Freshen up dry run mode for the installer/controller

The installer/controller dry run mode has not been kept up to date with
recent changes. This commit updates the dry run mode to skip actions
that would modify the system, such as writing to disk or starting
services when in dry run mode.

This will avoid damaging the host system when testing the installer in
dry run mode (used by https://github.com/openshift-assisted/assisted-swarm)

Author: omertuc

src/assisted_installer_controller/assisted_installer_controller.go

@@ -94,10 +94,10 @@ type ControllerConfig struct {
 	ControlPlaneCount       int    `envconfig:"CONTROL_PLANE_COUNT" required:"false" default:"3"`
 	WaitForClusterVersion   bool   `envconfig:"CHECK_CLUSTER_VERSION" required:"false" default:"false"`
 	MustGatherImage         string `envconfig:"MUST_GATHER_IMAGE" required:"false" default:""`
+	NotifyNumReboots        bool   `envconfig:"NOTIFY_NUM_REBOOTS" default:"false"`
 	DryRunEnabled           bool   `envconfig:"DRY_ENABLE" required:"false" default:"false"`
 	DryFakeRebootMarkerPath string `envconfig:"DRY_FAKE_REBOOT_MARKER_PATH" required:"false" default:""`
 	DryRunClusterHostsPath  string `envconfig:"DRY_CLUSTER_HOSTS_PATH"`
-	NotifyNumReboots        bool   `envconfig:"NOTIFY_NUM_REBOOTS" default:"false"`
 	// DryRunClusterHostsPath gets read parsed into ParsedClusterHosts by DryParseClusterHosts
 	ParsedClusterHosts config.DryClusterHosts
 }

src/installer/installer.go

@@ -220,6 +220,12 @@ func (i *installer) waitForWorkers(ctx context.Context) error {
 		i.log.Error(err)
 		return err
 	}
+
+	if i.DryRunEnabled {
+		// In dry run mode we assume we're not ABI
+		return nil
+	}
+
 	if invoker := common.GetInvoker(kc, i.log); invoker != common.InvokerAgent {
 		return nil
 	}
@@ -399,6 +405,10 @@ func (i *installer) writeImageToDisk(ignitionPath string) error {
 }
 
 func (i *installer) skipMcoReboot(ignitionPath string) {
+	if i.DryRunEnabled {
+		return
+	}
+
 	var mc *mcfgv1.MachineConfig
 	mc, err := i.getEncapsulatedMC(ignitionPath)
 	if err != nil {
@@ -447,9 +457,11 @@ func (i *installer) startBootstrap() error {
 	}
 
 	// reload systemd configurations from filesystem and regenerate dependency trees
-	err = i.ops.SystemctlAction("daemon-reload")
-	if err != nil {
-		return err
+	if !i.DryRunEnabled {
+		err = i.ops.SystemctlAction("daemon-reload")
+		if err != nil {
+			return err
+		}
 	}
 
 	/* in case hostname is illegal (localhost, contains capital letters, etc.), we need to set hostname to
@@ -466,7 +478,7 @@ func (i *installer) startBootstrap() error {
 	// restart NetworkManager to trigger NetworkManager/dispatcher.d/30-local-dns-prepender
 	// we don't do it on SNO because the "local-dns-prepender" is not even
 	// available on none-platform
-	if i.ControlPlaneCount != 1 {
+	if i.ControlPlaneCount != 1 && !i.DryRunEnabled {
 		err = i.ops.SystemctlAction("restart", "NetworkManager.service")
 		if err != nil {
 			i.log.Error(err)
@@ -482,7 +494,7 @@ func (i *installer) startBootstrap() error {
 	// If we're in a pure RHEL/CentOS environment, we need to overlay the node image
 	// first to have access to e.g. oc, kubelet, cri-o, etc...
 	// https://github.com/openshift/enhancements/pull/1637
-	if !i.ops.FileExists(openshiftClientBin) {
+	if !i.ops.FileExists(openshiftClientBin) && !i.DryRunEnabled {
 		err = i.ops.SystemctlAction("start", "--no-block", nodeImagePullService, nodeImageOverlayService)
 		if err != nil {
 			return err
@@ -504,6 +516,9 @@ func (i *installer) startBootstrap() error {
 	}
 
 	servicesToStart := []string{"bootkube.service", "approve-csr.service", "progress.service"}
+	if i.DryRunEnabled {
+		servicesToStart = []string{}
+	}
 	for _, service := range servicesToStart {
 		err = i.ops.SystemctlAction("start", service)
 		if err != nil {
@@ -614,9 +629,12 @@ func (i *installer) waitForControlPlane(ctx context.Context) error {
 	}
 
 	i.waitForBootkube(ctx)
-	if err = i.waitForETCDBootstrap(ctx); err != nil {
-		i.log.Error(err)
-		return err
+
+	if !i.DryRunEnabled {
+		if err = i.waitForETCDBootstrap(ctx); err != nil {
+			i.log.Error(err)
+			return err
+		}
 	}
 
 	// waiting for controller pod to be running

src/main/assisted-installer-controller/assisted_installer_main.go

@@ -98,7 +98,8 @@ func main() {
 		mockController := gomock.NewController(logger)
 		kc = k8s_client.NewMockK8SClient(mockController)
 		mock, _ := kc.(*k8s_client.MockK8SClient)
-		drymock.PrepareControllerDryMock(mock, logger, o, Options.ControllerConfig.ParsedClusterHosts)
+		drymock.PrepareControllerDryMock(mock, logger, o, Options.ControllerConfig.ParsedClusterHosts,
+			Options.ControllerConfig.ControlPlaneCount)
 	}
 
 	err = kc.SetProxyEnvVars()

src/main/drymock/dry_mode_k8s_mock.go

@@ -73,7 +73,11 @@ Dry`)
 // PrepareControllerDryMock utilizes k8s_client.MockK8SClient to fake the k8s API to make the
 // controller think it's running on an actual cluster, just enough to make it pass an installation.
 // Used in dry mode.
-func PrepareControllerDryMock(mockk8sclient *k8s_client.MockK8SClient, logger *logrus.Logger, o ops.Ops, clusterHosts config.DryClusterHosts) {
+func PrepareControllerDryMock(mockk8sclient *k8s_client.MockK8SClient, logger *logrus.Logger, o ops.Ops, clusterHosts config.DryClusterHosts, controlPlaneCount int) {
+	mockk8sclient.EXPECT().GetControlPlaneReplicas().Return(controlPlaneCount, nil).AnyTimes()
+	// Dry run doesn't support arbiter for now
+	mockk8sclient.EXPECT().GetArbiterReplicas().Return(0, nil).AnyTimes()
+
 	// Called by main
 	mockk8sclient.EXPECT().SetProxyEnvVars().Return(nil).AnyTimes()
 
@@ -107,7 +111,7 @@ func PrepareControllerDryMock(mockk8sclient *k8s_client.MockK8SClient, logger *l
 				// Host didn't even reboot yet, don't pretend it fetched the ignition
 				continue
 			}
-			mcsLogs += fmt.Sprintf("%s.(Ignition)\n", clusterHost.Ip)
+			mcsLogs += fmt.Sprintf("%s:12345 User-Agent:\"Ignition\"\n", clusterHost.Ip)
 		}
 		return mcsLogs, nil
 	}).AnyTimes()
@@ -280,7 +284,11 @@ dEFgad6P3hMZTOg7yVkMOd3QtgVQ9I8dXqS2nG9EMEh97WIhi6f5ztvcQvQ5tXjh
 // PrepareInstallerDryK8sMock utilizes k8s_client.MockK8SClient to fake the k8s API to make the
 // installer think it's talking with an actual cluster, just enough to make it pass an installation.
 // Used in dry mode.
-func PrepareInstallerDryK8sMock(mockk8sclient *k8s_client.MockK8SClient, logger logrus.FieldLogger, o ops.Ops, clusterHosts config.DryClusterHosts) {
+func PrepareInstallerDryK8sMock(mockk8sclient *k8s_client.MockK8SClient, logger logrus.FieldLogger, o ops.Ops, clusterHosts config.DryClusterHosts,
+	controlPlaneCount int) {
+	mockk8sclient.EXPECT().GetControlPlaneReplicas().Return(controlPlaneCount, nil).AnyTimes()
+	// Dry run doesn't support arbiter for now
+	mockk8sclient.EXPECT().GetArbiterReplicas().Return(0, nil).AnyTimes()
 	// The installer compares AI host objects to cluster Node objects (either by name or by IP) to check which AI hosts are already
 	// joined as nodes. This fakes the node list so that check will pass
 	mockk8sclient.EXPECT().ListNodesByRole(gomock.Any()).DoAndReturn(func() (*v1.NodeList, error) {
@@ -307,7 +315,8 @@ func NewDryRunK8SClientBuilder(installerConfig *config.Config, ops ops.Ops) func
 		mockController := gomock.NewController(ginkgo.GinkgoT())
 		kc = k8s_client.NewMockK8SClient(mockController)
 		mock, _ := kc.(*k8s_client.MockK8SClient)
-		PrepareInstallerDryK8sMock(mock, logger, ops, installerConfig.ParsedClusterHosts)
+		PrepareInstallerDryK8sMock(mock, logger, ops, installerConfig.ParsedClusterHosts,
+			installerConfig.ControlPlaneCount)
 		return kc, nil
 	}
 }

src/ops/ops.go

@@ -268,6 +268,14 @@ func (o *ops) configureFirstBootIgnition(liveLogger io.Writer, ignitionPath stri
 }
 
 func (o *ops) WriteImageToExistingRoot(liveLogger io.Writer, ignitionPath string, installerArgs []string) error {
+	if o.installerConfig.DryRunEnabled {
+		// In dry run, we use an executable called dry-installer rather than ostree.
+		// This executable is expected to pretend to be doing coreos-installer stuff and print fake
+		// progress. It's up to the dry-mode user to make sure such executable is available in PATH
+		_, err := o.ExecPrivilegeCommand(liveLogger, dryRunCoreosInstallerExecutable)
+		return err
+	}
+
 	if err := o.remountFilesystems(liveLogger); err != nil {
 		return err
 	}