AGENT-858: Agent day2 ignition services

cmd/node-joiner/main.go

@@ -1,9 +1,15 @@
 package main
 
 import (
+	"fmt"
+	"io"
+	"os"
+
 	"github.com/sirupsen/logrus"
 	"github.com/spf13/cobra"
+	terminal "golang.org/x/term"
 
+	"github.com/openshift/installer/cmd/openshift-install/command"
 	"github.com/openshift/installer/pkg/nodejoiner"
 )
 
@@ -33,14 +39,47 @@ func main() {
 	}
 
 	rootCmd := &cobra.Command{
-		Use: "node-joiner",
+		Use:              "node-joiner",
+		PersistentPreRun: runRootCmd,
 	}
 	rootCmd.PersistentFlags().String("kubeconfig", "", "Path to the kubeconfig file.")
 	rootCmd.PersistentFlags().String("dir", ".", "assets directory")
+	rootCmd.PersistentFlags().String("log-level", "info", "log level (e.g. \"debug | info | warn | error\")")
 
 	rootCmd.AddCommand(nodesAddCmd)
 	rootCmd.AddCommand(nodesMonitorCmd)
 	if err := rootCmd.Execute(); err != nil {
 		logrus.Fatal(err)
 	}
 }
+
+func runRootCmd(cmd *cobra.Command, args []string) {
+	logrus.SetOutput(io.Discard)
+	logrus.SetLevel(logrus.TraceLevel)
+
+	logLevel, err := cmd.Flags().GetString("log-level")
+	if err != nil {
+		logrus.Fatal(err)
+	}
+
+	level, err := logrus.ParseLevel(logLevel)
+	if err != nil {
+		level = logrus.InfoLevel
+	}
+
+	logrus.AddHook(command.NewFileHookWithNewlineTruncate(os.Stderr, level, &logrus.TextFormatter{
+		// Setting ForceColors is necessary because logrus.TextFormatter determines
+		// whether or not to enable colors by looking at the output of the logger.
+		// In this case, the output is io.Discard, which is not a terminal.
+		// Overriding it here allows the same check to be done, but against the
+		// hook's output instead of the logger's output.
+		ForceColors:            terminal.IsTerminal(int(os.Stderr.Fd())),
+		DisableTimestamp:       true,
+		DisableLevelTruncation: true,
+		DisableQuote:           true,
+	}))
+
+	if err != nil {
+		logrus.Fatal(fmt.Errorf("invalid log-level: %w", err))
+	}
+}

data/data/agent/files/usr/local/bin/add-node.sh

@@ -0,0 +1,47 @@
+#!/bin/bash
+set -e
+
+# shellcheck disable=SC1091
+source issue_status.sh
+
+BASE_URL="${SERVICE_BASE_URL}api/assisted-install/v2"
+
+cluster_id=""
+while [[ "${cluster_id}" = "" ]]
+do
+    # Get cluster id
+    cluster_id=$(curl -s -S "${BASE_URL}/clusters" | jq -r .[].id)
+    if [[ "${cluster_id}" = "" ]]; then
+        sleep 2
+    fi
+done
+
+printf '\nInfra env id is %s\n' "${INFRA_ENV_ID}" 1>&2
+
+status_issue="90_add-node"
+
+# Wait for the current host to be ready
+host_ready=false
+while [[ $host_ready == false ]]
+do
+    host_status=$(curl -s -S "${BASE_URL}/infra-envs/${INFRA_ENV_ID}/hosts" | jq -r ".[].status")
+    if [[ "${host_status}" != "known" ]]; then
+        printf '\\e{yellow}Waiting for the host to be ready' | set_issue "${status_issue}"
+        sleep 10
+    else
+        host_ready=true
+    fi
+done
+
+HOST_ID=$(curl -s "${BASE_URL}/infra-envs/${INFRA_ENV_ID}/hosts" | jq -r '.[].id')
+printf '\nHost %s is ready for installation\n' "${HOST_ID}" 1>&2
+clear_issue "${status_issue}"
+
+# Add the current host to the cluster
+res=$(curl -X POST -s -S -w "%{http_code}\\n" -o /dev/null "${BASE_URL}/infra-envs/${INFRA_ENV_ID}/hosts/${HOST_ID}/actions/install")
+if [[ $res = "202" ]]; then 
+    printf '\nHost installation started\n' 1>&2
+else
+    printf '\nHost installation failed\n' 1>&2
+    exit 1
+fi

data/data/agent/systemd/units/agent-add-node.service

@@ -0,0 +1,21 @@
+[Unit]
+Description=Adds the current node to an already existing cluster
+Wants=network-online.target
+Requires=apply-host-config.service
+PartOf=assisted-service-pod.service
+After=network-online.target apply-host-config.service
+ConditionPathExists=/etc/assisted/node0
+
+[Service]
+EnvironmentFile=/usr/local/share/assisted-service/assisted-service.env
+EnvironmentFile=/usr/local/share/start-cluster/start-cluster.env
+EnvironmentFile=/etc/assisted/rendezvous-host.env
+ExecStartPre=/usr/local/bin/wait-for-assisted-service.sh
+ExecStart=/usr/local/bin/add-node.sh
+
+KillMode=none
+Type=oneshot
+RemainAfterExit=true
+
+[Install]
+WantedBy=multi-user.target

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

@@ -0,0 +1,28 @@
+[Unit]
+Description=Imports an already existing 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/assisted-service/assisted-service.env
+EnvironmentFile=/etc/assisted/add-nodes.env
+ExecStartPre=/bin/rm -f %t/%n.ctr-id
+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 --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 OPENSHIFT_INSTALL_RELEASE_IMAGE_MIRROR --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

@@ -14,7 +14,7 @@ EnvironmentFile=/usr/local/share/assisted-service/assisted-service.env
 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
 

hack/build-node-joiner.sh

@@ -0,0 +1,50 @@
+#!/bin/sh
+
+set -ex
+
+# shellcheck disable=SC2068
+version() { IFS="."; printf "%03d%03d%03d\\n" $@; unset IFS;}
+
+minimum_go_version=1.21
+current_go_version=$(go version | cut -d " " -f 3)
+
+if [ "$(version "${current_go_version#go}")" -lt "$(version "$minimum_go_version")" ]; then
+     echo "Go version should be greater or equal to $minimum_go_version"
+     exit 1
+fi
+
+export CGO_ENABLED=0
+MODE="${MODE:-release}"
+
+GIT_COMMIT="${SOURCE_GIT_COMMIT:-$(git rev-parse --verify 'HEAD^{commit}')}"
+GIT_TAG="${BUILD_VERSION:-$(git describe --always --abbrev=40 --dirty)}"
+DEFAULT_ARCH="${DEFAULT_ARCH:-amd64}"
+GOFLAGS="${GOFLAGS:--mod=vendor}"
+GCFLAGS=""
+LDFLAGS="${LDFLAGS} -X github.com/openshift/installer/pkg/version.Raw=${GIT_TAG} -X github.com/openshift/installer/pkg/version.Commit=${GIT_COMMIT} -X github.com/openshift/installer/pkg/version.defaultArch=${DEFAULT_ARCH}"
+TAGS="${TAGS:-}"
+OUTPUT="${OUTPUT:-bin/node-joiner}"
+
+case "${MODE}" in
+release)
+	LDFLAGS="${LDFLAGS} -s -w"
+	TAGS="${TAGS} release"
+	;;
+dev)
+    GCFLAGS="${GCFLAGS} all=-N -l"
+	;;
+*)
+	echo "unrecognized mode: ${MODE}" >&2
+	exit 1
+esac
+
+if test "${SKIP_GENERATION}" != y
+then
+	# this step has to be run natively, even when cross-compiling
+	GOOS='' GOARCH='' go generate ./data
+fi
+
+echo "building node-joiner"
+
+# shellcheck disable=SC2086
+go build ${GOFLAGS} -gcflags "${GCFLAGS}" -ldflags "${LDFLAGS}" -tags "${TAGS}" -o "${OUTPUT}" ./cmd/node-joiner

pkg/asset/agent/image/agentartifacts.go

@@ -92,7 +92,7 @@ func (a *AgentArtifacts) Generate(dependencies asset.Parents) error {
 func (a *AgentArtifacts) fetchAgentTuiFiles(releaseImage string, pullSecret string, mirrorConfig []mirror.RegistriesConfig) ([]string, error) {
 	release := NewRelease(
 		Config{MaxTries: OcDefaultTries, RetryDelay: OcDefaultRetryDelay},
-		releaseImage, pullSecret, mirrorConfig)
+		releaseImage, pullSecret, mirrorConfig, nil)
 
 	agentTuiFilenames := []string{"/usr/bin/agent-tui", "/usr/lib64/libnmstate.so.*"}
 	files := []string{}

pkg/asset/agent/image/agentimage.go

@@ -20,8 +20,9 @@ import (
 )
 
 const (
-	agentISOFilename    = "agent.%s.iso"
-	iso9660Level1ExtLen = 3
+	agentISOFilename         = "agent.%s.iso"
+	agentAddNodesISOFilename = "agent-addnodes.%s.iso"
+	iso9660Level1ExtLen      = 3
 )
 
 // AgentImage is an asset that generates the bootable image used to install clusters.
@@ -34,6 +35,7 @@ type AgentImage struct {
 	rootFSURL            string
 	bootArtifactsBaseURL string
 	platform             hiveext.PlatformType
+	isoFilename          string
 }
 
 var _ asset.WritableAsset = (*AgentImage)(nil)
@@ -61,9 +63,11 @@ func (a *AgentImage) Generate(dependencies asset.Parents) error {
 	switch agentWorkflow.Workflow {
 	case workflow.AgentWorkflowTypeInstall:
 		a.platform = agentManifests.AgentClusterInstall.Spec.PlatformType
+		a.isoFilename = agentISOFilename
 
 	case workflow.AgentWorkflowTypeAddNodes:
 		a.platform = clusterInfo.PlatformType
+		a.isoFilename = agentAddNodesISOFilename
 
 	default:
 		return fmt.Errorf("AgentWorkflowType value not supported: %s", agentWorkflow.Workflow)
@@ -239,7 +243,7 @@ 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))
+	agentIsoFile := filepath.Join(directory, fmt.Sprintf(a.isoFilename, a.cpuArch))
 
 	// Remove symlink if it exists
 	os.Remove(agentIsoFile)