背景介绍:
项目目前在文件同步方面采用的是rsync,在尝试用分布式文件系统替换的时候,使用过MooseFS,效果差强人意,在了解到了GlusterFS之
后,决定尝试一下,因为它跟MooseFS相比,感觉部署上更加简单一些,同时没有元数据服务器的特点使其没有单点故障的存在,感觉非常不错。
环境介绍:
OS: CentOS 6.4 x86_64 Minimal
Servers: sc2-log1,sc2-log2,sc2-log3,sc2-log4
Client: sc2-ads15
具体步骤:
1. 在sc2-log{1-4}上安装GlusterFS软件包:
# wget -P /etc/yum.repos.d http://download.gluster.org/pub/gluster/glusterfs/LATEST/CentOS/glusterfs-epel.repo
# yum install -y glusterfs-3.4.2-1.el6 glusterfs-server-3.4.2-1.el6 glusterfs-fuse-3.4.2-1.el6
# /etc/init.d/glusterd start
# chkconfig glusterfsd on
2. 在sc2-log1上配置整个GlusterFS集群:
[root@sc2-log1 ~]# gluster peer probe sc2-log1
peer probe: success: on localhost not needed
[root@sc2-log1 ~]# gluster peer probe sc2-log2
peer probe: success
[root@sc2-log1 ~]# gluster peer probe sc2-log3
peer probe: success
[root@sc2-log1 ~]# gluster peer probe sc2-log4
peer probe: success
[root@sc2-log1 ~]# gluster peer status
Number of Peers: 3
Hostname: sc2-log2
Port: 24007
Uuid: 399973af-bae9-4326-9cbd-b5b05e5d2927
State: Peer in Cluster (Connected)
Hostname: sc2-log3
Port: 24007
Uuid: 833a7b8d-e3b3-4099-baf9-416ee7213337
State: Peer in Cluster (Connected)
Hostname: sc2-log4
Port: 24007
Uuid: 54bf115a-0119-4021-af80-7a6bca137fd9
State: Peer in Cluster (Connected)
3. 在sc2-log{1-4}上创建数据存放目录:
# mkdir -p /usr/local/share/{models,geoip,wurfl}
# ls -l /usr/local/share/
total 24
drwxr-xr-x 2 root root 4096 Apr 1 12:19 geoip
drwxr-xr-x 2 root root 4096 Apr 1 12:19 models
drwxr-xr-x 2 root root 4096 Apr 1 12:19 wurfl
4. 在sc2-log1上创建GlusterFS磁盘:
[root@sc2-log1 ~]# gluster volume create models replica 4
sc2-log1:/usr/local/share/models sc2-log2:/usr/local/share/models
sc2-log3:/usr/local/share/models sc2-log4:/usr/local/share/models force
volume create: models: success: please start the volume to access data
[root@sc2-log1 ~]# gluster volume create geoip replica 4 sc2-log1:/usr/local/share/geoip sc2-log2:/usr/local/share/geoip sc2-log3:/usr/local/share/geoip sc2-log4:/usr/local/share/geoip force
volume create: geoip: success: please start the volume to access data
[root@sc2-log1 ~]# gluster volume create wurfl replica 4 sc2-log1:/usr/local/share/wurfl sc2-log2:/usr/local/share/wurfl sc2-log3:/usr/local/share/wurfl sc2-log4:/usr/local/share/wurfl force
volume create: wurfl: success: please start the volume to access data
[root@sc2-log1 ~]# gluster volume start models
volume start: models: success
[root@sc2-log1 ~]# gluster volume start geoip
volume start: geoip: success
[root@sc2-log1 ~]# gluster volume start wurfl
volume start: wurfl: success
[root@sc2-log1 ~]# gluster volume info
Volume Name: models
Type: Replicate
Volume ID: b29b22bd-6d8c-45c0-b199-91fa5a76801f
Status: Started
Number of Bricks: 1 x 4 = 4
Transport-type: tcp
Bricks:
Brick1: sc2-log1:/usr/local/share/models
Brick2: sc2-log2:/usr/local/share/models
Brick3: sc2-log3:/usr/local/share/models
Brick4: sc2-log4:/usr/local/share/models
Volume Name: geoip
Type: Replicate
Volume ID: 69b0caa8-7c23-4712-beae-6f536b1cffa3
Status: Started
Number of Bricks: 1 x 4 = 4
Transport-type: tcp
Bricks:
Brick1: sc2-log1:/usr/local/share/geoip
Brick2: sc2-log2:/usr/local/share/geoip
Brick3: sc2-log3:/usr/local/share/geoip
Brick4: sc2-log4:/usr/local/share/geoip
Volume Name: wurfl
Type: Replicate
Volume ID: c723a99d-eeab-4865-819a-c0926cf7b88a
Status: Started
Number of Bricks: 1 x 4 = 4
Transport-type: tcp
Bricks:
Brick1: sc2-log1:/usr/local/share/wurfl
Brick2: sc2-log2:/usr/local/share/wurfl
Brick3: sc2-log3:/usr/local/share/wurfl
Brick4: sc2-log4:/usr/local/share/wurfl
5. 在sc2-ads15上部署客户端并mount GlusterFS文件系统:
[sc2-ads15][root@sc2-ads15 ~]# wget -P /etc/yum.repos.d
http://download.gluster.org/pub/gluster/glusterfs/LATEST/CentOS/glusterfs-epel.repo
[sc2-ads15][root@sc2-ads15 ~]# yum install -y glusterfs-3.4.2-1.el6 glusterfs-fuse-3.4.2-1.el6
[sc2-ads15][root@sc2-ads15 ~]# mkdir -p /mnt/{models,geoip,wurfl}
[sc2-ads15][root@sc2-ads15 ~]# mount -t glusterfs -o ro sc2-log3:models /mnt/models/
[sc2-ads15][root@sc2-ads15 ~]# mount -t glusterfs -o ro sc2-log3:geoip /mnt/geoip/
[sc2-ads15][root@sc2-ads15 ~]# mount -t glusterfs -o ro sc2-log3:wurfl /mnt/wurfl/
[sc2-ads15][root@sc2-ads15 ~]# df -h
Filesystem Size Used Avail Use% Mounted on
/dev/mapper/vg_t-lv_root
59G 7.7G 48G 14% /
tmpfs 3.9G 0 3.9G 0% /dev/shm
/dev/xvda1 485M 33M 428M 8% /boot
sc2-log3:models 98G 8.6G 85G 10% /mnt/models
sc2-log3:geoip 98G 8.6G 85G 10% /mnt/geoip
sc2-log3:wurfl 98G 8.6G 85G 10% /mnt/wurfl
6. 相关数据读写可用性测试:
在sc2-ads15挂载点上写入数据:
[sc2-ads15][root@sc2-ads15 ~]# umount /mnt/models
[sc2-ads15][root@sc2-ads15 ~]# mount -t glusterfs sc2-log3:models /mnt/models/
[sc2-ads15][root@sc2-ads15 ~]# echo "This is sc2-ads15" > /mnt/models/hello.txt
[sc2-ads15][root@sc2-ads15 ~]# mkdir /mnt/testdir
在sc2-log1数据目录中进行查看:
[root@sc2-log1 ~]# ls /usr/local/share/models/
hello.txt testdir
结果: 数据写入成功
在sc2-log1数据目录中直接写入数据:
[root@sc2-log1 ~]# echo "This is sc2-log1" > /usr/local/share/models/hello.2.txt
[root@sc2-log1 ~]# mkdir /usr/local/share/models/test2
在sc2-ads15挂载点上进行查看:
[sc2-ads15][root@sc2-ads15 ~]# ls /mnt/models
[sc2-ads15][root@sc2-ads15 ~]# ls -l /mnt/models
hello.txt testdir
结果: 数据写入失败
在sc2-log1挂载点上写入数据:
[root@sc2-log1 ~]# mount -t glusterfs sc2-log1:models /mnt/models/
[root@sc2-log1 ~]# echo "This is sc2-log1" > /mnt/models/hello.3.txt
[root@sc2-log1 ~]# mkdir /mnt/models/test3
在sc2-ads15挂载点上进行查看:
[sc2-ads15][root@sc2-ads15 models]# ls /mnt/models
hello.2.txt hello.3.txt hello.txt test2 test3 testdir
结果: 数据写入成功,同时之前写入失败的数据也成功加载了。
最终结论:
在数据目录中直接写入数据,会导致其它节点因为得不到通知而使数据同步失败。
正确的做法是所有的读写操作都通过挂载点来进行。
7. 其它操作笔记:
删除GlusterFS磁盘:
# gluster volume stop models
# gluster volume delete models
卸载GlusterFS磁盘:
# gluster peer detach sc2-log4
ACL访问控制:
# gluster volume set models auth.allow 10.60.1.*
添加GlusterFS节点:
# gluster peer probe sc2-log5
# gluster peer probe sc2-log6
# gluster volume add-brick models sc2-log5:/data/gluster sc2-log6:/data/gluster
迁移GlusterFS磁盘数据:
# gluster volume remove-brick models sc2-log1:/usr/local/share/models sc2-log5:/usr/local/share/models start
# gluster volume remove-brick models sc2-log1:/usr/local/share/models sc2-log5:/usr/local/share/models status
# gluster volume remove-brick models sc2-log1:/usr/local/share/models sc2-log5:/usr/local/share/models commit
修复GlusterFS磁盘数据(例如在sc2-log1宕机的情况下):
# gluster volume replace-brick models sc2-log1:/usr/local/share/models sc2-log5:/usr/local/share/models commit -force
# gluster volume heal models full