containerd_cri_plugin
containerd cri plugin
初始化
PluginConfig
containerdConfig “containerd”
ContainerdRootDir
ContainerdEndpoint
RootDir
StateDir
Snapshotter string
"snapshortter"
DefaultRuntime Runtime
"default_runtime"
UntrustedWorkloadRuntime Runtime
"untrusted_workload_runtime"
Runtimes map[string] Runtime
runtimes
NoPivot deprecated 只影响v1.linux
Type string “runtime_type”
Engine string "runtime_engine" deprecated 只影响v1.linux
Root string "runtime_root" 只影响linux.v1
Options *toml.Primitive "options"
CniConfig “cni”
NetworkPluginBinDir string “bin_dir”
NetworkPluginConDir "conf_dir"
NetworkPluginConfTemplate string "conf_template"
Registry "registry"
map [string] Mirrors "mirrors"
Endpoints []string "endpoint"
map [string] AuthConfig "auths"
Username string "username"
Password string "password"
Auth string "auth"
IdetityToken "indetitytoken"
1. 在main（）启动之前，init（）函数将cri plugin注册到
containerd InitContext中的plugin数组中，该init函数会
赋值plugin的InitFn为InitCRIService
2. containerd 进入main函数会LoadPlugin，调用cri plugin的
initCRIService（）函数进行初始化
1. 赋值InitContext Meta的Platform
2. 赋值Exports map的CRIVersion
3. 加载ic.Config.(*criconfig.PluginConfig)
4. getServiceOpts()
问题
5. 初始化cri所需service的client端 client,err := containerd.New("",
containerd.WithDefaultNamespace(), containerd.WithServices(serviceOpts))
CRIService Interface定义了几个方法接口的声明
及cirService对这些接口的定义
Run() 位于service.go
io.Closer 即Close()方法 位于service.go
Plugin.Service 即Register(*grpc.Server) 位于service.go
grpcService
runtime.RuntimeServiceServer实现 Runtime相关
runtimer.ImageServiceServer
api.CRIPluginServiceServer
Version（） version.go
RunPodSandbox() sandbox_run.go
StopPodSandbox() sandbox_stop.go
RemovePodSandbox() sandbox_remove.go
ListPodSandbox() sandbox_list.go
PodSandboxStatus() sandbox_status.go
CreateContainer() container_create.go
StartContainer() container_start.go
StopContainer() container_stop.go
RemoveContainer() container_remove.go
ListContainers() container_list.go
ContainerStatus() container_status.go
UpdateContainerResources() container_update_resource.go
ReopenContainerLog() container_log_reopen.go
ExecSync() container_exesync.go
Exec() container_exec.go
Attach() container_attach.go
PortForward() sandbox_portforward.go
ContainerStats() container_stats.go
ListContainerStats() container_stats_list.go
UpdateRuntimeConfig() update_runtime_config.go
Status() container_stats.go
ListImages() image_list.go
ImagesStatus()
PullImage() image_pull.go
ImageFsInfo() imagefs_info.go
LoadImage() image_load.go
6. NewCRIService,构造CRIService
c.imageFSPath = imageFSPath(config.ContainerdRootDir,
config.ContainerdConfig.Snapshotter)
从配置文件中cni conf dir和cni bin dir加载netPlugin
创建StreamServer
CONFIG
criService
config criconfig.Config
imageFSPath
apparmorEnabled bool
seccompEnabled bool
os osinterface.os
sandboxStore *snadboxstore.Store
sandboxNameIndex *registar.Registar
containerStore *containerstore.Store
containerNameIndex *registrar.Registrar
imageStore *imagestore.Store
snapshotStore *snapshotstore.Store
netPlugin cni.CNI
client *containerd.Client
streamServer streaming.Server
eventMonitor *eventMonitor
sandboxes map[string]Sandbox
idIndex *truncindex.TruncIndex
Metadata
Status StatusStorage
Container containerd.Container
NetNS *netns.NetNS
*store.StopCh
nameToKey map[string] string
KeyToName map[string] string
refCache map[string] string
client *containerd.Client
store *store
address = “”，貌似实际上并不会同
containerd建立client端连接
snapshots map[string] Snapshot
Key string
Kind snapshot.Kind
Inodes uint64
Timestamp int64
1. criService中containerd.Client如何同containerd中其他plugin service进行交互？
2. criService中streamServer用于何种场景 ？
3. eventMonitor监控什么事件，如何监控？
service
connMu
conn *grpc.Clientconn
runtime string
connector func() (*grpc.ClientConn, error)
c *criService
ch <- chan *events.Envelope
errCh <- chan error
ctx context.Context
cancel context.CancelFunc
backOff *backOff
4. containerd创建grpc server后，会将plugin services注册到rpc，对于cri service，调用
criService的Register()方法，将criService注册，对于criService，主要注册三个模块的
server处理方法
1. RuntimeService相关
2. Image Service相关
3. CRI Plugin Service相关，目前就Load Image一种服务
3. 调用Run()开始启动criService
1. 注册containerd 事件订阅
/stasks/exit /tasks/oom /images
2. 开始recover pod、container和images
1. 通过container.Store列出所有Sandbox 并加载Sandbox状态
2. 通过container.Store列出所有containers并记载container状态
3. 恢复镜像信息
4. 清除所有无对应containerd 容器的孤儿Sandbox和container目录
3. 开始监控事件
4. 启动snapshots syncer
5. 启动streaming server
6. 监听eventMonitorErr和streamServerErr事件，进行出错处理
4. 为何将shim进程强制杀死后无法rmp
Created With
MindMaster