[WIP] AGENT-850: day2 agent-import-cluster.service

cmd/openshift-install/agent.go

@@ -112,7 +112,19 @@ var (
 		},
 	}
 
-	agentTargets = []target{agentConfigTarget, agentManifestsTarget, agentImageTarget, agentPXEFilesTarget, agentConfigImageTarget, agentUnconfiguredIgnitionTarget}
+	agentImageAddNodesTarget = target{
+		name: "Agent ISO Image for adding nodes",
+		command: &cobra.Command{
+			Use:   "nodes-image",
+			Short: "Generates a bootable image containing all the information needed to added one or more nodes to an already existing cluster",
+			Args:  cobra.ExactArgs(0),
+		},
+		assets: []asset.WritableAsset{
+			&image.AgentImage{},
+		},
+	}
+
+	agentTargets = []target{agentConfigTarget, agentManifestsTarget, agentImageTarget, agentPXEFilesTarget, agentConfigImageTarget, agentUnconfiguredIgnitionTarget, agentImageAddNodesTarget}
 )
 
 func newAgentCreateCmd() *cobra.Command {

data/data/agent/files/usr/local/bin/get-container-images.sh

@@ -10,6 +10,8 @@ set -euo pipefail
 # The agent image will be also retrieved when its script is run
 cat <<EOF >/usr/local/share/assisted-service/agent-images.env
 SERVICE_IMAGE=$(image_for agent-installer-api-server)
+# TODO: remove when assisted-service changes to import cluster merges
+SERVICE_IMAGE=quay.io/rwsu1/assisted-service:latest
 CONTROLLER_IMAGE=$(image_for agent-installer-csr-approver)
 INSTALLER_IMAGE=$(image_for agent-installer-orchestrator)
 AGENT_DOCKER_IMAGE=$(image_for agent-installer-node-agent)

data/data/agent/files/usr/local/bin/import-cluster.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+set -e
+
+echo "Waiting for assisted-service to be ready"
+until curl --output /dev/null --silent --fail "${SERVICE_BASE_URL}/api/assisted-install/v2/infra-envs"; do
+    printf '.'
+    sleep 5
+done
+
+echo "Importing cluster from ${API_VIP_DNSNAME}"
+curl -d '{"name":"abi", "api_vip_dnsname":"'"${API_VIP_DNSNAME}"'", "openshift_cluster_id":"764eb48f-8ffb-4126-b7ed-0ca746365654"}' -H "Content-Type: application/json" -X POST "${SERVICE_BASE_URL}/api/assisted-install/v2/clusters/import"

data/data/agent/files/usr/local/bin/start-cluster-installation.sh

@@ -48,33 +48,45 @@ num_known_hosts() {
 
 while [[ "${total_required_nodes}" != $(num_known_hosts) ]]
 do
+
+    # For some reason the initial role patching doesn't seem to work properly, to be reviewed
+    echo "Patching host..."
+    export HOST_ID=$(curl -s ${BASE_URL}/infra-envs/${INFRA_ENV_ID}/hosts | jq -r '.[].id')
+    curl -X PATCH -d '{"host_role":"worker"}' -H "Content-Type: application/json" ${BASE_URL}/infra-envs/${INFRA_ENV_ID}/hosts/${HOST_ID}
+
     echo "Waiting for hosts to become ready for cluster installation..." 1>&2
     sleep 10
 done
 
 echo "All ${total_required_nodes} hosts are ready." 1>&2
 clear_issue "${status_issue}"
 
-while [[ "${cluster_status}" != "installed" ]]
-do
-    sleep 5
-    cluster_status=$(curl -s -S "${BASE_URL}/clusters" | jq -r .[].status)
-    echo "Cluster status: ${cluster_status}" 1>&2
-    # Start the cluster install, if it transitions back to Ready due to a failure,
-    # then it will be restarted
-    case "${cluster_status}" in
-        "ready")
-            echo "Starting cluster installation..." 1>&2
-            curl -s -S -X POST "${BASE_URL}/clusters/${cluster_id}/actions/install" \
-                -H 'accept: application/json' \
-                -d ''
-            echo "Cluster installation started" 1>&2
-            ;&
-        "installed" | "preparing-for-installation" | "installing")
-            printf '\\e{lightgreen}Cluster installation in progress\\e{reset}' | set_issue "${status_issue}"
-            ;;
-        *)
-            printf '\\e{lightred}Installation cannot proceed:\\e{reset} Cluster status: %s' "${cluster_status}" | set_issue "${status_issue}"
-            ;;
-    esac
-done
+# Install the host
+# echo "Installing host..."
+# curl -X POST ${BASE_URL}/infra-envs/${INFRA_ENV_ID}/hosts/${HOST_ID}/actions/install
+# TODO: temporary to test cluster import without initiating install
+echo "RWSU skip Installing host..."
+
+# while [[ "${cluster_status}" != "installed" ]]
+# do
+#     sleep 5
+#     cluster_status=$(curl -s -S "${BASE_URL}/clusters" | jq -r .[].status)
+#     echo "Cluster status: ${cluster_status}" 1>&2
+#     # Start the cluster install, if it transitions back to Ready due to a failure,
+#     # then it will be restarted
+#     case "${cluster_status}" in
+#         "ready")
+#             echo "Starting cluster installation..." 1>&2
+#             curl -s -S -X POST "${BASE_URL}/clusters/${cluster_id}/actions/install" \
+#                 -H 'accept: application/json' \
+#                 -d ''
+#             echo "Cluster installation started" 1>&2
+#             ;&
+#         "installed" | "preparing-for-installation" | "installing")
+#             printf '\\e{lightgreen}Cluster installation in progress\\e{reset}' | set_issue "${status_issue}"
+#             ;;
+#         *)
+#             printf '\\e{lightred}Installation cannot proceed:\\e{reset} Cluster status: %s' "${cluster_status}" | set_issue "${status_issue}"
+#             ;;
+#     esac
+# done

data/data/agent/files/usr/local/share/assisted-service/images.env.template

@@ -1,3 +1,4 @@
 ASSISTED_SERVICE_HOST=localhost:8090
 ASSISTED_SERVICE_SCHEME={{.ServiceProtocol}}
-OS_IMAGES=[{"openshift_version":"{{.OSImage.OpenshiftVersion}}","cpu_architecture":"{{.OSImage.CPUArchitecture}}","url":"{{.OSImage.URL}}","version":"{{.OSImage.Version}}"}]
+# TODO: temporary to test on specific version
+OS_IMAGES=[{"openshift_version":"{{.OSImage.OpenshiftVersion}}","cpu_architecture":"{{.OSImage.CPUArchitecture}}","url":"{{.OSImage.URL}}","version":"{{.OSImage.Version}}"},{"openshift_version":"4.15.0-rc.2","cpu_architecture":"x86_64","url":"https://rhcos.mirror.openshift.com/art/storage/prod/streams/4.14-9.2/builds/414.92.202307070025-0/x86_64/rhcos-414.92.202307070025-0-live.x86_64.iso","version":"414.92.202307070025-0"}]

data/data/agent/files/usr/local/share/import-cluster/import-cluster.env

@@ -0,0 +1,3 @@
+CLUSTER_ID=d8fc9f93-1b33-4b2a-943f-54afca271cb2
+CLUSTER_NAME=abi
+CLUSTER_API_VIP_DNS_NAME=api.ostest.test.metalkube.org

data/data/agent/files/usr/local/share/start-cluster/start-cluster.env.template

@@ -1,2 +1,2 @@
 REQUIRED_MASTER_NODES={{.ControlPlaneAgents}}
-REQUIRED_WORKER_NODES={{.WorkerAgents}}
+REQUIRED_WORKER_NODES=1

data/data/agent/systemd/units/agent-import-cluster.service.template

@@ -0,0 +1,27 @@
+[Unit]
+Description=Service that imports the cluster
+Wants=network-online.target assisted-service.service
+PartOf=assisted-service-pod.service
+After=network-online.target assisted-service.service
+ConditionPathExists=/etc/assisted/node0
+
+[Service]
+Environment=PODMAN_SYSTEMD_UNIT=%n
+Environment=OPENSHIFT_INSTALL_RELEASE_IMAGE_MIRROR={{.ReleaseImageMirror}}
+EnvironmentFile=/etc/assisted/rendezvous-host.env
+EnvironmentFile=/usr/local/share/assisted-service/agent-images.env
+EnvironmentFile=/usr/local/share/import-cluster/import-cluster.env
+ExecStartPre=/usr/local/bin/wait-for-assisted-service.sh
+ExecStartPre=/bin/rm -f %t/%n.ctr-id
+ExecStart=podman run --net host --cidfile=%t/%n.ctr-id --cgroups=no-conmon --log-driver=journald --rm --pod-id-file=%t/assisted-service-pod.pod-id --replace --name=agent-import-cluster -v /etc/assisted/manifests:/manifests -v /etc/assisted/extra-manifests:/extra-manifests -v /etc/pki/ca-trust:/etc/pki/ca-trust:z {{ if .HaveMirrorConfig }}-v /etc/containers:/etc/containers{{ end }} --env SERVICE_BASE_URL --env CLUSTER_ID --env CLUSTER_NAME --env CLUSTER_API_VIP_DNS_NAME $SERVICE_IMAGE /usr/local/bin/agent-installer-client importCluster
+ExecStop=/usr/bin/podman stop --ignore --cidfile=%t/%n.ctr-id
+ExecStopPost=/usr/bin/podman rm -f --ignore --cidfile=%t/%n.ctr-id
+
+KillMode=none
+Type=oneshot
+Restart=on-failure
+RestartSec=30
+RemainAfterExit=true
+
+[Install]
+WantedBy=multi-user.target

data/data/agent/systemd/units/agent-register-infraenv.service.template

@@ -2,7 +2,7 @@
 Description=Service that registers the infraenv
 Wants=network-online.target assisted-service.service
 PartOf=assisted-service-pod.service
-After=network-online.target assisted-service.service agent-register-cluster.service
+After=network-online.target assisted-service.service {{ if eq .WorkflowType "install" }}agent-register-cluster.service{{ end }}{{ if eq .WorkflowType "addnodes" }}agent-import-cluster.service{{ end }}
 ConditionPathExists=/etc/assisted/node0
 
 [Service]

data/data/agent/systemd/units/apply-host-config.service → data/data/agent/systemd/units/apply-host-config.service.template

@@ -11,10 +11,11 @@ Environment=PODMAN_SYSTEMD_UNIT=%n
 EnvironmentFile=/etc/assisted/rendezvous-host.env
 EnvironmentFile=/usr/local/share/assisted-service/agent-images.env
 EnvironmentFile=/usr/local/share/assisted-service/assisted-service.env
+Environment=WORKFLOW_TYPE={{.WorkflowType}}
 ExecStartPre=/bin/rm -f %t/%n.ctr-id
 ExecStartPre=/bin/mkdir -p %t/agent-installer /etc/assisted/hostconfig
 ExecStartPre=/usr/local/bin/wait-for-assisted-service.sh
-ExecStart=podman run --net host --cidfile=%t/%n.ctr-id --cgroups=no-conmon --log-driver=journald --restart=on-failure:10 --pod-id-file=%t/assisted-service-pod.pod-id --replace --name=apply-host-config -v /etc/assisted/hostconfig:/etc/assisted/hostconfig -v %t/agent-installer:/var/run/agent-installer:z --env SERVICE_BASE_URL --env INFRA_ENV_ID $SERVICE_IMAGE /usr/local/bin/agent-installer-client configure
+ExecStart=podman run --net host --cidfile=%t/%n.ctr-id --cgroups=no-conmon --log-driver=journald --restart=on-failure:10 --pod-id-file=%t/assisted-service-pod.pod-id --replace --name=apply-host-config -v /etc/assisted/hostconfig:/etc/assisted/hostconfig -v %t/agent-installer:/var/run/agent-installer:z --env SERVICE_BASE_URL --env INFRA_ENV_ID --env WORKFLOW_TYPE $SERVICE_IMAGE /usr/local/bin/agent-installer-client configure
 ExecStop=/usr/bin/podman stop --ignore --cidfile=%t/%n.ctr-id
 ExecStopPost=/usr/bin/podman rm -f --ignore --cidfile=%t/%n.ctr-id
 

pkg/asset/agent/image/agentimage.go

@@ -32,6 +32,7 @@ type AgentImage struct {
 	rootFSURL            string
 	bootArtifactsBaseURL string
 	platform             hiveext.PlatformType
+	kubeconfig           *manifests.KubeConfigFile
 }
 
 var _ asset.WritableAsset = (*AgentImage)(nil)
@@ -42,6 +43,7 @@ func (a *AgentImage) Dependencies() []asset.Asset {
 		&AgentArtifacts{},
 		&manifests.AgentManifests{},
 		&BaseIso{},
+		&manifests.KubeConfigFile{},
 	}
 }
 
@@ -50,21 +52,24 @@ func (a *AgentImage) Generate(dependencies asset.Parents) error {
 	agentArtifacts := &AgentArtifacts{}
 	agentManifests := &manifests.AgentManifests{}
 	baseIso := &BaseIso{}
-	dependencies.Get(agentArtifacts, agentManifests, baseIso)
+	kubeconfig := &manifests.KubeConfigFile{}
+	dependencies.Get(agentArtifacts, agentManifests, baseIso, kubeconfig)
 
 	a.cpuArch = agentArtifacts.CPUArch
 	a.rendezvousIP = agentArtifacts.RendezvousIP
 	a.tmpPath = agentArtifacts.TmpPath
 	a.isoPath = agentArtifacts.ISOPath
 	a.bootArtifactsBaseURL = agentArtifacts.BootArtifactsBaseURL
+	a.kubeconfig = kubeconfig
 
 	volumeID, err := isoeditor.VolumeIdentifier(a.isoPath)
 	if err != nil {
 		return err
 	}
 	a.volumeID = volumeID
 
-	a.platform = agentManifests.AgentClusterInstall.Spec.PlatformType
+	// a.platform = agentManifests.AgentClusterInstall.Spec.PlatformType
+	a.platform = "BareMetal"
 	if a.platform == hiveext.ExternalPlatformType {
 		// when the bootArtifactsBaseURL is specified, construct the custom rootfs URL
 		if a.bootArtifactsBaseURL != "" {
@@ -214,7 +219,11 @@ func (a *AgentImage) PersistToFile(directory string) error {
 		return errors.New("cannot generate ISO image due to configuration errors")
 	}
 
-	agentIsoFile := filepath.Join(directory, fmt.Sprintf(agentISOFilename, a.cpuArch))
+	name := fmt.Sprintf(agentISOFilename, a.cpuArch)
+	if a.kubeconfig != nil {
+		name = "workers-" + name
+	}
+	agentIsoFile := filepath.Join(directory, name)
 
 	// Remove symlink if it exists
 	os.Remove(agentIsoFile)