Integrate k3s and multus

Signed-off-by: Manuel Buil <[email protected]>

docs/adrs/multus.md

@@ -0,0 +1,44 @@
+# Record architecture decisions
+
+Date: 2024-02-14
+
+## Status
+
+Approved
+
+## Context
+
+### Multus
+
+Multus is a CNI multiplexer that allows pods to have multiple network interfaces. We have users that are operating K3s + Multus but it is not super obvious how to configure it to work with K3s and how to add the additional pieces needed (e.g. IPAM or additional CNI plugins). We could facilitate this by creating an integration with Multus.
+
+We we will wait a bit to include whereabouts. That project is using very old dependencies which will creep in CVEs
+
+### Design suggestion
+
+Add multus to the k3s-charts repo. That multus chart will consume the tarball we generate in rke2-charts, i.e. both rke2 and k3s will use the same chart with minimal diffs (e.g. the Chart name will be k3s-multus instead of rke2-multus).
+
+Then, multus will be consumed as traefik:
+* The chart gets downloaded with `make download`
+* The chart tarball gets embedded in k3s binary with `go generate` and included in `pkg/static/zz_generated_bindata.go`
+* The HelmChart manifest pointing to the chart tarball gets embedded in k3s binary with `go generate` and included in `pkg/deploy/zz_generated_bindata.go`
+
+K3s will include a new `--multus` boolean flag. When that flag is true, we would leave the HelmChart manifest installing multus and whereabouts.
+
+The multus chart will install a daemonset that:
+* deploys the necessary binaries (multus and common CNI plugins) in each node
+* generates the correct CNI plugin
+* Installs the required CRDs
+
+It sucks a bit that the daemonset stays dormant forever after doing the job instead of just dying, but the alternatives are worse
+
+## Alternatives
+
+* K3s creates a job that picks the multus and whereabouts CNI plugins from the `image-build-cni-plugins` and copies them to each node. However, configuring jobs to run on each node is not that easy and very error prone. Therefore, we decided to reject this idea
+
+* K3s includes the multus and whereabouts CNI plugins as part of its multi-exec cni binary. However, the whereabouts binary is using very old dependencies which would creep in CVEs. Moreover, the size of the K3s binary would increase more than 10%, something not acceptable for a something that the vast majority of K3s users will not enable
+
+
+## Decision
+
+YES

manifests/multus.yaml

@@ -0,0 +1,14 @@
+---
+apiVersion: helm.cattle.io/v1
+kind: HelmChart
+metadata:
+  name: multus
+  namespace: kube-system
+spec:
+  chart: https://%{KUBERNETES_API}%/static/charts/rke2-multus-v4.0.2-build2024020801.tgz
+  valuesContent: |-
+    config:
+      cni_conf:
+        confDir: /var/lib/rancher/k3s/agent/etc/cni/net.d
+        binDir: %{DATA_DIR}%
+        kubeconfig: /var/lib/rancher/k3s/agent/etc/cni/net.d/multus.d/multus.kubeconfig

pkg/cli/cmds/server.go

@@ -105,6 +105,7 @@ type Server struct {
 	EtcdS3Folder             string
 	EtcdS3Timeout            time.Duration
 	EtcdS3Insecure           bool
+	Multus					 bool
 	ServiceLBNamespace       string
 }
 
@@ -489,6 +490,11 @@ var ServerFlags = []cli.Flag{
 		Usage:       "(experimental/components) Enable embedded distributed container registry; requires use of embedded containerd",
 		Destination: &ServerConfig.EmbeddedRegistry,
 	},
+	&cli.BoolFlag{
+		Name:        "multus",
+		Usage:       "(experimental/networking) Enable multus",
+		Destination: &ServerConfig.Multus,
+	},
 	NodeNameFlag,
 	WithNodeIDFlag,
 	NodeLabels,

pkg/cli/server/server.go

@@ -173,6 +173,7 @@ func run(app *cli.Context, cfg *cmds.Server, leaderControllers server.CustomCont
 	serverConfig.ControlConfig.EncryptSecrets = cfg.EncryptSecrets
 	serverConfig.ControlConfig.EtcdExposeMetrics = cfg.EtcdExposeMetrics
 	serverConfig.ControlConfig.EtcdDisableSnapshots = cfg.EtcdDisableSnapshots
+	serverConfig.ControlConfig.Multus = cfg.Multus
 	serverConfig.ControlConfig.VLevel = cmds.LogConfig.VLevel
 	serverConfig.ControlConfig.VModule = cmds.LogConfig.VModule
 
@@ -396,6 +397,11 @@ func run(app *cli.Context, cfg *cmds.Server, leaderControllers server.CustomCont
 		serverConfig.ControlConfig.Disables["ccm"] = true
 	}
 
+	if !serverConfig.ControlConfig.Multus {
+		serverConfig.ControlConfig.Skips["multus"] = true
+		serverConfig.ControlConfig.Disables["multus"] = true
+	}
+
 	tlsMinVersionArg := getArgValueFromList("tls-min-version", serverConfig.ControlConfig.ExtraAPIArgs)
 	serverConfig.ControlConfig.TLSMinVersion, err = kubeapiserverflag.TLSVersion(tlsMinVersionArg)
 	if err != nil {

pkg/daemons/config/types.go

@@ -157,6 +157,7 @@ type CriticalControlArgs struct {
 	FlannelIPv6Masq       bool         `cli:"flannel-ipv6-masq"`
 	FlannelExternalIP     bool         `cli:"flannel-external-ip"`
 	EgressSelectorMode    string       `cli:"egress-selector-mode"`
+	Multus				  bool         `cli:"multus"`	
 	ServiceIPRange        *net.IPNet   `cli:"service-cidr"`
 	ServiceIPRanges       []*net.IPNet `cli:"service-cidr"`
 }

pkg/server/server.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"os"
+	"os/exec"
 	"path"
 	"path/filepath"
 	"runtime/debug"
@@ -279,6 +280,13 @@ func stageFiles(ctx context.Context, sc *Context, controlConfig *config.Control)
 		dnsIPFamilyPolicy = "RequireDualStack"
 	}
 
+	// Find the /var/lib/rancher/k3s/data/${SHA}/bin/ directory. Same procedure we use in pkg/agent/config/config.go
+	hostLocal, err := exec.LookPath("host-local")
+	if err != nil {
+		return errors.Wrap(err, "failed to find host-local")
+	}
+	CNIBinDir := filepath.Dir(hostLocal)
+
 	templateVars := map[string]string{
 		"%{CLUSTER_DNS}%":                 controlConfig.ClusterDNS.String(),
 		"%{CLUSTER_DNS_LIST}%":            fmt.Sprintf("[%s]", util.JoinIPs(controlConfig.ClusterDNSs)),
@@ -288,13 +296,15 @@ func stageFiles(ctx context.Context, sc *Context, controlConfig *config.Control)
 		"%{SYSTEM_DEFAULT_REGISTRY}%":     registryTemplate(controlConfig.SystemDefaultRegistry),
 		"%{SYSTEM_DEFAULT_REGISTRY_RAW}%": controlConfig.SystemDefaultRegistry,
 		"%{PREFERRED_ADDRESS_TYPES}%":     addrTypesPrioTemplate(controlConfig.FlannelExternalIP),
+		"%{DATA_DIR}%":                    CNIBinDir,
 	}
 
 	skip := controlConfig.Skips
 	if !skip["traefik"] && isHelmChartTraefikV1(sc) {
 		logrus.Warn("Skipping Traefik v2 deployment due to existing Traefik v1 installation")
 		skip["traefik"] = true
 	}
+	logrus.Infof("MANU - templateVars: %v", templateVars)
 	if err := deploy.Stage(dataDir, templateVars, skip); err != nil {
 		return err
 	}

scripts/download

@@ -41,5 +41,9 @@ git clone --single-branch --branch=${VERSION_CONTAINERD} --depth=1 https://${PKG
 
 for CHART_FILE in $(grep -rlF HelmChart manifests/ | xargs yq eval --no-doc .spec.chart | xargs -n1 basename); do
   CHART_NAME=$(echo $CHART_FILE | grep -oE '^(-*[a-z])+')
-  curl -sfL ${CHARTS_URL}/${CHART_NAME}/${CHART_FILE} -o ${CHARTS_DIR}/${CHART_FILE}
+  if [[ $CHART_NAME == *"rke"* ]]; then
+	  curl -sfL https://github.com/rancher/rke2-charts/raw/main/assets/rke2-multus/rke2-multus-v4.0.2-build2024020801.tgz -o ${CHARTS_DIR}/${CHART_FILE}
+  else
+	  curl -sfL ${CHARTS_URL}/${CHART_NAME}/${CHART_FILE} -o ${CHARTS_DIR}/${CHART_FILE}
+  fi
 done