Fixed e2e tests for HA master.

cluster/gce/debian/master-helper.sh

@@ -56,7 +56,7 @@ function replicate-master-instance() {
   ETCD_CA_KEY="$(echo "${kube_env}" | grep "ETCD_CA_KEY" |  sed "s/^.*: '//" | sed "s/'$//")"
   ETCD_CA_CERT="$(echo "${kube_env}" | grep "ETCD_CA_CERT" |  sed "s/^.*: '//" | sed "s/'$//")"
 
-  create-etcd-certs "${ETCD_CA_CERT}" "${ETCD_CA_KEY}"
+  create-etcd-certs "${REPLICA_NAME}" "${ETCD_CA_CERT}" "${ETCD_CA_KEY}"
 
   kube_env="$(echo "${kube_env}" | grep -v "ETCD_PEER_KEY")"
   kube_env="$(echo -e "${kube_env}\nETCD_PEER_KEY: '${ETCD_PEER_KEY_BASE64}'")"

cluster/gce/gci/master-helper.sh

@@ -52,7 +52,7 @@ function replicate-master-instance() {
   ETCD_CA_KEY="$(echo "${kube_env}" | grep "ETCD_CA_KEY" |  sed "s/^.*: '//" | sed "s/'$//")"
   ETCD_CA_CERT="$(echo "${kube_env}" | grep "ETCD_CA_CERT" |  sed "s/^.*: '//" | sed "s/'$//")"
 
-  create-etcd-certs "${ETCD_CA_CERT}" "${ETCD_CA_KEY}"
+  create-etcd-certs "${REPLICA_NAME}" "${ETCD_CA_CERT}" "${ETCD_CA_KEY}"
 
   kube_env="$(echo "${kube_env}" | grep -v "ETCD_PEER_KEY")"
   kube_env="$(echo -e "${kube_env}\nETCD_PEER_KEY: '${ETCD_PEER_KEY_BASE64}'")"

cluster/gce/util.sh

@@ -747,8 +747,9 @@ function get-master-disk-size() {
 #   KUBE_TEMP: temporary directory
 #
 # Args:
-#  $1: CA certificate
-#  $2: CA key
+#  $1: host name
+#  $2: CA certificate
+#  $3: CA key
 #
 # If CA cert/key is empty, the function will also generate certs for CA.
 #
@@ -759,8 +760,9 @@ function get-master-disk-size() {
 #   ETCD_PEER_CERT_BASE64
 #
 function create-etcd-certs {
-  local ca_cert=${1:-}
-  local ca_key=${2:-}
+  local host=${1}
+  local ca_cert=${2:-}
+  local ca_key=${3:-}
 
   mkdir -p "${KUBE_TEMP}/cfssl"
   pushd "${KUBE_TEMP}/cfssl"
@@ -810,8 +812,8 @@ EOF
     ./cfssl gencert -initca ca-csr.json | ./cfssljson -bare ca -
   fi
 
-  echo '{"CN":"'"${MASTER_NAME}"'","hosts":[""],"key":{"algo":"ecdsa","size":256}}' \
-      | ./cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=client-server -hostname="${MASTER_NAME}" - \
+  echo '{"CN":"'"${host}"'","hosts":[""],"key":{"algo":"ecdsa","size":256}}' \
+      | ./cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=client-server -hostname="${host}" - \
       | ./cfssljson -bare etcd
 
   ETCD_CA_KEY_BASE64=$(cat "ca-key.pem" | base64 | tr -d '\r\n')
@@ -878,7 +880,7 @@ function create-master() {
   MASTER_ADVERTISE_ADDRESS="${MASTER_RESERVED_IP}"
 
   create-certs "${MASTER_RESERVED_IP}"
-  create-etcd-certs
+  create-etcd-certs ${MASTER_NAME}
 
   # Sets MASTER_ROOT_DISK_SIZE that is used by create-master-instance
   get-master-root-disk-size
@@ -904,7 +906,7 @@ function add-replica-to-etcd() {
     --project "${PROJECT}" \
     --zone "${EXISTING_MASTER_ZONE}" \
     --command \
-      "curl localhost:${client_port}/v2/members -XPOST -H \"Content-Type: application/json\" -d '{\"peerURLs\":[\"https://${REPLICA_NAME}:${internal_port}\"]}'"
+      "curl localhost:${client_port}/v2/members -XPOST -H \"Content-Type: application/json\" -d '{\"peerURLs\":[\"https://${REPLICA_NAME}:${internal_port}\"]}' -s"
   return $?
 }
 
@@ -1381,7 +1383,7 @@ function kube-down() {
   if [[ "${REMAINING_MASTER_COUNT}" == "1" ]]; then
     if gcloud compute forwarding-rules describe "${MASTER_NAME}" --region "${REGION}" --project "${PROJECT}" &>/dev/null; then
       detect-master
-      local REMAINING_REPLICA_NAME="$(get-replica-name)"
+      local REMAINING_REPLICA_NAME="$(get-all-replica-names)"
       local REMAINING_REPLICA_ZONE=$(gcloud compute instances list "${REMAINING_REPLICA_NAME}" \
         --project "${PROJECT}" --format="value(zone)")
       gcloud compute forwarding-rules delete \
@@ -1476,6 +1478,21 @@ function kube-down() {
     # If there are no more remaining master replicas, we should update kubeconfig.
     export CONTEXT="${PROJECT}_${INSTANCE_PREFIX}"
     clear-kubeconfig
+  else
+  # If some master replicas remain: cluster has been changed, we need to re-validate it.
+    echo "... calling validate-cluster" >&2
+    # Override errexit
+    (validate-cluster) && validate_result="$?" || validate_result="$?"
+
+    # We have two different failure modes from validate cluster:
+    # - 1: fatal error - cluster won't be working correctly
+    # - 2: weak error - something went wrong, but cluster probably will be working correctly
+    # We just print an error message in case 2).
+    if [[ "${validate_result}" == "1" ]]; then
+      exit 1
+    elif [[ "${validate_result}" == "2" ]]; then
+      echo "...ignoring non-fatal errors in validate-cluster" >&2
+    fi
   fi
   set -e
 }
@@ -1511,6 +1528,19 @@ function get-all-replica-names() {
     --format "value(name)" | tr "\n" "," | sed 's/,$//')
 }
 
+# Prints the number of all of the master replicas in all zones.
+#
+# Assumed vars:
+#   MASTER_NAME
+function get-master-replicas-count() {
+  detect-project
+  local num_masters=$(gcloud compute instances list \
+    --project "${PROJECT}" \
+    --regexp "$(get-replica-name-regexp)" \
+    --format "value(zone)" | wc -l)
+  echo -n "${num_masters}"
+}
+
 # Prints regexp for full master machine name. In a cluster with replicated master,
 # VM names may either be MASTER_NAME or MASTER_NAME with a suffix for a replica.
 function get-replica-name-regexp() {
@@ -1786,7 +1816,7 @@ function test-setup() {
   # Detect the project into $PROJECT if it isn't set
   detect-project
 
-  if [[ ${MULTIZONE:-} == "true" ]]; then
+  if [[ ${MULTIZONE:-} == "true" && -n ${E2E_ZONES:-} ]]; then
     for KUBE_GCE_ZONE in ${E2E_ZONES}
     do
       KUBE_GCE_ZONE="${KUBE_GCE_ZONE}" KUBE_USE_EXISTING_MASTER="${KUBE_USE_EXISTING_MASTER:-}" "${KUBE_ROOT}/cluster/kube-up.sh"
@@ -1843,7 +1873,7 @@ function test-teardown() {
   delete-firewall-rules \
     "${NODE_TAG}-${INSTANCE_PREFIX}-http-alt" \
     "${NODE_TAG}-${INSTANCE_PREFIX}-nodeports"
-  if [[ ${MULTIZONE:-} == "true" ]]; then
+  if [[ ${MULTIZONE:-} == "true" && -n ${E2E_ZONES:-} ]]; then
       local zones=( ${E2E_ZONES} )
       # tear them down in reverse order, finally tearing down the master too.
       for ((zone_num=${#zones[@]}-1; zone_num>0; zone_num--))

cluster/validate-cluster.sh

@@ -52,7 +52,12 @@ CLUSTER_READY_ADDITIONAL_TIME_SECONDS="${CLUSTER_READY_ADDITIONAL_TIME_SECONDS:-
 
 EXPECTED_NUM_NODES="${NUM_NODES}"
 if [[ "${REGISTER_MASTER_KUBELET:-}" == "true" ]]; then
-  EXPECTED_NUM_NODES=$((EXPECTED_NUM_NODES+1))
+  if [[ "${KUBERNETES_PROVIDER:-}" == "gce" ]]; then
+    NUM_MASTERS=$(get-master-replicas-count)
+  else
+    NUM_MASTERS=1
+  fi
+  EXPECTED_NUM_NODES=$((EXPECTED_NUM_NODES+NUM_MASTERS))
 fi
 REQUIRED_NUM_NODES=$((EXPECTED_NUM_NODES - ALLOWED_NOTREADY_NODES))
 # Make several attempts to deal with slow cluster birth.

hack/e2e-internal/e2e-add-master.sh

@@ -20,6 +20,7 @@ if [[ ! -z "${1:-}" ]]; then
   export KUBE_GCE_ZONE="${1}"
 fi
 export KUBE_REPLICATE_EXISTING_MASTER=true
+export MULTIZONE=true
 
 source "${KUBE_ROOT}/hack/e2e-internal/e2e-up.sh"
 

test/e2e/framework/util.go

@@ -4011,6 +4011,43 @@ func WaitForClusterSize(c clientset.Interface, size int, timeout time.Duration)
 	return fmt.Errorf("timeout waiting %v for cluster size to be %d", timeout, size)
 }
 
+// waitForMasters waits until the cluster has the desired number of ready masters in it.
+func WaitForMasters(masterPrefix string, c clientset.Interface, size int, timeout time.Duration) error {
+	for start := time.Now(); time.Since(start) < timeout; time.Sleep(20 * time.Second) {
+		nodes, err := c.Core().Nodes().List(api.ListOptions{})
+		if err != nil {
+			Logf("Failed to list nodes: %v", err)
+			continue
+		}
+
+		// Filter out nodes that are not master replicas
+		FilterNodes(nodes, func(node api.Node) bool {
+			res, err := regexp.Match(masterPrefix+"(-...)?", ([]byte)(node.Name))
+			if err != nil {
+				Logf("Failed to match regexp to node name: %v", err)
+				return false
+			}
+			return res
+		})
+
+		numNodes := len(nodes.Items)
+
+		// Filter out not-ready nodes.
+		FilterNodes(nodes, func(node api.Node) bool {
+			return IsNodeConditionSetAsExpected(&node, api.NodeReady, true)
+		})
+
+		numReady := len(nodes.Items)
+
+		if numNodes == size && numReady == size {
+			Logf("Cluster has reached the desired number of masters %d", size)
+			return nil
+		}
+		Logf("Waiting for the number of masters %d, current %d, not ready master nodes %d", size, numNodes, numNodes-numReady)
+	}
+	return fmt.Errorf("timeout waiting %v for the number of masters to be %d", timeout, size)
+}
+
 // GetHostExternalAddress gets the node for a pod and returns the first External
 // address. Returns an error if the node the pod is on doesn't have an External
 // address.

test/e2e/ha_master.go

@@ -23,6 +23,7 @@ import (
 	"path"
 	"strconv"
 	"strings"
+	"time"
 
 	. "github.com/onsi/ginkgo"
 	clientset "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
@@ -63,7 +64,6 @@ func createNewRC(c clientset.Interface, ns string, name string) {
 func verifyNumberOfMasterReplicas(expected int) {
 	output, err := exec.Command("gcloud", "compute", "instances", "list",
 		"--project="+framework.TestContext.CloudConfig.ProjectID,
-		"--zones="+framework.TestContext.CloudConfig.Zone,
 		"--regexp="+framework.TestContext.CloudConfig.MasterName+"(-...)?",
 		"--filter=status=RUNNING",
 		"--format=[no-heading]").CombinedOutput()
@@ -73,7 +73,7 @@ func verifyNumberOfMasterReplicas(expected int) {
 	replicas := bytes.Count(output, newline)
 	framework.Logf("Num master replicas/expected: %d/%d", replicas, expected)
 	if replicas != expected {
-		framework.Failf("Wrong number of master replicas")
+		framework.Failf("Wrong number of master replicas %d expected %d", replicas, expected)
 	}
 }
 
@@ -131,6 +131,8 @@ var _ = framework.KubeDescribe("HA-master [Feature:HAMaster]", func() {
 		for _, zone := range additionalReplicaZones {
 			removeMasterReplica(zone)
 		}
+		framework.WaitForMasters(framework.TestContext.CloudConfig.MasterName, c, 1, 10*time.Minute)
+		verifyNumberOfMasterReplicas(1)
 	})
 
 	type Action int
@@ -151,6 +153,7 @@ var _ = framework.KubeDescribe("HA-master [Feature:HAMaster]", func() {
 			additionalReplicaZones = removeZoneFromZones(additionalReplicaZones, zone)
 		}
 		verifyNumberOfMasterReplicas(len(additionalReplicaZones) + 1)
+		framework.WaitForMasters(framework.TestContext.CloudConfig.MasterName, c, len(additionalReplicaZones)+1, 10*time.Minute)
 
 		// Verify that API server works correctly with HA master.
 		rcName := "ha-master-" + strconv.Itoa(len(existingRCs))
@@ -159,16 +162,19 @@ var _ = framework.KubeDescribe("HA-master [Feature:HAMaster]", func() {
 		verifyRCs(c, ns, existingRCs)
 	}
 
-	It("pods survive addition/removal same zone [Slow]", func() {
+	It("survive addition/removal replicas same zone [Serial][Disruptive]", func() {
 		zone := framework.TestContext.CloudConfig.Zone
 		step(None, "")
-		step(AddReplica, zone)
-		step(AddReplica, zone)
-		step(RemoveReplica, zone)
-		step(RemoveReplica, zone)
+		numAdditionalReplicas := 2
+		for i := 0; i < numAdditionalReplicas; i++ {
+			step(AddReplica, zone)
+		}
+		for i := 0; i < numAdditionalReplicas; i++ {
+			step(RemoveReplica, zone)
+		}
 	})
 
-	It("pods survive addition/removal different zones [Slow]", func() {
+	It("survive addition/removal replicas different zones [Serial][Disruptive]", func() {
 		zone := framework.TestContext.CloudConfig.Zone
 		region := findRegionForZone(zone)
 		zones := findZonesForRegion(region)