@@ -10,6 +10,7 @@ import (
1010 goruntime "runtime"
1111 "strconv"
1212 "strings"
13+ "sync"
1314 "time"
1415
1516 systemd "github.com/coreos/go-systemd/v22/daemon"
@@ -32,6 +33,7 @@ import (
3233 "github.com/k3s-io/k3s/pkg/nodeconfig"
3334 "github.com/k3s-io/k3s/pkg/profile"
3435 "github.com/k3s-io/k3s/pkg/rootless"
36+ "github.com/k3s-io/k3s/pkg/signals"
3537 "github.com/k3s-io/k3s/pkg/spegel"
3638 "github.com/k3s-io/k3s/pkg/util"
3739 "github.com/k3s-io/k3s/pkg/version"
@@ -147,34 +149,29 @@ func run(ctx context.Context, cfg cmds.Agent, proxy proxy.Proxy) error {
147149 }
148150 }
149151
150- if nodeConfig .Docker {
151- if err := executor .Docker (ctx , nodeConfig ); err != nil {
152- return err
153- }
154- } else if nodeConfig .ContainerRuntimeEndpoint == "" {
155- if err := containerd .SetupContainerdConfig (nodeConfig ); err != nil {
156- return err
157- }
158- if err := executor .Containerd (ctx , nodeConfig ); err != nil {
159- return err
160- }
161- } else {
162- if err := executor .CRI (ctx , nodeConfig ); err != nil {
163- return err
164- }
165- }
152+ // Create a new context to use for agent components that is cancelled on a
153+ // delay after the signal context. This allows other things (like etcd) to
154+ // clean up, before agent components exit when their contexts are cancelled.
155+ ctx = util .DelayCancel (ctx , util .DefaultContextDelay )
166156
167157 notifySocket := os .Getenv ("NOTIFY_SOCKET" )
168158 os .Unsetenv ("NOTIFY_SOCKET" )
169159
160+ go func () {
161+ if err := startCRI (ctx , nodeConfig ); err != nil {
162+ signals .RequestShutdown (pkgerrors .WithMessage (err , "failed to start container runtime" ))
163+ }
164+ }()
165+
170166 if err := setupTunnelAndRunAgent (ctx , nodeConfig , cfg , proxy ); err != nil {
171167 return err
172168 }
173169
174170 go func () {
175171 <- executor .APIServerReadyChan ()
176- if err := startNetwork (ctx , nodeConfig ); err != nil {
177- logrus .Fatalf ("Failed to start networking: %v" , err )
172+ if err := startNetwork (ctx , & sync.WaitGroup {}, nodeConfig ); err != nil {
173+ signals .RequestShutdown (pkgerrors .WithMessage (err , "failed to start networking" ))
174+ return
178175 }
179176
180177 // By default, the server is responsible for notifying systemd
@@ -189,9 +186,23 @@ func run(ctx context.Context, cfg cmds.Agent, proxy proxy.Proxy) error {
189186 return nil
190187}
191188
189+ // startCRI starts the configured CRI, or waits for an external CRI to be ready.
190+ func startCRI (ctx context.Context , nodeConfig * daemonconfig.Node ) error {
191+ if nodeConfig .Docker {
192+ return executor .Docker (ctx , nodeConfig )
193+ } else if nodeConfig .ContainerRuntimeEndpoint == "" {
194+ if err := containerd .SetupContainerdConfig (nodeConfig ); err != nil {
195+ return err
196+ }
197+ return executor .Containerd (ctx , nodeConfig )
198+ } else {
199+ return executor .CRI (ctx , nodeConfig )
200+ }
201+ }
202+
192203// startNetwork updates the network annotations on the node, and starts flannel
193204// and the kube-router netpol controller, if enabled.
194- func startNetwork (ctx context.Context , nodeConfig * daemonconfig.Node ) error {
205+ func startNetwork (ctx context.Context , wg * sync. WaitGroup , nodeConfig * daemonconfig.Node ) error {
195206 // Use the kubelet kubeconfig to update annotations on the local node
196207 kubeletClient , err := util .GetClientSet (nodeConfig .AgentConfig .KubeConfigKubelet )
197208 if err != nil {
@@ -203,13 +214,13 @@ func startNetwork(ctx context.Context, nodeConfig *daemonconfig.Node) error {
203214 }
204215
205216 if ! nodeConfig .NoFlannel {
206- if err := flannel .Run (ctx , nodeConfig ); err != nil {
217+ if err := flannel .Run (ctx , wg , nodeConfig ); err != nil {
207218 return err
208219 }
209220 }
210221
211222 if ! nodeConfig .AgentConfig .DisableNPC {
212- if err := netpol .Run (ctx , nodeConfig ); err != nil {
223+ if err := netpol .Run (ctx , wg , nodeConfig ); err != nil {
213224 return err
214225 }
215226 }
@@ -264,7 +275,7 @@ func getConntrackConfig(nodeConfig *daemonconfig.Node) (*kubeproxyconfig.KubePro
264275// RunStandalone bootstraps the executor, but does not run the kubelet or containerd.
265276// This allows other bits of code that expect the executor to be set up properly to function
266277// even when the agent is disabled.
267- func RunStandalone (ctx context.Context , cfg cmds.Agent ) error {
278+ func RunStandalone (ctx context.Context , wg * sync. WaitGroup , cfg cmds.Agent ) error {
268279 proxy , err := createProxyAndValidateToken (ctx , & cfg )
269280 if err != nil {
270281 return err
@@ -308,7 +319,7 @@ func RunStandalone(ctx context.Context, cfg cmds.Agent) error {
308319
309320// Run sets up cgroups, configures the LB proxy, and triggers startup
310321// of containerd and kubelet.
311- func Run (ctx context.Context , cfg cmds.Agent ) error {
322+ func Run (ctx context.Context , wg * sync. WaitGroup , cfg cmds.Agent ) error {
312323 if err := cgroups .Validate (); err != nil {
313324 return err
314325 }
0 commit comments