Merge pull request #72 from justinsb/upup_polish_2

upup: misc polish marching towards v1

upup/cmd/cloudup/main.go

@@ -157,10 +157,18 @@ func (c *CreateClusterCmd) Run() error {
 		if c.Config.KubernetesVersion == "" {
 			return fmt.Errorf("Must either specify a KubernetesVersion (-kubernetes-version) or provide an asset with the release bundle")
 		}
-		defaultReleaseAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/release/v%s/kubernetes-server-linux-amd64.tar.gz", c.Config.KubernetesVersion)
-		glog.Infof("Adding default kubernetes release asset: %s", defaultReleaseAsset)
-		// TODO: Verify it exists, get the hash (that will check that KubernetesVersion is valid)
-		c.Config.Assets = append(c.Config.Assets, defaultReleaseAsset)
+		//defaultReleaseAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/release/v%s/kubernetes-server-linux-amd64.tar.gz", c.Config.KubernetesVersion)
+		//glog.Infof("Adding default kubernetes release asset: %s", defaultReleaseAsset)
+
+		defaultKubeletAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/release/v%s/bin/linux/amd64/kubelet", c.Config.KubernetesVersion)
+		glog.Infof("Adding default kubelet release asset: %s", defaultKubeletAsset)
+
+		defaultKubectlAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/release/v%s/bin/linux/amd64/kubectl", c.Config.KubernetesVersion)
+		glog.Infof("Adding default kubelet release asset: %s", defaultKubectlAsset)
+
+		// TODO: Verify assets exist, get the hash (that will check that KubernetesVersion is valid)
+
+		c.Config.Assets = append(c.Config.Assets, defaultKubeletAsset, defaultKubectlAsset)
 	}
 
 	if c.Config.NodeUp.Location == "" {
@@ -169,12 +177,6 @@ func (c *CreateClusterCmd) Run() error {
 		c.Config.NodeUp.Location = location
 	}
 
-	if useProtokube {
-		location := "https://kubeupv2.s3.amazonaws.com/protokube/protokube.tar.gz"
-		glog.Infof("Using default protokube location: %q", location)
-		c.Config.Assets = append(c.Config.Assets, location)
-	}
-
 	var cloud fi.Cloud
 
 	var project string

upup/models/cloudup/resources/config.yaml.template

@@ -37,6 +37,9 @@ Tokens:
   "system:logging": {{ GetOrCreateSecret "system:logging" }}
   "system:monitoring": {{ GetOrCreateSecret "system:monitoring" }}
   "system:dns": {{ GetOrCreateSecret "system:dns" }}
+{{ if HasTag "_kope_routing" }}
+  "kope-routing": {{ GetOrCreateSecret "kope-routing" }}
+{{ end }}
 
 Tags:
 {{ range $tag := Args }}

upup/models/nodeup/_kope_routing/files/etc/kubernetes/manifests/kope-routing.manifest.template

@@ -0,0 +1,47 @@
+# kope-routing podspec
+apiVersion: v1
+kind: Pod
+metadata:
+  name: kope-routing
+  namespace: kube-system
+spec:
+  hostNetwork: true
+  containers:
+  - name: kope-routing
+    image: kope/route-controller
+    #resources:
+    #  requests:
+    #    cpu: {{ .KubeProxy.CPURequest }}
+    command:
+    - /bin/sh
+    - -c
+#    - /usr/bin/route-controller --boot-id=/var/lib/kope-routing/boot-id --kubeconfig=/var/lib/kope-routing/kubeconfig --healthz-port=10247 1>>/var/log/kope-routing.log 2>&1
+    - /usr/bin/route-controller --provider=gre --boot-id=/var/lib/kope-routing/boot-id --kubeconfig=/var/lib/kope-routing/kubeconfig --healthz-port=10247
+    securityContext:
+      privileged: true
+    volumeMounts:
+    - mountPath: /etc/ssl/certs
+      name: ssl-certs-host
+      readOnly: true
+#    - mountPath: /var/log
+#      name: varlog
+#      readOnly: false
+    - mountPath: /var/lib/kope-routing/kubeconfig
+      name: kubeconfig
+      readOnly: true
+    - mountPath: /var/lib/kope-routing/boot-id
+      name: host-boot-id
+      readOnly: true
+  volumes:
+  - hostPath:
+      path: /usr/share/ca-certificates
+    name: ssl-certs-host
+  - hostPath:
+      path: /var/lib/kope-routing/kubeconfig
+    name: kubeconfig
+  - hostPath:
+      path: /proc/sys/kernel/random/boot_id
+    name: host-boot-id
+#  - hostPath:
+#      path: /var/log
+#    name: varlog

upup/models/nodeup/_kope_routing/files/var/lib/kope-routing/kubeconfig.template

@@ -0,0 +1,17 @@
+apiVersion: v1
+kind: Config
+users:
+- name: kope-routing
+  user:
+    token: {{ .GetToken "kope-routing" }}
+clusters:
+- name: local
+  cluster:
+    certificate-authority-data: {{ Base64Encode .CACertificate.AsString }}
+    server: https://{{ .MasterInternalName }}
+contexts:
+- context:
+    cluster: local
+    user: kope-routing
+  name: service-account-context
+current-context: service-account-context

upup/models/nodeup/_kope_routing/files/var/lib/kope-routing/kubeconfig.template.meta

@@ -0,0 +1,3 @@
+{
+  "mode": "0400"
+}

upup/models/nodeup/_kope_routing/files/var/log/kope-routing.log.meta

@@ -0,0 +1,3 @@
+{
+  "ifNotExists": true
+}

upup/pkg/fi/cloudup/awstasks/security_group.go

@@ -36,7 +36,7 @@ func (e *SecurityGroup) Find(c *fi.Context) (*SecurityGroup, error) {
 		return nil, nil
 	}
 
-	filters := cloud.BuildFilters(nil) // TODO: Do we need any filters here - done by group-name
+	filters := cloud.BuildFilters(e.Name)
 	filters = append(filters, awsup.NewEC2Filter("vpc-id", *vpcID))
 	filters = append(filters, awsup.NewEC2Filter("group-name", *e.Name))
 

upup/pkg/fi/cloudup/loader.go

@@ -161,7 +161,7 @@ func (l *Loader) Build(baseDir string) (map[string]fi.Task, error) {
 	if err != nil {
 		return nil, err
 	}
-	glog.Infof("options: %s", fi.DebugAsJsonStringIndent(l.config))
+	glog.V(1).Infof("options: %s", fi.DebugAsJsonStringIndent(l.config))
 
 	// Second pass: load everything else
 	tw = &loader.TreeWalker{
@@ -218,7 +218,7 @@ func (l *Loader) processDeferrals() error {
 								return fmt.Errorf("Unable to find task %q, referenced from %s:%s", *name, taskKey, path)
 							}
 
-							glog.Infof("Replacing task %q at %s:%s", *name, taskKey, path)
+							glog.V(4).Infof("Replacing task %q at %s:%s", *name, taskKey, path)
 							v.Set(reflect.ValueOf(primary))
 						}
 						return utils.SkipReflection
@@ -372,7 +372,7 @@ func (l *Loader) loadObjectMap(key string, data map[string]interface{}) (map[str
 
 		o := reflect.New(t)
 
-		glog.Warningf("replace with partial unmarshal...")
+		// TODO replace with partial unmarshal...
 		jsonValue, err := json.Marshal(v)
 		if err != nil {
 			return nil, fmt.Errorf("error marshalling to json: %v", err)
@@ -381,7 +381,7 @@ func (l *Loader) loadObjectMap(key string, data map[string]interface{}) (map[str
 		if err != nil {
 			return nil, fmt.Errorf("error parsing %q: %v", key, err)
 		}
-		glog.Infof("Built %s:%s => %v", key, k, o.Interface())
+		glog.V(4).Infof("Built %s:%s => %v", key, k, o.Interface())
 
 		if inferredName {
 			hn, ok := o.Interface().(fi.HasName)

upup/pkg/fi/executor.go

@@ -5,8 +5,11 @@ import (
 	"github.com/golang/glog"
 	"strings"
 	"sync"
+	"time"
 )
 
+const MaxAttemptsWithNoProgress = 3
+
 type executor struct {
 	context *Context
 }
@@ -43,10 +46,13 @@ func (e *executor) RunTasks(taskMap map[string]Task) error {
 		}
 	}
 
+	noProgressCount := 0
 	for {
 		var canRun []*taskState
+		doneCount := 0
 		for _, ts := range taskStates {
 			if ts.done {
+				doneCount++
 				continue
 			}
 			ready := true
@@ -61,6 +67,7 @@ func (e *executor) RunTasks(taskMap map[string]Task) error {
 			}
 		}
 
+		glog.Infof("Tasks: %d done / %d total; %d can run", doneCount, len(taskStates), len(canRun))
 		if len(canRun) == 0 {
 			break
 		}
@@ -86,12 +93,19 @@ func (e *executor) RunTasks(taskMap map[string]Task) error {
 
 		if !progress {
 			if len(errors) != 0 {
-				// TODO: Sleep and re-attempt?
-				return fmt.Errorf("did not make any progress executing task.  Example error: %v", errors[0])
+				noProgressCount++
+				if noProgressCount == MaxAttemptsWithNoProgress {
+					return fmt.Errorf("did not make any progress executing task.  Example error: %v", errors[0])
+				} else {
+					glog.Infof("No progress made, sleeping before retrying failed tasks")
+					time.Sleep(10 * time.Second)
+				}
 			} else {
 				// Logic error!
 				panic("did not make progress executing tasks; but no errors reported")
 			}
+		} else {
+			noProgressCount = 0
 		}
 	}
 

upup/pkg/fi/nodeup/config.go

@@ -120,10 +120,11 @@ type APIServerConfig struct {
 type KubeControllerManagerConfig struct {
 	CloudProvider string `flag:"cloud-provider"`
 
-	Master            string `flag:"master"`
-	ClusterName       string `flag:"cluster-name"`
-	ClusterCIDR       string `flag:"cluster-cidr"`
-	AllocateNodeCIDRs *bool  `flag:"allocate-node-cidrs"`
+	Master               string `flag:"master"`
+	ClusterName          string `flag:"cluster-name"`
+	ClusterCIDR          string `flag:"cluster-cidr"`
+	AllocateNodeCIDRs    *bool  `flag:"allocate-node-cidrs"`
+	ConfigureCloudRoutes *bool  `flag:"configure-cloud-routes"`
 	// TODO: Name verbosity or LogLevel
 	LogLevel    int   `flag:"v"`
 	LeaderElect *bool `flag:"leader-elect"`

upup/pkg/kutil/delete_cluster.go

@@ -39,9 +39,14 @@ func (c *DeleteCluster) ListResources() (map[string]DeletableResource, error) {
 
 	resources := make(map[string]DeletableResource)
 
-	filters := cloud.BuildFilters(nil)
 	tags := cloud.BuildTags(nil, nil)
 
+	var filters []*ec2.Filter
+	for k, v := range tags {
+		filter := awsup.NewEC2Filter("tag:"+k, v)
+		filters = append(filters, filter)
+	}
+
 	{
 		glog.V(2).Infof("Listing all Autoscaling groups matching cluster tags")
 		var asgNames []*string