test: move driver deployment into e2e.test

The immediate advantage is that we can get rid of the logic in the Jenkinsfile which tries to determine whether the running cluster can be reused. It also becomes shorter. Retries get removed because we should get to a point where tests are so reliable that we don't need them. The long-term advantage is that we can add tests for various kinds of deploying PMEM-CSI more easily.
pohly · Mar 19, 2020 · df17bda · df17bda
1 parent 0eeb2e2
commit df17bda
Show file tree

Hide file tree

Showing 44 changed files with 1,159 additions and 280 deletions.
diff --git a/Jenkinsfile b/Jenkinsfile
@@ -226,7 +226,6 @@ pipeline {
         stage('Build test image') {
             options {
                 timeout(time: 60, unit: "MINUTES")
-                retry(2)
             }
 
             steps {
@@ -244,104 +243,41 @@ pipeline {
             }
         }
 
-        stage('testing 1.16 LVM') {
+        stage('testing 1.16') {
             options {
-                timeout(time: 90, unit: "MINUTES")
-                retry(2)
+                timeout(time: 120, unit: "MINUTES")
             }
             steps {
-                TestInVM("lvm", "testing", "fedora", "", "1.16")
+                TestInVM("fedora", "", "1.16")
             }
         }
 
-        stage('testing 1.16 direct') {
-            options {
-                timeout(time: 180, unit: "MINUTES")
-                retry(2)
-            }
-            steps {
-                TestInVM("direct", "testing", "fedora", "", "1.16")
-            }
-        }
-
-        stage('testing 1.14 LVM') {
-            when { not { changeRequest() } }
-            options {
-                timeout(time: 90, unit: "MINUTES")
-                retry(2)
-            }
-            steps {
-                TestInVM("lvm", "testing", "fedora", "", "1.14")
-            }
-        }
-
-        stage('testing 1.14 direct') {
-            when { not { changeRequest() } }
-            options {
-                timeout(time: 180, unit: "MINUTES")
-                retry(2)
-            }
-            steps {
-                TestInVM("direct", "testing", "fedora", "", "1.14")
-            }
-        }
-
-        /*
-          In production we can only run E2E testing, no sanity testing.
-          Therefore it is faster.
-        */
-
-        stage('production 1.14 LVM') {
-            when { not { changeRequest() } }
-            options {
-                timeout(time: 45, unit: "MINUTES")
-                retry(2)
-            }
-            steps {
-                TestInVM("lvm", "production", "fedora", "", "1.14")
-            }
-        }
-
-        stage('production 1.14 direct') {
-            when { not { changeRequest() } }
-            options {
-                timeout(time: 45, unit: "MINUTES")
-                retry(2)
-            }
-            steps {
-                TestInVM("direct", "production", "fedora", "", "1.14")
-            }
-        }
-
-        stage('production 1.15 LVM') {
+        stage('testing 1.15') {
             when { not { changeRequest() } }
             options {
-                timeout(time: 45, unit: "MINUTES")
-                retry(2)
+                timeout(time: 120, unit: "MINUTES")
             }
             steps {
-                TestInVM("lvm", "production", "fedora", "", "1.15")
+                TestInVM("fedora", "", "1.15")
             }
         }
 
-        stage('production 1.15 direct') {
+        stage('testing 1.14') {
             when { not { changeRequest() } }
             options {
-                timeout(time: 45, unit: "MINUTES")
-                retry(2)
+                timeout(time: 120, unit: "MINUTES")
             }
             steps {
-                TestInVM("direct", "production", "fedora", "", "1.15")
+                TestInVM("fedora", "", "1.14")
             }
         }
 
-        stage('production 1.17, Clear Linux') {
+        stage('Clear Linux') {
             options {
-                timeout(time: 90, unit: "MINUTES")
-                retry(2)
+                timeout(time: 120, unit: "MINUTES")
             }
             steps {
-                TestInVM("lvm", "production", "clear", "${env.CLEAR_LINUX_VERSION_1_17}", "")
+                TestInVM("clear", "${env.CLEAR_LINUX_VERSION_1_17}", "")
             }
         }
 
@@ -443,7 +379,7 @@ String RunInBuilder() {
     "
 }
 
-void TestInVM(deviceMode, deploymentMode, distro, distroVersion, kubernetesVersion) {
+void TestInVM(distro, distroVersion, kubernetesVersion) {
     try {
         /*
         We have to run "make start" in the current directory
@@ -481,8 +417,6 @@ void TestInVM(deviceMode, deploymentMode, distro, distroVersion, kubernetesVersi
            ${RunInBuilder()} \
                   -e CLUSTER=${env.CLUSTER} \
                   -e TEST_LOCAL_REGISTRY=\$(ip addr show dev docker0 | grep ' inet ' | sed -e 's/.* inet //' -e 's;/.*;;'):5000 \
-                  -e TEST_DEVICEMODE=${deviceMode} \
-                  -e TEST_DEPLOYMENTMODE=${deploymentMode} \
                   -e TEST_CHECK_SIGNED_FILES=false \
                   -e TEST_CHECK_KVM=false \
                   -e TEST_QEMU_CPU=host,-vmx \
@@ -495,13 +429,7 @@ void TestInVM(deviceMode, deploymentMode, distro, distroVersion, kubernetesVersi
                            loggers=; \
                            atexit () { set -x; kill \$loggers; kill \$( ps --no-header -o %p ); }; \
                            trap atexit EXIT; \
-                           echo CLUSTER=\$CLUSTER TEST_LOCAL_REGISTRY=\$TEST_LOCAL_REGISTRY TEST_DISTRO=\$TEST_DISTRO TEST_DISTRO_VERSION=\$TEST_DISTRO_VERSION TEST_KUBERNETES_VERSION=\$TEST_KUBERNETES_VERSION >_work/new-cluster-config && \
-                           if [ -e _work/\$CLUSTER/cluster-config ] && ! diff _work/\$CLUSTER/cluster-config _work/new-cluster-config; then \
-                               echo QEMU cluster configuration has changed, need to stop old cluster. && \
-                               make stop; \
-                           fi && \
                            make start && \
-                           mv _work/new-cluster-config _work/\$CLUSTER/cluster-config && \
                            _work/${env.CLUSTER}/ssh.0 kubectl get pods --all-namespaces -o wide && \
                            for pod in ${env.LOGGING_PODS}; do \
                                _work/${env.CLUSTER}/ssh.0 kubectl logs -f -n kube-system \$pod-pmem-csi-${env.CLUSTER}-master | sed -e \"s/^/\$pod: /\" & \
@@ -520,7 +448,7 @@ void TestInVM(deviceMode, deploymentMode, distro, distroVersion, kubernetesVersi
                                  done | sed -e \"s/^/\$hostname: /\" ) & \
                                loggers=\"\$loggers \$!\"; \
                            done && \
-                           testrun=\$(echo '${distro}-${distroVersion}-${kubernetesVersion}-${deviceMode}-${deploymentMode}' | sed -e s/--*/-/g | tr . _ ) && \
+                           testrun=\$(echo '${distro}-${distroVersion}-${kubernetesVersion}' | sed -e s/--*/-/g | tr . _ ) && \
                            make test_e2e TEST_E2E_REPORT_DIR=${WORKSPACE}/build/reports.tmp/\$testrun \
                                          TEST_E2E_SKIP=\$(if [ \"${env.CHANGE_ID}\" ] && [ \"${env.CHANGE_ID}\" != null ]; then echo \\\\[Slow\\\\]; fi) \
                            ' \

diff --git a/README.md b/README.md
@@ -1127,11 +1127,6 @@ Once `make start` completes, the cluster is ready for interactive use via
 set `KUBECONFIG` as shown at the end of the `make start` output
 and use `kubectl` binary on the host running VMs.
 
-When the cluster is already running, `make start` will re-deploy the
-PMEM-CSI driver without recreating the virtual machines. `kubectl
-apply` is used for this, which may limit the kind of changes that can
-be made on-the-fly.
-
 Use `make stop` to stop and remove the virtual machines.
 
 `make restart` can be used to cleanly reboot all virtual
@@ -1149,6 +1144,13 @@ the master node.
 
 `_work/pmem-govm/ssh.1` starts a shell on the first worker node.
 
+### Deploying PMEM-CSI on a test cluster
+
+After `make start`, PMEM-CSI is *not* installed yet. Either install
+manually as [described for a normal
+cluster](#run-pmem-csi-on-kubernetes) or use the
+[setup-deployment.sh](./test/setup-deployment.sh) script.
+
 ### Configuration options
 
 Several aspects of the cluster and build setup can be configured by overriding
@@ -1161,10 +1163,10 @@ Multiple different clusters can be brought up in parallel by changing
 the default `pmem-govm` cluster name via the `CLUSTER` env variable.
 
 For example, this invocation sets up a cluster using the non-default
-direct device mode:
+Fedora distro:
 
 ``` sh
-TEST_DEVICEMODE=direct CLUSTER=clear-govm-direct make start
+TEST_DISTRO=fedora CLUSTER=fedora-govm make start
 ```
 
 ### Running E2E tests

diff --git a/cmd/pmem-csi-driver/main.go b/cmd/pmem-csi-driver/main.go
@@ -11,9 +11,15 @@ import (
 	"flag"
 	"os"
 
+	"k8s.io/klog"
+
 	"github.com/intel/pmem-csi/pkg/pmem-csi-driver"
 )
 
+func init() {
+	klog.InitFlags(nil)
+}
+
 func main() {
 	flag.Parse()
 	os.Exit(pmemcsidriver.Main())

diff --git a/deploy/kubernetes-1.14/direct/pmem-csi.yaml b/deploy/kubernetes-1.14/direct/pmem-csi.yaml
@@ -1,12 +1,16 @@
 apiVersion: v1
 kind: ServiceAccount
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-controller
   namespace: default
 ---
 apiVersion: rbac.authorization.k8s.io/v1
 kind: Role
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-external-provisioner-cfg
   namespace: default
 rules:
@@ -36,6 +40,8 @@ rules:
 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRole
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-external-provisioner-runner
 rules:
 - apiGroups:
@@ -109,6 +115,8 @@ rules:
 apiVersion: rbac.authorization.k8s.io/v1
 kind: RoleBinding
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-csi-provisioner-role-cfg
   namespace: default
 roleRef:
@@ -123,6 +131,8 @@ subjects:
 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRoleBinding
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-csi-provisioner-role
 roleRef:
   apiGroup: rbac.authorization.k8s.io
@@ -136,39 +146,51 @@ subjects:
 apiVersion: v1
 kind: Service
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-controller
+  namespace: default
 spec:
   ports:
   - port: 10000
   selector:
     app: pmem-csi-controller
+    pmem-csi.intel.com/deployment: direct-production
 ---
 apiVersion: v1
 kind: Service
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-metrics
+  namespace: default
 spec:
   ports:
   - port: 10010
   selector:
     app: pmem-csi-controller
+    pmem-csi.intel.com/deployment: direct-production
   type: NodePort
 ---
 apiVersion: apps/v1
 kind: StatefulSet
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-controller
   namespace: default
 spec:
   replicas: 1
   selector:
     matchLabels:
       app: pmem-csi-controller
+      pmem-csi.intel.com/deployment: direct-production
   serviceName: pmem-csi-controller
   template:
     metadata:
       labels:
         app: pmem-csi-controller
+        pmem-csi.intel.com/deployment: direct-production
         pmem-csi.intel.com/webhook: ignore
     spec:
       containers:
@@ -226,15 +248,20 @@ spec:
 apiVersion: apps/v1
 kind: DaemonSet
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi-node
+  namespace: default
 spec:
   selector:
     matchLabels:
       app: pmem-csi-node
+      pmem-csi.intel.com/deployment: direct-production
   template:
     metadata:
       labels:
         app: pmem-csi-node
+        pmem-csi.intel.com/deployment: direct-production
         pmem-csi.intel.com/webhook: ignore
     spec:
       containers:
@@ -334,6 +361,8 @@ spec:
 apiVersion: storage.k8s.io/v1beta1
 kind: CSIDriver
 metadata:
+  labels:
+    pmem-csi.intel.com/deployment: direct-production
   name: pmem-csi.intel.com
 spec:
   attachRequired: false