目录
主机名 | 系统 | ip地址 | 角色 |
---|---|---|---|
ops-k8s-175 | ubuntu16.04 | 192.168.75.175 | k8s-master,glusterfs,heketi |
ops-k8s-176 | ubuntu16.04 | 192.168.75.176 | k8s-node,glusterfs |
ops-k8s-177 | ubuntu16.04 | 192.168.75.177 | k8s-node,glusterfs |
ops-k8s-178 | ubuntu16.04 | 192.168.175.178 | k8s-node,glusterfs |
# 在所有节点执行:
apt-get install glusterfs-server glusterfs-common glusterfs-client fuse
systemctl start glusterfs-server
systemctl enable glusterfs-server
# 在175上执行:
gluster peer probe 192.168.75.176
gluster peer probe 192.168.75.177
gluster peer probe 192.168.75.178
创建测试卷
# 创建
gluster volume create test-volume replica 2 192.168.75.175:/home/gluterfs/data 192.168.75.176:/home/glusterfs/data force
# 激活卷
gluster volume start test-volume
# 挂载
mount -t glusterfs 192.168.75.175:/test-volume /mnt/mytest
扩容测试卷
# 向卷中添加brick
gluster volume add-brick test-volume 192.168.75.177:/home/gluterfs/data 192.168.75.178:/home/glusterfs/data force
删除测试卷
gluster volume stop test-volume
gluster volume delete test-volume
Heketi提供了一个RESTful管理界面,可以用来管理GlusterFS卷的生命周期。 通过Heketi,就可以像使用OpenStack Manila,Kubernetes和OpenShift一样申请可以动态配置GlusterFS卷。Heketi会动态在集群内选择bricks构建所需的volumes,这样以确保数据的副本会分散到集群不同的故障域内。同时Heketi还支持任意数量的ClusterFS集群,以保证接入的云服务器不局限于单个GlusterFS集群。
heketi项目地址:https://github.com/heketi/heketi
下载heketi相关包:
https://github.com/heketi/heketi/releases/download/v5.0.1/heketi-client-v5.0.1.linux.amd64.tar.gz
https://github.com/heketi/heketi/releases/download/v5.0.1/heketi-v5.0.1.linux.amd64.tar.gz
修改heketi配置文件/etc/heketi/heketi.json,内容如下:
......
#修改端口,防止端口冲突
"port": "18080",
......
#允许认证
"use_auth": true,
......
#admin用户的key改为adminkey
"key": "adminkey"
......
#修改执行插件为ssh,并配置ssh的所需证书,注意要能对集群中的机器免密ssh登陆,使用ssh-copy-id把pub key拷到每台glusterfs服务器上
"executor": "ssh",
"sshexec": {
"keyfile": "/root/.ssh/id_rsa",
"user": "root",
"port": "22",
"fstab": "/etc/fstab"
},
......
# 定义heketi数据库文件位置
"db": "/var/lib/heketi/heketi.db"
......
#调整日志输出级别
"loglevel" : "warning"
需要说明的是,heketi有三种executor,分别为mock、ssh、kubernetes,建议在测试环境使用mock,生产环境使用ssh,当glusterfs以容器的方式部署在kubernetes上时,才使用kubernetes。我们这里将glusterfs和heketi独立部署,使用ssh的方式。
在上面我们配置heketi的时候使用了ssh的executor,那么就需要heketi服务器能通过ssh密钥的方式连接到所有glusterfs节点进行管理操作,所以需要先生成ssh密钥
ssh-keygen -t rsa -q -f /etc/heketi/heketi_key -N ''
chmod 600 /etc/heketi/heketi_key.pub
# ssh公钥传递,这里只以一个节点为例
ssh-copy-id -i /etc/heketi/heketi_key.pub root@192.168.75.175
# 验证是否能通过ssh密钥正常连接到glusterfs节点
ssh -i /etc/heketi/heketi_key root@192.168.75.175
nohup heketi -config=/etc/heketi/heketi.json &
在我实际生产中,使用docker-compose来管理heketi,而不直接手动启动,下面直接给出docker-compose配置示例:
version: "2"
services:
heketi:
container_name: heketi
image: dk-reg.op.douyuyuba.com/library/heketi:5
volumes:
- "/etc/heketi:/etc/heketi"
- "/var/lib/heketi:/var/lib/heketi"
- "/etc/localtime:/etc/localtime"
network_mode: host
heketi-cli --user admin -server http://192.168.75.175:18080 --secret adminkey --json cluster create
{"id":"d102a74079dd79aceb3c70d6a7e8b7c4","nodes":[],"volumes":[]}
由于我们开启了heketi认证,所以每次执行heketi-cli操作时,都需要带上一堆的认证字段,比较麻烦,我在这里创建一个别名来避免相关操作:
alias heketi-cli='heketi-cli --server "http://192.168.75.175:18080" --user "admin" --secret "adminkey"'
下面添加节点
heketi-cli --json node add --cluster "d102a74079dd79aceb3c70d6a7e8b7c4" --management-host-name 192.168.75.175 --storage-host-name 192.168.75.175 --zone 1
heketi-cli --json node add --cluster "d102a74079dd79aceb3c70d6a7e8b7c4" --management-host-name 192.168.75.176 --storage-host-name 192.168.75.176 --zone 1
heketi-cli --json node add --cluster "d102a74079dd79aceb3c70d6a7e8b7c4" --management-host-name 192.168.75.177 --storage-host-name 192.168.75.177 --zone 1
heketi-cli --json node add --cluster "d102a74079dd79aceb3c70d6a7e8b7c4" --management-host-name 192.168.75.178 --storage-host-name 192.168.75.178 --zone 1
看到有些文档说需要在centos上部署时,需要注释每台glusterfs上的/etc/sudoers中的Defaults requiretty,不然加第二个node死活报错,最后把日志级别调高才看到日志里有记录sudo提示require tty。由于我这里直接部署在ubuntu上,所有不存在上述问题。如果有遇到这种问题的,可以照着操作下。
这里需要特别说明的是,目前heketi仅支持使用裸分区或裸磁盘添加为device,不支持文件系统。
# --node参数给出的id是上一步创建node时生成的,这里只给出一个添加的示例,实际配置中,要添加每一个节点的每一块用于存储的硬盘
heketi-cli -json device add -name="/dev/vda2" --node "c3638f57b5c5302c6f7cd5136c8fdc5e"
上面展示了如何手动一步步生成cluster,往cluster中添加节点,添加device的操作,在我们实际生产配置中,可以直接通过配置文件完成。
创建一个/etc/heketi/topology-sample.json的文件,内容如下:
{
"clusters": [
{
"nodes": [
{
"node": {
"hostnames": {
"manage": [
"192.168.75.175"
],
"storage": [
"192.168.75.175"
]
},
"zone": 1
},
"devices": [
"/dev/vda2"
]
},
{
"node": {
"hostnames": {
"manage": [
"192.168.75.176"
],
"storage": [
"192.168.75.176"
]
},
"zone": 1
},
"devices": [
"/dev/vda2"
]
},
{
"node": {
"hostnames": {
"manage": [
"192.168.75.177"
],
"storage": [
"192.168.75.177"
]
},
"zone": 1
},
"devices": [
"/dev/vda2"
]
},
{
"node": {
"hostnames": {
"manage": [
"192.168.75.178"
],
"storage": [
"192.168.75.178"
]
},
"zone": 1
},
"devices": [
"/dev/vda2"
]
}
]
}
]
}
创建:
heketi-cli topology load --json topology-sample.json
这里仅仅是做一个测试,实际使用中,会由kubernetes自动创建pvc
如果添加的volume小的话可能会提示No Space,要解决这一问题要在heketi.json添加"brick_min_size_gb" : 1 ,1为1G
......
"brick_min_size_gb" : 1,
"db": "/var/lib/heketi/heketi.db"
......
size要比brick_min_size_gb大,如果设成1还是报min brick limit,replica必须大于1
heketi-cli --json volume create --size 3 --replica 2
在执行创建的时候,抛出了如下异常:
Error: /usr/sbin/thin_check: execvp failed: No such file or directory
WARNING: Integrity check of metadata for pool vg_d9fb2bec56cfdf73e21d612b1b3c1feb/tp_e94d763a9b687bfc8769ac43b57fa41e failed.
/usr/sbin/thin_check: execvp failed: No such file or directory
Check of pool vg_d9fb2bec56cfdf73e21d612b1b3c1feb/tp_e94d763a9b687bfc8769ac43b57fa41e failed (status:2). Manual repair required!
Failed to activate thin pool vg_d9fb2bec56cfdf73e21d612b1b3c1feb/tp_e94d763a9b687bfc8769ac43b57fa41e.
这需要在所有glusterfs节点机上安装thin-provisioning-tools包:
apt-get -y install thin-provisioning-tools
成功创建的返回输出如下:
heketi-cli --json volume create --size 3 --replica 2
{"size":3,"name":"vol_7fc61913851227ca2c1237b4c4d51997","durability":{"type":"replicate","replicate":{"replica":2},"disperse":{"data":4,"redundancy":2}},"snapshot":{"enable":false,"factor":1},"id":"7fc61913851227ca2c1237b4c4d51997","cluster":"dae1ab512dfad0001c3911850cecbd61","mount":{"glusterfs":{"hosts":["192.168.75.175","192.168.75.178"],"device":"192.168.75.175:vol_7fc61913851227ca2c1237b4c4d51997","options":{"backup-volfile-servers":"192.168.75.178"}}},"bricks":[{"id":"004f34fd4eb9e04ca3e1ca7cc1a2dd2c","path":"/var/lib/heketi/mounts/vg_d9fb2bec56cfdf73e21d612b1b3c1feb/brick_004f34fd4eb9e04ca3e1ca7cc1a2dd2c/brick","device":"d9fb2bec56cfdf73e21d612b1b3c1feb","node":"20d14c78691d9caef050b5dc78079947","volume":"7fc61913851227ca2c1237b4c4d51997","size":3145728},{"id":"2876e9a7574b0381dc0479aaa2b64d46","path":"/var/lib/heketi/mounts/vg_b7fd866d3ba90759d0226e26a790d71f/brick_2876e9a7574b0381dc0479aaa2b64d46/brick","device":"b7fd866d3ba90759d0226e26a790d71f","node":"9cddf0ac7899676c86cb135be16649f5","volume":"7fc61913851227ca2c1237b4c4d51997","size":3145728}]}
参考https://kubernetes.io/docs/concepts/storage/persistent-volumes/#persistentvolumeclaims
添加storageclass-glusterfs.yaml文件,内容如下:
apiVersion: storage.k8s.io/v1beta1
kind: StorageClass
metadata:
name: glusterfs
provisioner: kubernetes.io/glusterfs
parameters:
resturl: "http://192.168.75.175:18080"
restauthenabled: "true"
restuser: "admin"
restuserkey: "adminkey"
volumetype: "replicate:2"
kubectl apply -f storageclass-glusterfs.yaml
这是直接将userkey明文写入配置文件创建storageclass的方式,官方推荐将key使用secret保存。示例如下:
# glusterfs-secret.yaml内容如下:
apiVersion: v1
kind: Secret
metadata:
name: heketi-secret
namespace: default
data:
# base64 encoded password. E.g.: echo -n "mypassword" | base64
key: TFRTTkd6TlZJOEpjUndZNg==
type: kubernetes.io/glusterfs
# storageclass-glusterfs.yaml内容修改如下:
apiVersion: storage.k8s.io/v1beta1
kind: StorageClass
metadata:
name: glusterfs
provisioner: kubernetes.io/glusterfs
parameters:
resturl: "http://192.168.75.175:18080"
clusterid: "dae1ab512dfad0001c3911850cecbd61"
restauthenabled: "true"
restuser: "admin"
secretNamespace: "default"
secretName: "heketi-secret"
#restuserkey: "adminkey"
gidMin: "40000"
gidMax: "50000"
volumetype: "replicate:2"
更详细的用法参考:Storage Classes | Kubernetes
glusterfs-pvc.yaml内容如下:
kind: PersistentVolumeClaim
apiVersion: v1
metadata:
name: glusterfs-mysql1
namespace: default
annotations:
volume.beta.kubernetes.io/storage-class: "glusterfs"
spec:
accessModes:
- ReadWriteMany
resources:
requests:
storage: 2Gi
kubectl create -f glusterfs-pvc.yaml
mysql-deployment.yaml内容如下:
kind: Deployment
apiVersion: extensions/v1beta1
metadata:
name: mysql
namespace: default
spec:
replicas: 1
template:
metadata:
labels:
name: mysql
spec:
containers:
- name: mysql
image: mysql:5.7
imagePullPolicy: IfNotPresent
env:
- name: MYSQL_ROOT_PASSWORD
value: root123456
ports:
- containerPort: 3306
volumeMounts:
- name: gluster-mysql-data
mountPath: "/var/lib/mysql"
volumes:
- name: glusterfs-mysql-data
persistentVolumeClaim:
claimName: glusterfs-mysql1
kubectl create -f /etc/kubernetes/mysql-deployment.yaml
需要说明的是,我这里使用的动态pvc的方式来创建glusterfs挂载盘,还有一种手动创建pvc的方式,可以参考:一篇讲透Kubernetes与GlusterFS之间的爱恨情仇(一) - 服务公告 - 恒生研究院