deployment: refactor config manager to support NRI enabling in CRI-O

This commit extends config manager code and the plugins helm charts so that CRI-O users are also able to enable NRI via our charts if they wish to. Same parameter is used to opt in for the feature in Helm charts and we don't require users to indicate what container runtime is being used. Instead the config manager auto-detects the runtime and does the necessary changes to its configuration file. In scenarios with multiple active runtimes (e.g., CRI-O and containerd), the manager gracefully exits and throws an error. Signed-off-by: Feruzjon Muyassarov <[email protected]>
containers · Oct 3, 2023 · 0008b76 · 0008b76
1 parent e00e6f0
commit 0008b76
Show file tree

Hide file tree

Showing 7 changed files with 112 additions and 45 deletions.
diff --git a/cmd/config-manager/Dockerfile b/cmd/config-manager/Dockerfile
@@ -14,6 +14,6 @@ COPY cmd/config-manager/*.go ./
 RUN go build -tags osusergo,netgo -ldflags "-extldflags=-static" -o config-manager
 
 # Final Image
-FROM gcr.io/distroless/static
+FROM gcr.io/distroless/static:debug
 COPY --from=build /go/builder/config-manager /bin/config-manager
 ENTRYPOINT ["/bin/config-manager"]
diff --git a/cmd/config-manager/main.go b/cmd/config-manager/main.go
@@ -29,32 +29,45 @@ import (
 )
 
 const (
-	tomlFilePath = "/etc/containerd/config.toml"
-	nriPluginKey = "io.containerd.nri.v1.nri"
-	disableKey   = "disable"
-	replaceMode  = "replace"
-	resultDone   = "done"
-	unit         = "containerd.service"
+	containerdConfig = "/etc/containerd/config.toml"
+	nriPluginKey     = "io.containerd.nri.v1.nri"
+	replaceMode      = "replace"
+	resultDone       = "done"
+	containerdUnit   = "containerd.service"
+	crioUnit         = "crio.service"
 )
 
 func main() {
-	tomlMap, err := readConfig(tomlFilePath)
+	unit, err := detectRuntime()
 	if err != nil {
-		log.Fatalf("Error reading TOML file: %v", err)
+		log.Fatalf("failed to autodetect container runtime: %v", err)
 	}
 
-	updatedTomlMap := updateNRIPlugin(tomlMap)
+	// Edit containerd config only if containerd is detected to be the runtime.
+	if unit == containerdUnit {
+		tomlMap, err := readConfig(containerdConfig)
+		if err != nil {
+			log.Fatalf("Error reading TOML file: %v", err)
+		}
 
-	err = writeConfig(tomlFilePath, updatedTomlMap)
-	if err != nil {
-		log.Fatalf("failed to write updated config into a file %q:, %v", tomlFilePath, err)
+		updatedTomlMap := updateContainerdConfig(tomlMap)
+
+		err = writeConfig(containerdConfig, updatedTomlMap)
+		if err != nil {
+			log.Fatalf("failed to write updated config into a file %q:, %v", containerdConfig, err)
+		}
 	}
 
-	err = restartSystemdUnit(unit)
-	if err != nil {
-		log.Fatalf("failed to restart containerd: %v", err)
+	// If we reach this point, it indicates that the CRI-O runtime is in use,
+	// and we perform a restart only. NRI related modifications to the CRI-O
+	// are accomplished through a mounted drop-in file.
+	if err = restartSystemdUnit(unit); err != nil {
+		log.Fatalf("failed to restart %q unit: %v", unit, err)
 	}
+
+	log.Println("Enabled NRI for", unit)
 }
+
 func writeConfig(file string, config map[string]interface{}) error {
 	var buf bytes.Buffer
 	enc := tomlv2.NewEncoder(&buf)
@@ -90,7 +103,7 @@ func readConfig(file string) (map[string]interface{}, error) {
 	return tomlMap, nil
 }
 
-func updateNRIPlugin(config map[string]interface{}) map[string]interface{} {
+func updateContainerdConfig(config map[string]interface{}) map[string]interface{} {
 	plugins, exists := config["plugins"].(map[string]interface{})
 	if !exists {
 		log.Println("Top level plugins section not found, adding it to enable NRI...")
@@ -105,15 +118,37 @@ func updateNRIPlugin(config map[string]interface{}) map[string]interface{} {
 		plugins[nriPluginKey] = nri
 	}
 
-	nri[disableKey] = false
-	log.Println("Enabled NRI...")
+	nri["disable"] = false
 	return config
 }
 
+func detectRuntime() (string, error) {
+	conn, err := dbus.NewSystemConnectionContext(context.Background())
+	if err != nil {
+		return "", fmt.Errorf("failed to create DBus connection: %w", err)
+	}
+	defer conn.Close()
+
+	// Filter out active container runtime (CRI-O or containerd) systemd units on the node.
+	// It is expected that only one container runtime systemd unit should be active at a time
+	// (either containerd or CRI-O).If more than one container runtime systemd unit is found
+	// to be in an active state, the process fails.
+	units, err := conn.ListUnitsByPatternsContext(context.Background(), []string{"active"}, []string{containerdUnit, crioUnit})
+	if err != nil {
+		return "", fmt.Errorf("failed to detect container runtime in use: %w", err)
+	}
+
+	if len(units) > 1 {
+		return "", fmt.Errorf("detected more than one container runtime on the host, expected one")
+	}
+
+	return units[0].Name, nil
+}
+
 func restartSystemdUnit(unit string) error {
 	conn, err := dbus.NewSystemConnectionContext(context.Background())
 	if err != nil {
-		return fmt.Errorf("failed to create DBus connection for unit %q: %w", unit, err)
+		return fmt.Errorf("failed to create DBus connection: %w", err)
 	}
 	defer conn.Close()
 

diff --git a/deployment/helm/resource-management-policies/balloons/templates/daemonset.yaml b/deployment/helm/resource-management-policies/balloons/templates/daemonset.yaml
@@ -17,14 +17,25 @@ spec:
       serviceAccount: nri-resource-policy-balloons
       nodeSelector:
         kubernetes.io/os: "linux"
-      {{- if .Values.nri.patchContainerdConfig }}
+      {{- if .Values.nri.patchRuntime }}
       initContainers:
-      - name: patch-containerd
+      - name: patch-runtime
         image: {{ .Values.initContainerImage.name }}:{{ .Values.initContainerImage.tag | default .Chart.AppVersion }}
         imagePullPolicy: {{ .Values.initContainerImage.pullPolicy }}
+        restartPolicy: Never
+        command: ["sh","-c"]
+        args:
+        - |
+          if [ ! -f /etc/crio/crio.conf.d/10-enable-nri.conf ];then
+          cat <<EOF >/etc/crio/crio.conf.d/10-enable-nri.conf
+          [crio.nri]
+          enable_nri = true
+          EOF
+          fi
+          /bin/config-manager
         volumeMounts:
-        - name: containerd-config
-          mountPath: /etc/containerd/config.toml
+        - name: etc
+          mountPath: /etc
         - name: dbus-socket
           mountPath: /var/run/dbus/system_bus_socket
         securityContext:
@@ -91,11 +102,11 @@ spec:
         hostPath:
           path: /var/run/nri
           type: Directory
-      {{- if .Values.nri.patchContainerdConfig }}
-      - name: containerd-config
+      {{- if .Values.nri.patchRuntime }}
+      - name: etc
         hostPath:
-          path: /etc/containerd/config.toml
-          type: File
+          path: /etc
+          type: Directory
       - name: dbus-socket
         hostPath:
           path: /var/run/dbus/system_bus_socket

diff --git a/deployment/helm/resource-management-policies/balloons/values.yaml b/deployment/helm/resource-management-policies/balloons/values.yaml
@@ -20,7 +20,8 @@ resources:
   memory: 512Mi
 
 nri:
-  patchContainerdConfig: false
+  patchRuntime: false
+
 
 initContainerImage:
   name: ghcr.io/containers/nri-plugins/nri-resource-policy-config-manager

diff --git a/deployment/helm/resource-management-policies/topology-aware/templates/daemonset.yaml b/deployment/helm/resource-management-policies/topology-aware/templates/daemonset.yaml
@@ -17,14 +17,25 @@ spec:
       serviceAccount: nri-resource-policy-topology-aware
       nodeSelector:
         kubernetes.io/os: "linux"
-      {{- if .Values.nri.patchContainerdConfig }}
+      {{- if .Values.nri.patchRuntime }}
       initContainers:
-      - name: patch-containerd
+      - name: patch-runtime
         image: {{ .Values.initContainerImage.name }}:{{ .Values.initContainerImage.tag | default .Chart.AppVersion }}
         imagePullPolicy: {{ .Values.initContainerImage.pullPolicy }}
+        restartPolicy: Never
+        command: ["sh","-c"]
+        args:
+        - |
+          if [ ! -f /etc/crio/crio.conf.d/10-enable-nri.conf ];then
+          cat <<EOF >/etc/crio/crio.conf.d/10-enable-nri.conf
+          [crio.nri]
+          enable_nri = true
+          EOF
+          fi
+          /bin/config-manager
         volumeMounts:
-        - name: containerd-config
-          mountPath: /etc/containerd/config.toml
+        - name: etc
+          mountPath: /etc
         - name: dbus-socket
           mountPath: /var/run/dbus/system_bus_socket
         securityContext:
@@ -91,11 +102,11 @@ spec:
         hostPath:
           path: /var/run/nri
           type: Directory
-      {{- if .Values.nri.patchContainerdConfig }}
-      - name: containerd-config
+      {{- if .Values.nri.patchRuntime }}
+      - name: etc
         hostPath:
-          path: /etc/containerd/config.toml
-          type: File
+          path: /etc
+          type: Directory
       - name: dbus-socket
         hostPath:
           path: /var/run/dbus/system_bus_socket

diff --git a/deployment/helm/resource-management-policies/topology-aware/values.yaml b/deployment/helm/resource-management-policies/topology-aware/values.yaml
@@ -20,7 +20,7 @@ resources:
   memory: 512Mi
 
 nri:
-  patchContainerdConfig: false
+  patchRuntime: false
 
 initContainerImage:
   name: ghcr.io/containers/nri-plugins/nri-resource-policy-config-manager

diff --git a/docs/resource-policy/installation.md b/docs/resource-policy/installation.md
@@ -18,23 +18,32 @@ following components: DaemonSet, ConfigMap, CustomResourceDefinition, and RBAC-r
 - Container runtime:
     - containerD:
         - At least [containerd 1.7.0](https://github.com/containerd/containerd/releases/tag/v1.7.0)
-            release version to use the NRI feature
+            release version to use the NRI feature.
+
         - Enable NRI feature by following [these](https://github.com/containerd/containerd/blob/main/docs/NRI.md#enabling-nri-support-in-containerd)
           detailed instructions. You can optionally enable the NRI in containerd using the Helm chart
-          during the chart installation simply by setting the `nri.patchContainerdConfig` parameter.
+          during the chart installation simply by setting the `nri.patchRuntime` parameter.
           For instance,
 
           ```sh
-          helm install topology-aware --namespace kube-system --set nri.patchContainerdConfig=true deployment/helm/resource-management-policies/topology-aware/
+          helm install topology-aware --namespace kube-system --set nri.patchRuntime=true deployment/helm/resource-management-policies/topology-aware/
           ```
 
-          Enabling `nri.patchContainerdConfig` creates an init container to turn on
+          Enabling `nri.patchRuntime` creates an init container to turn on
           NRI feature in containerd and only after that proceed the plugin installation.
 
     - CRI-O
         - At least [v1.26.0](https://github.com/cri-o/cri-o/releases/tag/v1.26.0) release version to
             use the NRI feature
         - Enable NRI feature by following [these](https://github.com/cri-o/cri-o/blob/main/docs/crio.conf.5.md#crionri-table) detailed instructions.
+          You can optionally enable the NRI in CRI-O using the Helm chart
+          during the chart installation simply by setting the `nri.patchRuntime` parameter.
+          For instance,
+
+          ```sh
+          helm install topology-aware --namespace kube-system --set nri.patchRuntime=true deployment/helm/resource-management-policies/topology-aware/
+          ```
+
 - Kubernetes 1.24+
 - Helm 3.0.0+
 
@@ -94,14 +103,14 @@ along with the default values, for the Topology-aware and Balloons plugins Helm
 
 | Name               | Default                                                                                                                       | Description                                          |
 | ------------------ | ----------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------- |
-| `image.name`       | [ghcr.io/containers/nri-plugins/nri-resource-policy-topology-aware](ghcr.io/containers/nri-plugins/nri-resource-policy-topology-aware)    | container image name                                 |
+| `image.name`       | [ghcr.io/containers/nri-plugins/nri-resource-policy-topology-aware](ghcr.io/containers/nri-plugins/nri-resource-policy-topology-aware)    | container image name                     |
 | `image.tag`        | unstable                                                                                                                      | container image tag                                  |
 | `image.pullPolicy` | Always                                                                                                                        | image pull policy                                    |
 | `resources.cpu`    | 500m                                                                                                                          | cpu resources for the Pod                            |
 | `resources.memory` | 512Mi                                                                                                                         | memory qouta for the Pod                             | 
 | `hostPort`         | 8891                                                                                                                          | metrics port to expose on the host                   |
 | `config`           | <pre><code>ReservedResources:</code><br><code>  cpu: 750m</code></pre>                                                        | plugin configuration data                            |
-| `nri.patchContainerdConfig`       | false                                                                                                          | enable/disable NRI in containerd.                    |
+| `nri.patchRuntime` | false                                                                                                                         | enable NRI in containerd or CRI-O                    |
 | `initImage.name`   | [ghcr.io/containers/nri-plugins/config-manager](ghcr.io/containers/nri-plugins/config-manager)                                | init container image name                            |
 | `initImage.tag`    | unstable                                                                                                                      | init container image tag                             |
 | `initImage.pullPolicy` | Always                                                                                                                    | init container image pull policy                     |
@@ -117,7 +126,7 @@ along with the default values, for the Topology-aware and Balloons plugins Helm
 | `resources.memory` | 512Mi                                                                                                                         | memory qouta for the Pod                             | 
 | `hostPort`         | 8891                                                                                                                          | metrics port to expose on the host                   |
 | `config`           | <pre><code>ReservedResources:</code><br><code>  cpu: 750m</code></pre>                                                        | plugin configuration data                            |
-| `nri.patchContainerdConfig`       | false                                                                                                          | enable/disable NRI in containerd.                    |
+| `nri.patchRuntime` | false                                                                                                                         | enable NRI in containerd or CRI-O                    |
 | `initImage.name`   | [ghcr.io/containers/nri-plugins/config-manager](ghcr.io/containers/nri-plugins/config-manager)                                | init container image name                            |
 | `initImage.tag`    | unstable                                                                                                                      | init container image tag                             |
 | `initImage.pullPolicy` | Always                                                                                                                    | init container image pull policy                     |