERSION=v1.14.1 ./build/run.sh make kubelet GOFLAGS="-tags=nokmem"
|
||
```
|
||
|
||
如果是低版本需要修改代码重新编译。kubelet 在创建 pod 对应的 cgroup 目录时，也会调用 libcontianer 中的代码对 cgroup 做设置，在 `pkg/kubelet/cm/cgroup_manager_linux.go` 的 `Create` 方法中，会调用 `Manager.Apply` 方法，最终调用 `vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/memory.go` 中的 `MemoryGroup.Apply` 方法，开启 kmem accounting。这里也需要进行处理，可以将这部分代码注释掉然后重新编译 kubelet。
|
||
|
||
## 参考资料
|
||
|
||
* 一行 kubernetes 1.9 代码引发的血案（与 CentOS 7.x 内核兼容性问题）: [http://dockone.io/article/4797](http://dockone.io/article/4797)
|
||
* Cgroup泄漏--潜藏在你的集群中: [https://tencentcloudcontainerteam.github.io/2018/12/29/cgroup-leaking/](https://tencentcloudcontainerteam.github.io/2018/12/29/cgroup-leaking/)
|