[csi]浅聊ceph-csi组件

描述

??ceph-csi扩展各种存储类型的卷的管理能力，实现第三方存储ceph的各种操作能力与k8s存储系统的结合。调用第三方存储ceph的接口或命令，从而提供ceph数据卷的创建/删除、挂载/解除挂载的具体操作实现。前面分析组件中的对于数据卷的创建/删除、挂载/解除挂载操作，全是调用ceph-csi，然后由ceph-csi调用ceph提供的命令或接口来完成最终的操作。

ceph-csi服务组成

??ceph-csi含有rbdType、cephfsType、livenessType三大类型服务。rbdType主要进行rbd的操作完成与ceph的交互，cephfsType主要进行cephfs的操作完成与ceph交互，livenessType该服务主要是定时向csi endpoint探测csi组件的存活（向指定的socket地址发送probe请求），然后统计到prometheus指标中。

• ControllerServer：主要负责创建、删除cephfs/rbd存储等操作。
• NodeServer：部署在k8s中的每个node上，主要负责cephfs、rbd在node节点上相关的操作，如将存储挂载到node上，解除node上存储挂载等操作。
• IdentityServer：主要是返回自身服务的相关信息，如返回服务身份信息（名称与版本等信息）、返回服务具备的能力、暴露存活探测接口（用于给别的组件/服务探测该服务是否存活）等。
  

ceph-csi及相关组件部署

• 部署步骤
  部署请参考我的这个文章《k8s基于csi使用rbd存储》。

• 部署的组件介绍[rbd csi为例]
  csi-rbdplugin-provisioner.yaml部署的相关组件：csi-provisioner、csi-snapshotter、csi-attacher、csi-resizer、csi-rbdplugin、liveness-prometheus5个容器，作用分别如下。

  • csi-provisioner：实际上是external-provisioner组件。在create pvc时，csi-provisioner参与存储资源与pv对象的创建。csi-provisioner组件监听到pvc创建事件后，负责拼接请求，调用ceph-csi组件（即csi-rbdplugin容器）的CreateVolume方法来创建存储，创建存储成功后，创建pv对象；delete pvc时，csi-provisioner参与存储资源与pv对象的删除。当pvc被删除时，pv controller会将其绑定的pv对象状态由bound更新为release，csi-provisioner监听到pv更新事件后，调用ceph-csi组件（即csi-rbdplugin容器）的DeleteVolume方法来删除存储，并删除pv对象。
    ?
  • csi-snapshotter：实际上是external-snapshotter组件，负责处理存储快照相关的操作。
    ?
  • csi-attacher：实际上是external-attacher组件，只负责操作VolumeAttachment对象，实际上并没有操作存储。
    ?
  • csi-resizer：实际上是external-resizer组件，负责处理存储扩容相关的操作。
    ?
  • csi-rbdplugin：实际上是ceph-csi组件，rbdType-ControllerServer/IdentityServer类型的服务。create pvc时，external-provisioner组件（即csi-provisioner容器）监听到pvc创建事件后，负责拼接请求，然后调用csi-rbdplugin容器的CreateVolume方法来创建存储；delete pvc时，pv对象状态由bound变为release，external-provisioner组件（即csi-provisioner容器）监听到pv更新事件后，负责拼接请求，调用csi-rbdplugin容器的DeleteVolume方法来删除存储。
    ?
  • liveness-prometheus：实际上是ceph-csi组件，livenessType类型的服务。负责探测并上报csi-rbdplugin服务的存活情况。

  csi-rbdplugin.yaml部署的相关组件：包含了driver-registrar、csi-rbdplugin、liveness-prometheus 3个容器，作用分别如下。

  • driver-registrar：向kubelet传入csi-rbdplugin容器提供服务的socket地址、版本信息和驱动名称（如rbd.csi.ceph.com）等，将csi-rbdplugin容器服务注册给kubelet。
    ?
  • csi-rbdplugin：实际上是ceph-csi组件，rbdType-NoderServer/IdentityServer类型的服务。create pod cliam pvc时，kubelet会调用csi-rbdplugin容器将创建好的存储从ceph集群挂载到pod所在的node上，然后再挂载到pod相应的目录上；delete pod cliam pvc时，kubelet会调用csi-rbdplugin容器的相应方法，解除存储在pod目录上的挂载，再解除存储在node上的挂载。
    ?
  • liveness-prometheus：实际上是ceph-csi组件，livenessType类型的服务。负责探测并上报csi-rbdplugin服务的存活情况。

csi如何进行pvc的create和delete

# 环境信息

$ kubectl get pod NAME                                         READY   STATUS    RESTARTS   AGEcsi-rbdplugin-9tfnm                          3/3     Running   0          26hcsi-rbdplugin-provisioner-5cc9f558c7-d2stz   7/7     Running   0          26h$ kubectl get pvc,pvNAME                                  STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS   AGEpersistentvolumeclaim/raw-block-pvc   Bound    pvc-4e52c163-a593-4cc1-af59-23367d1e7573   2Gi        RWO            csi-rbd-sc     9m47sNAME                                                        CAPACITY   ACCESS MODES   RECLAIM POLICY   STATUS   CLAIM                   STORAGECLASS   REASON   AGEpersistentvolume/pvc-4e52c163-a593-4cc1-af59-23367d1e7573   2Gi        RWO            Delete           Bound    default/raw-block-pvc   csi-rbd-sc              9m47s

# 创建操作： 创建pvc的日志
$ kubectl apply -f raw-block-pvc.yaml
$ kubectl logs -f --tail=100 csi-rbdplugin-provisioner-5cc9f558c7-d2stz -c csi-provisioner
$ kubectl logs -f --tail=100 csi-rbdplugin-provisioner-5cc9f558c7-d2stz -c csi-rbdplugin

# csi-provisioner log: 接收到创建pvc的指令，下发指令创建I0310 09:21:35.323991       1 connection.go:183] GRPC call: /csi.v1.Controller/CreateVolumeI0310 09:21:35.390997       1 controller.go:777] create volume rep: {CapacityBytes:2147483648 VolumeId:0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-7b50e081-a053-11ec-b2dd-fa163ed7971b VolumeContext:map[clusterID:4a9e463a-4853-4237-a5c5-9ae9d25bacda csi.storage.k8s.io/pv/name:pvc-4e52c163-a593-4cc1-af59-23367d1e7573 csi.storage.k8s.io/pvc/name:raw-block-pvc csi.storage.k8s.io/pvc/namespace:default imageFeatures:layering imageName:csi-vol-7b50e081-a053-11ec-b2dd-fa163ed7971b journalPool:kubernetes pool:kubernetes] ContentSource:<nil> AccessibleTopology:[] XXX_NoUnkeyedLiteral:{} XXX_unrecognized:[] XXX_sizecache:0}I0310 09:21:35.391046       1 controller.go:861] successfully created PV pvc-4e52c163-a593-4cc1-af59-23367d1e7573 for PVC raw-block-pvc and csi volume name 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-7b50e081-a053-11ec-b2dd-fa163ed7971b# csi-rbdplugin log:  收到创建存储指令，调用rbd创建pv成功I0310 09:21:35.350755       1 rbd_journal.go:482] ID: 1609 Req-ID: pvc-4e52c163-a593-4cc1-af59-23367d1e7573 generated Volume ID (0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-7b50e081-a053-11ec-b2dd-fa163ed7971b) and image name (csi-vol-7b50e081-a053-11ec-b2dd-fa163ed7971b) for request name (pvc-4e52c163-a593-4cc1-af59-23367d1e7573)I0310 09:21:35.350822       1 rbd_util.go:352] ID: 1609 Req-ID: pvc-4e52c163-a593-4cc1-af59-23367d1e7573 rbd: create kubernetes/csi-vol-7b50e081-a053-11ec-b2dd-fa163ed7971b size 2048M (features: [layering]) using mon 172.20.163.52:6789,172.20.163.52:6789,172.20.163.52:6789I0310 09:21:35.374165       1 controllerserver.go:666] ID: 1609 Req-ID: pvc-4e52c163-a593-4cc1-af59-23367d1e7573 created image kubernetes/csi-vol-7b50e081-a053-11ec-b2dd-fa163ed7971b backed for request name pvc-4e52c163-a593-4cc1-af59-23367d1e7573

# 删除操作：删除pvc的日志
$ kubectl delete -f raw-block-pvc.yaml
$ kubectl logs -f --tail=100 csi-rbdplugin-provisioner-5cc9f558c7-d2stz -c csi-provisioner
$ kubectl logs -f --tail=100 csi-rbdplugin-provisioner-5cc9f558c7-d2stz -c csi-rbdplugin

# csi-provisioner log: 接收到删除pvc的请求，进行下发指令执行删除pvI0310 09:11:52.301723       1 controller.go:1413] delete "pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee": startedI0310 09:11:52.306652       1 connection.go:183] GRPC call: /csi.v1.Controller/DeleteVolumeI0310 09:11:52.306671       1 connection.go:184] GRPC request: {"secrets":"***stripped***","volume_id":"0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b"}I0310 09:11:53.088151       1 connection.go:186] GRPC response: {}I0310 09:11:53.088204       1 connection.go:187] GRPC error: <nil>I0310 09:11:53.088220       1 controller.go:1428] delete "pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee": volume deletedI0310 09:11:53.098260       1 controller.go:1478] delete "pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee": persistentvolume deletedI0310 09:11:53.098290       1 controller.go:1483] delete "pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee": succeededI0310 09:11:54.915543       1 leaderelection.go:278] successfully renewed lease default/rbd-csi-ceph-com# csi-rbdplugin log: 收到删除存储指令，调用rbd删除pv成功I0310 09:11:52.390569       1 rbd_util.go:644] ID: 1598 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b rbd: delete csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971b-temp using mon 172.20.163.52:6789,172.20.163.52:6789,172.20.163.52:6789, pool kubernetesI0310 09:11:52.394786       1 controllerserver.go:947] ID: 1598 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b deleting image csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971bI0310 09:11:52.394815       1 rbd_util.go:644] ID: 1598 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b rbd: delete csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971b using mon 172.20.163.52:6789,172.20.163.52:6789,172.20.163.52:6789, pool kubernetesask to remove image "kubernetes/csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971b" with id "11cbf1b83337" from trashI0310 09:11:53.087702       1 omap.go:123] ID: 1598 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b removed omap keys (pool="kubernetes", namespace="", name="csi.volumes.default"): [csi.volume.pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee]

csi如何进行pod rbd存储的create和delete

# 创建操作：挂载rbd存储到pod进行跟踪log[在node节点查看]
$ kubectl create -f raw-block-pod.yaml
$ kubectl logs -f --tail=10 csi-rbdplugin-9tfnm -c csi-rbdplugin

# csi通过rbd进行创建块设备，并映射到宿主机/dev/rbd0I0310 08:31:22.769929   11099 cephcmds.go:63] ID: 1900 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b command succeeded: rbd [--id kubernetes -m 172.20.163.52:6789,172.20.163.52:6789,172.20.163.52:6789 --keyfile=***stripped*** map kubernetes/csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971b --device-type krbd --options noudev]I0310 08:31:22.769972   11099 nodeserver.go:391] ID: 1900 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b rbd image: kubernetes/csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971b was successfully mapped at /dev/rbd0# 格式化/dev/rbd0块设备为ext4文件系统并成功挂载给podI0310 08:31:22.823277   11099 mount_linux.go:376] Checking for issues with fsck on disk: /dev/rbd0I0310 08:31:22.894904   11099 mount_linux.go:477] Attempting to mount disk /dev/rbd0 in ext4 format at /var/lib/kubelet/plugins/kubernetes.io/csi/pv/pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee/globalmount/0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971bI0310 08:31:22.894960   11099 mount_linux.go:183] Mounting cmd (mount) with arguments (-t ext4 -o _netdev,discard,defaults /dev/rbd0 /var/lib/kubelet/plugins/kubernetes.io/csi/pv/pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee/globalmount/0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b)I0310 08:31:22.997909   11099 resizefs_linux.go:124] ResizeFs.needResize - checking mounted volume /dev/rbd0I0310 08:31:23.000412   11099 resizefs_linux.go:128] Ext size: filesystem size=2147483648, block size=4096I0310 08:31:23.000433   11099 resizefs_linux.go:140] Volume /dev/rbd0: device size=2147483648, filesystem size=2147483648, block size=4096I0310 08:31:23.000502   11099 nodeserver.go:351] ID: 1900 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b rbd: successfully mounted volume 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b to stagingTargetPath /var/lib/kubelet/plugins/kubernetes.io/csi/pv/pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee/globalmount/0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b

# 删除操作：已挂载rbd存储的pod进行删除跟踪log[在node节点查看]
$ kubectl delete pod pod-with-raw-block-volume
$ kubectl logs -f --tail=10 csi-rbdplugin-9tfnm -c csi-rbdplugin

I0310 07:57:38.477003   11099 mount_linux.go:294] Unmounting /var/lib/kubelet/pods/b81968d7-1f46-4076-8f90-36c2b1e2ea86/volumes/kubernetes.io~csi/pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee/mount# 从pod中umount掉这个卷I0310 07:57:38.485433   11099 nodeserver.go:864] ID: 1862 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b rbd: successfully unbound volume 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b from /var/lib/kubelet/pods/b81968d7-1f46-4076-8f90-36c2b1e2ea86/volumes/kubernetes.io~csi/pvc-a92ceaaf-e400-4a74-a578-1fa0292ebdee/mount# 从宿主机rbd umap掉块设备I0310 07:57:38.777236   11099 cephcmds.go:63] ID: 1864 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b command succeeded: rbd [unmap kubernetes/csi-vol-b8f108b8-a022-11ec-b2dd-fa163ed7971b --device-type krbd --options noudev]I0310 07:57:38.777270   11099 nodeserver.go:977] ID: 1864 Req-ID: 0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b successfully unmapped volume (0001-0024-4a9e463a-4853-4237-a5c5-9ae9d25bacda-0000000000000005-b8f108b8-a022-11ec-b2dd-fa163ed7971b)

参考文献

[1] ceph.com 作者 202203
[2]良凯尔 作者 202203