PHPor 的Blog

对于kvm虚拟机，由于宿主机和guest之间的独立性很大，当guest把很多内存用于系统cache的话，宿主机也没有办法识别这部分内存，也没有办法建议guest让出一部分内存。

kvm有个内存气球的概念，原以为可以在不改变guest已定义的内存大小的情况下可以让guest让出一部分非关键性的内存占用，而实际上balloon却是在修改（变大和变小）guest的总内存大小，而且在减小guest内存大小的时候，也是武断地减少内存大小，一点儿不考虑guest的感受，不惜杀死进程来满足宿主机的请求，下面是逐步balloon的过程：

先让出free部分
free不够再让出cache部分（并非所有的buff/cache都是完全让出的）
实在还不够，杀进程

参考资料：

https://www.linux-kvm.org/page/Projects/auto-ballooning 这里有提到选项automatic=true

http://www.openstack.cn/?p=4540

环境：

# docker info
Containers: 56
 Running: 33
 Paused: 0
 Stopped: 23
Images: 31
Server Version: 1.12.5
Storage Driver: devicemapper
 Pool Name: data-docker_thinpool
 Pool Blocksize: 524.3 kB
 Base Device Size: 32.21 GB
 Backing Filesystem: xfs
 Data file:
 Metadata file:
 Data Space Used: 516.2 GB
 Data Space Total: 644.2 GB
 Data Space Available: 128.1 GB
 Metadata Space Used: 45.47 MB
 Metadata Space Total: 16.98 GB
 Metadata Space Available: 16.93 GB
 Thin Pool Minimum Free Space: 64.42 GB
 Udev Sync Supported: true
 Deferred Removal Enabled: true
 Deferred Deletion Enabled: false
 Deferred Deleted Device Count: 0
 Library Version: 1.02.135-RHEL7 (2016-11-16)
Logging Driver: journald
Cgroup Driver: cgroupfs
Plugins:
 Volume: local
 Network: host bridge null overlay
Swarm: inactive
Runtimes: runc docker-runc
Default Runtime: runc
Security Options: seccomp
Kernel Version: 3.10.0-514.6.1.el7.x86_64
Operating System: CentOS Linux 7 (Core)
OSType: linux
Architecture: x86_64
Number of Docker Hooks: 2
CPUs: 40
Total Memory: 94.14 GiB
Name: VM-2-10-12
ID: NYBE:NZML:4KQQ:PF2J:RXCB:IPPI:Y3BI:CY7E:RVAC:WVWV:VDM2:3EEK
Docker Root Dir: /data1/docker
Debug Mode (client): false
Debug Mode (server): false
Registry: https://index.docker.io/v1/
Insecure Registries:
 docker-registry.i.bbtfax.com:5000
 127.0.0.0/8
Registries: docker.io (secure)

# docker info

Containers: 56

Running: 33

Paused: 0

Stopped: 23

Images: 31

Server Version: 1.12.5

Storage Driver: devicemapper

Pool Name: data-docker_thinpool

Pool Blocksize: 524.3 kB

Base Device Size: 32.21 GB

Backing Filesystem: xfs

Data file:

Metadata file:

Data Space Used: 516.2 GB

Data Space Total: 644.2 GB

Data Space Available: 128.1 GB

Metadata Space Used: 45.47 MB

Metadata Space Total: 16.98 GB

Metadata Space Available: 16.93 GB

Thin Pool Minimum Free Space: 64.42 GB

Udev Sync Supported: true

Deferred Removal Enabled: true

Deferred Deletion Enabled: false

Deferred Deleted Device Count: 0

Library Version: 1.02.135-RHEL7 (2016-11-16)

Logging Driver: journald

Cgroup Driver: cgroupfs

Plugins:

Volume: local

Network: host bridge null overlay

Swarm: inactive

Runtimes: runc docker-runc

Default Runtime: runc

Security Options: seccomp

Kernel Version: 3.10.0-514.6.1.el7.x86_64

Operating System: CentOS Linux 7 (Core)

OSType: linux

Architecture: x86_64

Number of Docker Hooks: 2

CPUs: 40

Total Memory: 94.14 GiB

Name: VM-2-10-12

ID: NYBE:NZML:4KQQ:PF2J:RXCB:IPPI:Y3BI:CY7E:RVAC:WVWV:VDM2:3EEK

Docker Root Dir: /data1/docker

Debug Mode (client): false

Debug Mode (server): false

Registry: https://index.docker.io/v1/

Insecure Registries:

docker-registry.i.bbtfax.com:5000

127.0.0.0/8

Registries: docker.io (secure)

现象：

# docker exec -it c6176f37c4b6 bash
rpc error: code = 13 desc = invalid header field value "oci runtime error: exec failed: container_linux.go:247: starting container process caused \"process_linux.go:75: starting setns process caused \\\"fork/exec /proc/self/exe: no such file or directory\\\"\"\n"

1 2	# docker exec -it c6176f37c4b6 bash rpc error: code = 13 desc = invalid header field value "oci runtime error: exec failed: container_linux.go:247: starting container process caused \"process_linux.go:75: starting setns process caused \\\"fork/exec /proc/self/exe: no such file or directory\\\"\"\n"

乍一看， /proc/self/exe 文件找不见？

一般来讲，文件找不见并不奇怪，怪就怪在是 /proc/self/exe 找不见就不太应该了；

因为docker exec 最终是由libcontainerd进程来出来的，strace跟进发现，是chdir到 /root/data1/docker/devicemapper/mnt/4723e8178992b32b7284aa48c1c62f4011a6b785aca0c54e18d7ce5cc23b22dc/rootfs 时，找不到目标目录导致的，于是我就迅速地看了一下，该目录确实不存在，但是对于正常的能够exec的容器来讲，相应的rootfs目录也是不存在的

思考中。。。

docker玩的就是名字空间和cgroup，所以不能不想到这些；libcontainerd也有自己的（mnt）名字空间，我们进入libcontainerd进程的文件系统就可以查看到上面目录的存在了，而且，正常的容器存在相应的目录，异常的容器不存在相应的目录；

通过mount命令可以发现mount的规律，从容器的config.json (/var/run/docker/libcontainerd/c6176f37c4b67b03d4187edef6d1131cd44ab80bd0f0c20b24a7a20056967652/config.json) 中查看到对应的mount的位置，通过nsenter进入libcontainerd的mnt名字空间手动mount上去就好了，如下：

# nsenter  -m -t 3639 bash
# mount /dev/mapper/docker-253\:3-3221225568-4723e8178992b32b7284aa48c1c62f4011a6b785aca0c54e18d7ce5cc23b22dc -o rw,relatime,nouuid,attr2,inode64,sunit=512,swidth=1024,noquota -t xfs /data1/docker/devicemapper/mnt/4723e8178992b32b7284aa48c1c62f4011a6b785aca0c54e18d7ce5cc23b22dc

# nsenter -m -t 3639 bash

# mount /dev/mapper/docker-253\:3-3221225568-4723e8178992b32b7284aa48c1c62f4011a6b785aca0c54e18d7ce5cc23b22dc -o rw,relatime,nouuid,attr2,inode64,sunit=512,swidth=1024,noquota -t xfs /data1/docker/devicemapper/mnt/4723e8178992b32b7284aa48c1c62f4011a6b785aca0c54e18d7ce5cc23b22dc

写个脚本自动修复之：

#!/bin/bash
# author: phpor
#
LIBCONTAINERD_DIR=/var/run/docker/libcontainerd

function main() {
        local pidOfCotainerd=$(pidof docker-containerd-current)
        local mountinfo=$(< /proc/$pidOfCotainerd/mountinfo)
        for config in $LIBCONTAINERD_DIR/*/config.json;do
                local cid=$(awk -F'/' '{print $6}' <<<$config)
                local rootpath=$(jq -r .root.path $config|sed 's/\/rootfs$//')
                grep "$rootpath" <<<$mountinfo >/dev/null
                if [[ $? -eq 0 ]]; then
                        echo $cid $rootpath OK
                else
                        echo $cid $rootpath Should repair
                        local device=/dev/mapper/$(docker inspect $cid|jq -r .[0].GraphDriver.Data.DeviceName)
                        nsenter -m -p -t $pidOfCotainerd mount -t xfs -o rw,nouuid,attr2,inode64,sunit=512,swidth=1024,noquota $device $rootpath
                fi
        done
}
main

#!/bin/bash

# author: phpor

LIBCONTAINERD_DIR=/var/run/docker/libcontainerd

function main() {

local pidOfCotainerd=$(pidof docker-containerd-current)

local mountinfo=$(< /proc/$pidOfCotainerd/mountinfo)

for config in $LIBCONTAINERD_DIR/*/config.json;do

local cid=$(awk -F'/' '{print $6}' <<<$config)

local rootpath=$(jq -r .root.path $config|sed 's/\/rootfs$//')

grep "$rootpath" <<<$mountinfo >/dev/null

if [[ $? -eq 0 ]]; then

echo $cid $rootpath OK

else

echo $cid $rootpath Should repair

local device=/dev/mapper/$(docker inspect $cid|jq -r .[0].GraphDriver.Data.DeviceName)

nsenter -m -p -t $pidOfCotainerd mount -t xfs -o rw,nouuid,attr2,inode64,sunit=512,swidth=1024,noquota $device $rootpath

done

}

main

那么该mount点是如何丢掉的呢？重启dockerd能否自动修复该问题呢？（应该重启一下容器就行）稍后再研究

centos7 之vnc黑色桌面

windows virtio

kvm 之mem balloon

docker 之 exec失败的问题

windows 内存计算

libudev on centos7

vnc in openstack

RUNNING CONTAINERS WITH RUNC

cloud-init 工作原理 – 每天5分钟玩转 OpenStack（171） – CloudMan6的博客 – CSDN博客

Management Tools – KVM