Add E2E Split Server to Drone, support parrallel testing in Drone (#9940

)

* Fix SE old test name
* E2E: support multiple VMs at once in CI with time prefix
* Add local binary support to split server test, add to drone CI
* Cleanup old VMs in drone

Signed-off-by: Derek Nola <[email protected]>

.drone.yml

@@ -617,33 +617,24 @@ steps:
   - mkdir -p dist/artifacts
   - cp /tmp/artifacts/* dist/artifacts/
   - docker stop registry && docker rm registry
-  # Cleanup VMs running, happens if a previous test panics
-  # Cleanup inactive domains, happens if previous test is canceled
-  - |
-    VMS=$(virsh list --name | grep '_server-\|_agent-' || true)
-    if [ -n "$VMS" ]; then
-      for vm in $VMS
-      do
-        virsh destroy $vm
-        virsh undefine $vm --remove-all-storage
-      done
-    fi 
-    VMS=$(virsh list --name --inactive | grep '_server-\|_agent-' || true)
-    if [ -n "$VMS" ]; then
-      for vm in $VMS
-      do
-        virsh undefine $vm
-      done
-    fi 
+  # Cleanup VMs that are older than 2h. Happens if a previous test panics or is canceled
+  - tests/e2e/scripts/cleanup_vms.sh
   - docker run -d -p 5000:5000 -e REGISTRY_PROXY_REMOTEURL=https://registry-1.docker.io --name registry registry:2
-  - cd tests/e2e/validatecluster
-  - vagrant destroy -f
-  - go test -v -timeout=45m ./validatecluster_test.go -ci -local
-  - cp ./coverage.out /tmp/artifacts/validate-coverage.out
-  - cd ../secretsencryption
-  - vagrant destroy -f
-  - go test -v -timeout=30m ./secretsencryption_test.go -ci -local
-  - cp ./coverage.out /tmp/artifacts/se-coverage.out
+  - |
+    cd tests/e2e/validatecluster
+    vagrant destroy -f
+    go test -v -timeout=45m ./validatecluster_test.go -ci -local
+    cp ./coverage.out /tmp/artifacts/validate-coverage.out
+  - |
+    cd ../secretsencryption
+    vagrant destroy -f
+    go test -v -timeout=30m ./secretsencryption_test.go -ci -local
+    cp ./coverage.out /tmp/artifacts/se-coverage.out
+  - |
+    cd ../splitserver
+    vagrant destroy -f
+    go test -v -timeout=30m ./splitserver_test.go -ci -local
+    cp ./coverage.out /tmp/artifacts/split-coverage.out
   - |
     if [ "$DRONE_BUILD_EVENT" = "pull_request" ]; then
       cd ../upgradecluster
@@ -669,6 +660,7 @@ steps:
     files:
       - /tmp/artifacts/validate-coverage.out
       - /tmp/artifacts/se-coverage.out
+      - /tmp/artifacts/split-coverage.out
       - /tmp/artifacts/upgrade-coverage.out
     flags: 
       - e2etests

tests/e2e/dualstack/Vagrantfile

@@ -90,6 +90,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
 
   if NODE_ROLES.kind_of?(String)

tests/e2e/embeddedmirror/Vagrantfile

@@ -97,6 +97,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/externalip/Vagrantfile

@@ -65,6 +65,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
 
   if NODE_ROLES.kind_of?(String)

tests/e2e/privateregistry/Vagrantfile

@@ -103,6 +103,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/rootless/Vagrantfile

@@ -45,6 +45,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/rotateca/Vagrantfile

@@ -65,6 +65,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/s3/Vagrantfile

@@ -69,6 +69,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/scripts/cleanup_vms.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+
+# Clean up any VMS that are older than 2 hours.
+# 
+# We embed the time in the VM name, so we can easily filter them out.
+
+# Get the current time in seconds since the epoch
+current_time=$(date +%s)
+
+# Get the list of VMs
+vms=$(virsh list --name --all)
+time_regex="_([0-9]+)_(server|agent)"
+# Cleanup running VMs, happens if a previous test panics
+for vm in $vms; do
+    if [[ $vm =~ $time_regex ]]; then
+        vm_time="${BASH_REMATCH[1]}"
+        age=$((current_time - vm_time))
+        if [ $age -gt 7200 ]; then
+            virsh destroy $vm
+            virsh undefine $vm --remove-all-storage
+        fi
+    fi
+done
+
+# Cleanup inactive domains, happens if previous test is canceled
+vms=$(virsh list --name --inactive)
+for vm in $vms; do
+    if [[ $vm =~ $time_regex ]]; then
+        vm_time="${BASH_REMATCH[1]}"
+        age=$((current_time - vm_time))
+        if [ $age -gt 7200 ]; then
+            virsh undefine $vm  --remove-all-storage
+        fi
+    fi
+done

tests/e2e/secretsencryption/Vagrantfile

@@ -9,7 +9,7 @@ GOCOVER = (ENV['E2E_GOCOVER'] || "")
 NODE_CPUS = (ENV['E2E_NODE_CPUS'] || 2).to_i
 NODE_MEMORY = (ENV['E2E_NODE_MEMORY'] || 2048).to_i
 # Virtualbox >= 6.1.28 require `/etc/vbox/network.conf` for expanded private networks 
-NETWORK_PREFIX = "10.10.10"
+NETWORK_PREFIX = "10.10.13"
 install_type = ""
 
 def provision(vm, role, role_num, node_num)
@@ -53,6 +53,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/secretsencryption_old/Vagrantfile

@@ -52,6 +52,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/secretsencryption_old/secretsencryption_test.go

@@ -22,7 +22,7 @@ var local = flag.Bool("local", false, "deploy a locally built K3s binary")
 // Environment Variables Info:
 // E2E_RELEASE_VERSION=v1.23.1+k3s2 or nil for latest commit from master
 
-func Test_E2ESecretsEncryption(t *testing.T) {
+func Test_E2ESecretsEncryptionOld(t *testing.T) {
 	RegisterFailHandler(Fail)
 	flag.Parse()
 	suiteConfig, reporterConfig := GinkgoConfiguration()

tests/e2e/snapshotrestore/Vagrantfile

@@ -82,6 +82,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS

tests/e2e/splitserver/Vagrantfile

@@ -1,14 +1,15 @@
 ENV['VAGRANT_NO_PARALLEL'] = 'no'
 NODE_ROLES = (ENV['E2E_NODE_ROLES'] ||
-  ["server-etcd-0", "server-etcd-1", "server-etcd-2", "server-cp-0", "server-cp-1", "agent-0"])
+  ["server-etcd-0", "server-etcd-1", "server-etcd-2", "server-cp-0", "server-cp-1", "agent-0", "agent-1"])
 NODE_BOXES = (ENV['E2E_NODE_BOXES'] ||
   ['generic/ubuntu2310', 'generic/ubuntu2310', 'generic/ubuntu2310', 'generic/ubuntu2310', 'generic/ubuntu2310'])
 GITHUB_BRANCH = (ENV['E2E_GITHUB_BRANCH'] || "master")
 RELEASE_VERSION = (ENV['E2E_RELEASE_VERSION'] || "")
+GOCOVER = (ENV['E2E_GOCOVER'] || "")
 NODE_CPUS = (ENV['E2E_NODE_CPUS'] || 2).to_i
 NODE_MEMORY = (ENV['E2E_NODE_MEMORY'] || 2048).to_i
 # Virtualbox >= 6.1.28 require `/etc/vbox/network.conf` for expanded private networks 
-NETWORK_PREFIX = "10.10.10"
+NETWORK_PREFIX = "10.10.12"
 
 def provision(vm, role, role_num, node_num)
   vm.box = NODE_BOXES[node_num]
@@ -22,12 +23,14 @@ def provision(vm, role, role_num, node_num)
   defaultOSConfigure(vm)
 
   install_type = getInstallType(vm, RELEASE_VERSION, GITHUB_BRANCH)
+  addCoverageDir(vm, role, GOCOVER)
 
   vm.provision "ping k3s.io", type: "shell", inline: "ping -c 2 k3s.io"
-
   if node_num == 0 && !role.include?("server") && !role.include?("etcd")
-    puts "first node must be a etcd server"
-    abort
+    if ARGV.include?("up") || (ARGV.include?("reload") && ARGV.include?("--provision"))
+      puts "Error: first node provisioned must be a etcd server"
+      abort
+    end
   elsif role.include?("server") && role.include?("etcd") && role_num == 0
     vm.provision 'k3s-install', type: 'k3s', run: 'once' do |k3s|
       k3s.args = "server"
@@ -102,6 +105,8 @@ Vagrant.configure("2") do |config|
   config.vm.provider "libvirt" do |v|
     v.cpus = NODE_CPUS
     v.memory = NODE_MEMORY
+    # We replicate the default prefix, but add a timestamp to enable parallel runs and cleanup of old VMs
+    v.default_prefix = File.basename(Dir.getwd) + "_" + Time.now.to_i.to_s + "_"
   end
   config.vm.provider "virtualbox" do |v|
     v.cpus = NODE_CPUS
@@ -116,6 +121,7 @@ Vagrant.configure("2") do |config|
   end
 
   NODE_ROLES.each_with_index do |role, i|
+    # Find the first number in the role name
     role_num = role.split("-", -1).pop.to_i
     config.vm.define role do |node|
       provision(node.vm, role, role_num, i)

tests/e2e/splitserver/splitserver_test.go

@@ -1,16 +1,19 @@
 package splitserver
 
 import (
+	"context"
 	"flag"
 	"fmt"
 	"os"
 	"strconv"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/k3s-io/k3s/tests/e2e"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
+	"golang.org/x/sync/errgroup"
 )
 
 // Valid nodeOS: generic/ubuntu2310, opensuse/Leap-15.3.x86_64
@@ -19,21 +22,25 @@ var etcdCount = flag.Int("etcdCount", 3, "number of server nodes only deploying
 var controlPlaneCount = flag.Int("controlPlaneCount", 1, "number of server nodes acting as control plane")
 var agentCount = flag.Int("agentCount", 1, "number of agent nodes")
 var ci = flag.Bool("ci", false, "running on CI")
+var local = flag.Bool("local", false, "deploy a locally built K3s binary")
 var hardened = flag.Bool("hardened", false, "true or false")
 
 // Environment Variables Info:
 // E2E_RELEASE_VERSION=v1.23.1+k3s2 or nil for latest commit from master
 
-func createSplitCluster(nodeOS string, etcdCount, controlPlaneCount, agentCount int) ([]string, []string, []string, error) {
+// createSplitCluster creates a split server cluster with the given nodeOS, etcdCount, controlPlaneCount, and agentCount.
+// It duplicates and merges functionality found in the e2e.CreateCluster and e2e.CreateLocalCluster functions.
+func createSplitCluster(nodeOS string, etcdCount, controlPlaneCount, agentCount int, local bool) ([]string, []string, []string, error) {
 	etcdNodeNames := make([]string, etcdCount)
+	cpNodeNames := make([]string, controlPlaneCount)
+	agentNodeNames := make([]string, agentCount)
+
 	for i := 0; i < etcdCount; i++ {
 		etcdNodeNames[i] = "server-etcd-" + strconv.Itoa(i)
 	}
-	cpNodeNames := make([]string, controlPlaneCount)
 	for i := 0; i < controlPlaneCount; i++ {
 		cpNodeNames[i] = "server-cp-" + strconv.Itoa(i)
 	}
-	agentNodeNames := make([]string, agentCount)
 	for i := 0; i < agentCount; i++ {
 		agentNodeNames[i] = "agent-" + strconv.Itoa(i)
 	}
@@ -43,21 +50,68 @@ func createSplitCluster(nodeOS string, etcdCount, controlPlaneCount, agentCount
 	nodeBoxes := strings.Repeat(nodeOS+" ", etcdCount+controlPlaneCount+agentCount)
 	nodeBoxes = strings.TrimSpace(nodeBoxes)
 
+	allNodes := append(etcdNodeNames, cpNodeNames...)
+	allNodes = append(allNodes, agentNodeNames...)
+
 	var testOptions string
 	for _, env := range os.Environ() {
 		if strings.HasPrefix(env, "E2E_") {
 			testOptions += " " + env
 		}
 	}
 
-	cmd := fmt.Sprintf(`E2E_NODE_ROLES="%s" E2E_NODE_BOXES="%s" %s vagrant up &> vagrant.log`, nodeRoles, nodeBoxes, testOptions)
+	// Provision the first etcd node. In GitHub Actions, this also imports the VM image into libvirt, which
+	// takes time and can cause the next vagrant up to fail if it is not given enough time to complete.
+	cmd := fmt.Sprintf(`E2E_NODE_ROLES="%s" E2E_NODE_BOXES="%s" vagrant up --no-provision %s &> vagrant.log`, nodeRoles, nodeBoxes, etcdNodeNames[0])
 	fmt.Println(cmd)
 	if _, err := e2e.RunCommand(cmd); err != nil {
-		fmt.Println("Error Creating Cluster", err)
-		return nil, nil, nil, err
+		return etcdNodeNames, cpNodeNames, agentNodeNames, err
+	}
+
+	// Bring up the rest of the nodes in parallel
+	errg, _ := errgroup.WithContext(context.Background())
+	for _, node := range allNodes[1:] {
+		cmd := fmt.Sprintf(`E2E_NODE_ROLES="%s" E2E_NODE_BOXES="%s" vagrant up --no-provision %s &>> vagrant.log`, nodeRoles, nodeBoxes, node)
+		errg.Go(func() error {
+			_, err := e2e.RunCommand(cmd)
+			return err
+		})
+		// libVirt/Virtualbox needs some time between provisioning nodes
+		time.Sleep(10 * time.Second)
+	}
+	if err := errg.Wait(); err != nil {
+		return etcdNodeNames, cpNodeNames, agentNodeNames, err
+	}
+
+	if local {
+		testOptions += " E2E_RELEASE_VERSION=skip"
+		for _, node := range allNodes {
+			cmd := fmt.Sprintf(`E2E_NODE_ROLES=%s vagrant scp ../../../dist/artifacts/k3s  %s:/tmp/`, node, node)
+			if _, err := e2e.RunCommand(cmd); err != nil {
+				return etcdNodeNames, cpNodeNames, agentNodeNames, fmt.Errorf("failed to scp k3s binary to %s: %v", node, err)
+			}
+			if _, err := e2e.RunCmdOnNode("mv /tmp/k3s /usr/local/bin/", node); err != nil {
+				return etcdNodeNames, cpNodeNames, agentNodeNames, err
+			}
+		}
+	}
+	// Install K3s on all nodes in parallel
+	errg, _ = errgroup.WithContext(context.Background())
+	for _, node := range allNodes {
+		cmd = fmt.Sprintf(`E2E_NODE_ROLES="%s" E2E_NODE_BOXES="%s" %s vagrant provision %s &>> vagrant.log`, nodeRoles, nodeBoxes, testOptions, node)
+		errg.Go(func() error {
+			_, err := e2e.RunCommand(cmd)
+			return err
+		})
+		// K3s needs some time between joining nodes to avoid learner issues
+		time.Sleep(10 * time.Second)
+	}
+	if err := errg.Wait(); err != nil {
+		return etcdNodeNames, cpNodeNames, agentNodeNames, err
 	}
 	return etcdNodeNames, cpNodeNames, agentNodeNames, nil
 }
+
 func Test_E2ESplitServer(t *testing.T) {
 	RegisterFailHandler(Fail)
 	flag.Parse()
@@ -78,7 +132,7 @@ var _ = Describe("Verify Create", Ordered, func() {
 	Context("Cluster :", func() {
 		It("Starts up with no issues", func() {
 			var err error
-			etcdNodeNames, cpNodeNames, agentNodeNames, err = createSplitCluster(*nodeOS, *etcdCount, *controlPlaneCount, *agentCount)
+			etcdNodeNames, cpNodeNames, agentNodeNames, err = createSplitCluster(*nodeOS, *etcdCount, *controlPlaneCount, *agentCount, *local)
 			Expect(err).NotTo(HaveOccurred(), e2e.GetVagrantLog(err))
 			fmt.Println("CLUSTER CONFIG")
 			fmt.Println("OS:", *nodeOS)
@@ -229,6 +283,11 @@ var _ = AfterEach(func() {
 })
 
 var _ = AfterSuite(func() {
+	if !failed {
+		allNodes := append(cpNodeNames, etcdNodeNames...)
+		allNodes = append(allNodes, agentNodeNames...)
+		Expect(e2e.GetCoverageReport(allNodes)).To(Succeed())
+	}
 	if !failed || *ci {
 		Expect(e2e.DestroyCluster()).To(Succeed())
 		Expect(os.Remove(kubeConfigFile)).To(Succeed())