重庆分公司,新征程启航
为企业提供网站建设、域名注册、服务器等服务
Ceph块设备,以前称为 RADOS 块设备,为客户机提供可靠的、分布式的和高性能的块存储磁盘。
创新互联是一家专注于成都网站建设、网站制作与策划设计,鄯善网站建设哪家好?创新互联做网站,专注于网站建设十多年,网设计领域的专业建站公司;建站业务涵盖:鄯善等地区。鄯善做网站价格咨询:13518219792RADOS 块设备利用 librbd 库并以顺序的形式在 Ceph 集群中的多个 osd 上存储数据块。RBD是由 Ceph 的 RADOS 层支持的,因此每个块设备都分布在多个 Ceph 节点上,提供了高性能和优异的可靠性。RBD 有 linux 内核的本地支持。
任何普通的 linux 主机都可以充当 Ceph 的客户机。客户端通过网络与 Ceph 存储集群交互以存储或检索用户数据。Ceph RBD 支持已经添加到 Linux主线内核中,从 2.6.34和以后的版本开始。
192.168.3.158为客户端做如下操作
# 创建 ceph 块客户端用户名和认证密钥
[ceph@ceph265 my-cluster]$ ceph auth get-or-create client.rbd mon 'allow r' osd 'allow class-read object_prefix rbd_children, allow rwx pool=rbd' | tee ./ceph.client.rbd.keyring [client.rbd] key = AQBLBwRepKVJABAALyRx67z6efeI4xogPqHkyw== 注: client.rbd 为客户端名 mon 之后的全为授权配置拷贝配置文件及密钥到客户机
[ceph@ceph265 my-cluster]$ scp ceph.client.rbd.keyring root@192.168.3.158:/etc/ceph [ceph@ceph265 my-cluster]$ scp ceph.conf root@192.168.3.158:/etc/ceph# 检查是否符合块设备环境要求
uname -r modprobe rbd#安装 ceph 客户端
# wget -O /etc/yum.repos.d/ceph.repo https://raw.githubusercontent.com/aishangwei/ceph-demo/master/ceph-deploy/ceph.repo查看密钥文件
[root@ceph258 ~]# cat /etc/ceph/ceph.client.rbd.keyring [client.rdb] key = AQBLBwRepKVJABAALyRx67z6efeI4xogPqHkyw== [root@ceph258 ~]# ceph -s --name client.rbd在服务器192.168.3.165执行如下命令
(1) 创建块设备
默认创建块设备,会直接创建在 rbd 池中,但使用 deploy 安装后,该 rbd 池并没有创建。
# 创建池和块
$ ceph osd lspools # 查看集群存储池 $ ceph osd pool create rbd 50 pool 'rbd' created # 50 为 place group 数量,由于我们后续测试,也需要更多的 pg,所以这里设置为50确定pg_num取值是强制性的,因为不能自动计算。下面是几个常用的值:
少于5个OSD时可把pg_num设置为128
OSD数量在5到10个时,可把pg_num设置为512
OSD数量在10到50个时,可把pg_num设置为4096
OSD数量大于50时,你得理解权衡方法、以及如何自己计算pg_num取值
(2)客户端创建 块设备
创建一个容量为 5105M 的 rbd 块设备
[root@ceph258 ~]# rbd create rbd2 --size 5105 --name client.rbd192.168.3.158 客户端查看 rbd2 块设备
[root@ceph258 ~]# rbd ls --name client.rbd rbd2 [root@ceph258 ~]# rbd ls -p rbd --name client.rbd rbd2 [root@ceph258 ~]# rbd list --name client.rbd rbd2查看 rbd2块设备信息
[root@ceph258 ~]# rbd --image rbd2 info --name client.rbd# 映射到客户端,应该会报错
[root@ceph258 ~]# rbd map --image rbd2 --name client.rbdlayering:分层支持
- exclusive-lock:排它锁定支持对
- object-map:对象映射支持(需要排它锁定(exclusive-lock))
- deep-flatten:快照平支持(snapshot flatten support)
- fast-diff:在client-node1上使用krbd(内核rbd)客户机进行快速diff计算(需要对象映射),我们将无法在CentOS内核3.10上映射块设备映像,因为该内核不支持对象映射(object-map)、深平(deep-flatten)和快速dif(fast-dif)(在内核4.9中引入了支持)。为了解决这个问题,我们将禁用不支持的特性,有几个选项可以做到这一点:
1)动态禁用
rbd feature disable rbdl exclusive-lock object-map deep-flatten fast-diff--name client.rbd2)创建RBD镜像时,只启用分层特性。
rbd create rbd2 --size 10240 --image-feature layering--name client.rbd3)ceph配置文件中禁用
rbd_default_features=1# 我们这里动态禁用
[root@ceph258 ~]# rbd feature disable rbd2 exclusive-lock object-map fast-diff deep-flatten --name client.rbd对 rbd2 进行映射
[root@ceph258 ~]# rbd map --image rbd2 --name client.rbd查看本机已经映射的 rbd 镜像
[root@ceph258 ~]# rbd showmapped --name client.rbd查看磁盘 rbd0 大小
格式化 rbd0
创建挂载目录并进行挂载
[root@ceph258 ~]# mkdir /mnt/ceph-disk1 [root@ceph258 ~]# mount /dev/rbd0 /mnt/ceph-disk1/# 写入数据测试
[root@ceph258 ~]# dd if=/dev/zero of=/mnt/ceph-disk1/file1 count=100 bs=1M# 做成服务,开机自动挂载
[root@ceph203-]# wget -O /usr/local/bin/rbd-mount https://raw.githubusercontent.com/aishangwei/ceph-demo/master/client/rbd-mount# vim /usr/local/bin/rbd-mount
[root@ceph258 ~]# chmod +x /usr/local/bin/rbd-mount [root@ceph258~ ]# wget -O /etc/systemd/system/rbd-mount.service https://raw.githubusercontent.com/aishangwei/ceph-demo/master/client/rbd-mount.service [root@ceph258 ~]# systemctl daemon-reload [root@ceph258 ~]# systemctl enable rbd-mount.service Created symlink from /etc/systemd/system/multi-user.target.wants/rbd-mount.service to /etc/systemd/system/rbd-mount.service.卸载手动挂载的目录,进行服务自动挂载测试
[root@ceph258 ~]# umount /mnt/ceph-disk1/ [root@ceph258 ~]# systemctl status rbd-mountCeph: RBD在线扩容容量
Ceph管理端的操作
查询 pool 总容量及已经分配容量
[root@ceph265 ~]# ceph df查看已经存在的 pool
[root@ceph265 ~]# ceph osd lspools查看已经有的 rbd
开始对 rbd2 进行动态扩容
[root@ceph265 ~]# rbd resize rbd/rbd2 --size 7168Ceph客户端的操作
[root@ceph258 ~]# rbd showmapped[root@ceph258 ~]# df -h[root@ceph258 ~]# xfs_growfs -d /mnt/ceph-disk1输入ceph-deploy mon create-initial命令获取密钥key,会在当前目录(如我的是~/etc/ceph/)下生成几个key,但报错如下。意思是:就是配置失败的两个结点的配置文件的内容于当前节点不一致,提示使用--overwrite-conf参数去覆盖不一致的配置文件。
# ceph-deploy mon create-initial ... [ceph3][DEBUG ] remote hostname: ceph3 [ceph3][DEBUG ] write cluster configuration to /etc/ceph/{cluster}.conf [ceph_deploy.mon][ERROR ] RuntimeError: config file /etc/ceph/ceph.conf exists with different content; use --overwrite-conf to overwrite [ceph_deploy][ERROR ] GenericError: Failed to create 2 monitors ...输入命令如下(此处我共配置了三个结点ceph2~3):
# ceph-deploy --overwrite-conf mon create ceph{3,1,2} ... [ceph3][DEBUG ] remote hostname: ceph3 [ceph3][DEBUG ] write cluster configuration to /etc/ceph/{cluster}.conf [ceph3][DEBUG ] create the mon path if it does not exist [ceph3][DEBUG ] checking for done path: /var/lib/ceph/mon/ceph-ceph3/done ...之后配置成功,可继续进行初始化磁盘操作。
从上面集群状态信息可查,每个osd上的pg数量=21<最小的数目30个。pgs为32,因为我之前设置的是2副本的配置,所以当有3个osd的时候,每个osd上均分了32÷3*2=21个pgs,也就是出现了如上的错误 小于最小配置30个。
集群这种状态如果进行数据的存储和操作,会发现集群卡死,无法响应io,同时会导致大面积的osd down。
解决办法:
增加pg数
因为我的一个pool有8个pgs,所以我需要增加两个pool才能满足osd上的pg数量=48÷3*2=32>最小的数目30。
[root@ceph2 ceph]# ceph osd pool create mytest 8 pool 'mytest' created [root@ceph2 ceph]# ceph osd pool create mytest1 8 pool 'mytest1' created [root@ceph2 ceph]# ceph -s cluster: id: 8e2248e4-3bb0-4b62-ba93-f597b1a3bd40 health: HEALTH_OK services: mon: 3 daemons, quorum ceph3,ceph2,ceph4 mgr: ceph3(active), standbys: ceph2, ceph4 osd: 3 osds: 3 up, 3 in rgw: 1 daemon active data: pools: 6 pools, 48 pgs objects: 219 objects, 1.1 KiB usage: 3.0 GiB used, 245 GiB / 248 GiB avail pgs: 48 active+clean集群健康状态显示正常。
如果此时,查看集群状态是HEALTH_WARN application not enabled on 1 pool(s):
[root@ceph2 ceph]# ceph -s cluster: id: 13430f9a-ce0d-4d17-a215-272890f47f28 health: HEALTH_WARN application not enabled on 1 pool(s) [root@ceph2 ceph]# ceph health detail HEALTH_WARN application not enabled on 1 pool(s) POOL_APP_NOT_ENABLED application not enabled on 1 pool(s) application not enabled on pool 'mytest' use 'ceph osd pool application enable运行ceph health detail命令发现是新加入的存储池mytest没有被应用程序标记,因为之前添加的是RGW实例,所以此处依提示将mytest被rgw标记即可:
[root@ceph2 ceph]# ceph osd pool application enable mytest rgw enabled application 'rgw' on pool 'mytest'再次查看集群状态发现恢复正常
[root@ceph2 ceph]# ceph health HEALTH_OK以下以删除mytest存储池为例,运行ceph osd pool rm mytest命令报错,显示需要在原命令的pool名字后再写一遍该pool名字并最后加上--yes-i-really-really-mean-it参数
[root@ceph2 ceph]# ceph osd pool rm mytest Error EPERM: WARNING: this will *PERMANENTLY DESTROY* all data stored in pool mytest. If you are *ABSOLUTELY CERTAIN* that is what you want, pass the pool name *twice*, followed by --yes-i-really-really-mean-it.按照提示要求复写pool名字后加上提示参数如下,继续报错:
[root@ceph2 ceph]# ceph osd pool rm mytest mytest --yes-i-really-really-mean-it Error EPERM: pool deletion is disabled; you must first set the mon_allow_pool_delete config option to true before you can destroy a pool错误信息显示,删除存储池操作被禁止,应该在删除前现在ceph.conf配置文件中增加mon_allow_pool_delete选项并设置为true。所以分别登录到每一个节点并修改每一个节点的配置文件。操作如下:
[root@ceph2 ceph]# vi ceph.conf [root@ceph2 ceph]# systemctl restart ceph-mon.target在ceph.conf配置文件底部加入如下参数并设置为true,保存退出后使用systemctl restart ceph-mon.target命令重启服务。
[mon]
mon allow pool delete = true
其余节点操作同理。
[root@ceph3 ceph]# vi ceph.conf [root@ceph3 ceph]# systemctl restart ceph-mon.target [root@ceph4 ceph]# vi ceph.conf [root@ceph4 ceph]# systemctl restart ceph-mon.target再次删除,即成功删除mytest存储池。
[root@ceph2 ceph]# ceph osd pool rm mytest mytest --yes-i-really-really-mean-it pool 'mytest' removed笔者将ceph集群中的三个节点分别关机并重启后,查看ceph集群状态如下:
[root@ceph2 ~]# ceph -s cluster: id: 13430f9a-ce0d-4d17-a215-272890f47f28 health: HEALTH_WARN 1 MDSs report slow metadata IOs 324/702 objects misplaced (46.154%) Reduced data availability: 126 pgs inactive Degraded data redundancy: 144/702 objects degraded (20.513%), 3 pgs degraded, 126 pgs undersized services: mon: 3 daemons, quorum ceph3,ceph2,ceph4 mgr: ceph2(active), standbys: ceph3, ceph4 mds: cephfs-1/1/1 up {0=ceph2=up:creating} osd: 3 osds: 3 up, 3 in; 162 remapped pgs data: pools: 8 pools, 288 pgs objects: 234 objects, 2.8 KiB usage: 3.0 GiB used, 245 GiB / 248 GiB avail pgs: 43.750% pgs not active 144/702 objects degraded (20.513%) 324/702 objects misplaced (46.154%) 162 active+clean+remapped 123 undersized+peered 3 undersized+degraded+peered查看
[root@ceph2 ~]# ceph health detail HEALTH_WARN 1 MDSs report slow metadata IOs; 324/702 objects misplaced (46.154%); Reduced data availability: 126 pgs inactive; Degraded data redundancy: 144/702 objects degraded (20.513%), 3 pgs degraded, 126 pgs undersized MDS_SLOW_METADATA_IO 1 MDSs report slow metadata IOs mdsceph2(mds.0): 9 slow metadata IOs are blocked > 30 secs, oldest blocked for 42075 secs OBJECT_MISPLACED 324/702 objects misplaced (46.154%) PG_AVAILABILITY Reduced data availability: 126 pgs inactive pg 8.28 is stuck inactive for 42240.369934, current state undersized+peered, last acting [0] pg 8.2a is stuck inactive for 45566.934835, current state undersized+peered, last acting [0] pg 8.2d is stuck inactive for 42240.371314, current state undersized+peered, last acting [0] pg 8.2f is stuck inactive for 45566.913284, current state undersized+peered, last acting [0] pg 8.32 is stuck inactive for 42240.354304, current state undersized+peered, last acting [0] .... pg 8.28 is stuck undersized for 42065.616897, current state undersized+peered, last acting [0] pg 8.2a is stuck undersized for 42065.613246, current state undersized+peered, last acting [0] pg 8.2d is stuck undersized for 42065.951760, current state undersized+peered, last acting [0] pg 8.2f is stuck undersized for 42065.610464, current state undersized+peered, last acting [0] pg 8.32 is stuck undersized for 42065.959081, current state undersized+peered, last acting [0] ....可见在数据修复中, 出现了inactive和undersized的值, 则是不正常的现象
解决方法:
①处理inactive的pg:
重启一下osd服务即可
[root@ceph2 ~]# systemctl restart ceph-osd.target继续查看集群状态发现,inactive值的pg已经恢复正常,此时还剩undersized的pg。
[root@ceph2 ~]# ceph -s cluster: id: 13430f9a-ce0d-4d17-a215-272890f47f28 health: HEALTH_WARN 1 filesystem is degraded 241/723 objects misplaced (33.333%) Degraded data redundancy: 59 pgs undersized services: mon: 3 daemons, quorum ceph3,ceph2,ceph4 mgr: ceph2(active), standbys: ceph3, ceph4 mds: cephfs-1/1/1 up {0=ceph2=up:rejoin} osd: 3 osds: 3 up, 3 in; 229 remapped pgs rgw: 1 daemon active data: pools: 8 pools, 288 pgs objects: 241 objects, 3.4 KiB usage: 3.0 GiB used, 245 GiB / 248 GiB avail pgs: 241/723 objects misplaced (33.333%) 224 active+clean+remapped 59 active+undersized 5 active+clean io: client: 1.2 KiB/s rd, 1 op/s rd, 0 op/s wr②处理undersized的pg:
学会出问题先查看健康状态细节,仔细分析发现虽然设定的备份数量是3,但是PG 12.x却只有两个拷贝,分别存放在OSD 0~2的某两个上。
[root@ceph2 ~]# ceph health detail HEALTH_WARN 241/723 objects misplaced (33.333%); Degraded data redundancy: 59 pgs undersized OBJECT_MISPLACED 241/723 objects misplaced (33.333%) PG_DEGRADED Degraded data redundancy: 59 pgs undersized pg 12.8 is stuck undersized for 1910.001993, current state active+undersized, last acting [2,0] pg 12.9 is stuck undersized for 1909.989334, current state active+undersized, last acting [2,0] pg 12.a is stuck undersized for 1909.995807, current state active+undersized, last acting [0,2] pg 12.b is stuck undersized for 1910.009596, current state active+undersized, last acting [1,0] pg 12.c is stuck undersized for 1910.010185, current state active+undersized, last acting [0,2] pg 12.d is stuck undersized for 1910.001526, current state active+undersized, last acting [1,0] pg 12.e is stuck undersized for 1909.984982, current state active+undersized, last acting [2,0] pg 12.f is stuck undersized for 1910.010640, current state active+undersized, last acting [2,0]进一步查看集群osd状态树,发现ceph3和cepn3宕机再恢复后,osd.1 和osd.2进程已不在ceph3和cepn3上。
[root@ceph2 ~]# ceph osd tree ID CLASS WEIGHT TYPE NAME STATUS REWEIGHT PRI-AFF -1 0.24239 root default -9 0.16159 host centos7evcloud 1 hdd 0.08080 osd.1 up 1.00000 1.00000 2 hdd 0.08080 osd.2 up 1.00000 1.00000 -3 0.08080 host ceph2 0 hdd 0.08080 osd.0 up 1.00000 1.00000 -5 0 host ceph3 -7 0 host ceph4分别查看osd.1 和osd.2服务状态。
解决方法:
分别进入到ceph3和ceph4节点中重启osd.1 和osd.2服务,将这两个服务重新映射到ceph3和ceph4节点中。
[root@ceph2 ~]# ssh ceph3 [root@ceph3 ~]# systemctl restart ceph-osd@1.service [root@ceph3 ~]# ssh ceph4 [root@ceph4 ~]# systemctl restart ceph-osd@2.service最后查看集群osd状态树发现这两个服务重新映射到ceph3和ceph4节点中。
[root@ceph4 ~]# ceph osd tree ID CLASS WEIGHT TYPE NAME STATUS REWEIGHT PRI-AFF -1 0.24239 root default -9 0 host centos7evcloud -3 0.08080 host ceph2 0 hdd 0.08080 osd.0 up 1.00000 1.00000 -5 0.08080 host ceph3 1 hdd 0.08080 osd.1 up 1.00000 1.00000 -7 0.08080 host ceph4 2 hdd 0.08080 osd.2 up 1.00000 1.00000集群状态也显示了久违的HEALTH_OK。
[root@ceph4 ~]# ceph -s cluster: id: 13430f9a-ce0d-4d17-a215-272890f47f28 health: HEALTH_OK services: mon: 3 daemons, quorum ceph3,ceph2,ceph4 mgr: ceph2(active), standbys: ceph3, ceph4 mds: cephfs-1/1/1 up {0=ceph2=up:active} osd: 3 osds: 3 up, 3 in rgw: 1 daemon active data: pools: 8 pools, 288 pgs objects: 241 objects, 3.6 KiB usage: 3.1 GiB used, 245 GiB / 248 GiB avail pgs: 288 active+clean挂载命令如下:
mount -t ceph 10.0.86.246:6789,10.0.86.221:6789,10.0.86.253:6789:/ /mnt/mycephfs/ -o name=admin,secret=AQBAI/JbROMoMRAAbgRshBRLLq953AVowLgJPw==卸载CephFS后再挂载时报错:mount error(2): No such file or directory
说明:首先检查/mnt/mycephfs/目录是否存在并可访问,我的是存在的但依然报错No such file or directory。但是我重启了一下osd服务意外好了,可以正常挂载CephFS。
[root@ceph2 ~]# systemctl restart ceph-osd.target [root@ceph2 ~]# mount -t ceph 10.0.86.246:6789,10.0.86.221:6789,10.0.86.253:6789:/ /mnt/mycephfs/ -o name=admin,secret=AQBAI/JbROMoMRAAbgRshBRLLq953AVowLgJPw==可见挂载成功~!
[root@ceph2 ~]# df -h Filesystem Size Used Avail Use% Mounted on /dev/vda2 48G 7.5G 41G 16% / devtmpfs 1.9G 0 1.9G 0% /dev tmpfs 2.0G 8.0K 2.0G 1% /dev/shm tmpfs 2.0G 17M 2.0G 1% /run tmpfs 2.0G 0 2.0G 0% /sys/fs/cgroup tmpfs 2.0G 24K 2.0G 1% /var/lib/ceph/osd/ceph-0 tmpfs 396M 0 396M 0% /run/user/0 10.0.86.246:6789,10.0.86.221:6789,10.0.86.253:6789:/ 249G 3.1G 246G 2% /mnt/mycephfs参考链接
https://blog.csdn.net/SL_World/article/details/84584366
另外有需要云服务器可以了解下创新互联cdcxhl.cn,海内外云服务器15元起步,三天无理由+7*72小时售后在线,公司持有idc许可证,提供“云服务器、裸金属服务器、高防服务器、香港服务器、美国服务器、虚拟主机、免备案服务器”等云主机租用服务以及企业上云的综合解决方案,具有“安全稳定、简单易用、服务可用性高、性价比高”等特点与优势,专为企业上云打造定制,能够满足用户丰富、多元化的应用场景需求。