metropolis/node/kubernetes/controller-manager.go

// Copyright The Monogon Project Authors.
// SPDX-License-Identifier: Apache-2.0

package kubernetes

import (
	"context"
	"encoding/pem"
	"fmt"
	"net"
	"os/exec"

	"source.monogon.dev/metropolis/node/kubernetes/pki"
	"source.monogon.dev/osbase/fileargs"
	"source.monogon.dev/osbase/supervisor"
)

type controllerManagerConfig struct {
	clusterNet net.IPNet
	serviceNet net.IPNet
	// All PKI-related things are in DER
	kubeConfig            []byte
	rootCA                []byte
	serviceAccountPrivKey []byte // In PKCS#8 form
	serverCert            []byte
	serverKey             []byte
}

func getPKIControllerManagerConfig(ctx context.Context, kpki *pki.PKI) (*controllerManagerConfig, error) {
	var config controllerManagerConfig
	var err error
	config.rootCA, _, err = kpki.Certificate(ctx, pki.IdCA)
	if err != nil {
		return nil, fmt.Errorf("failed to get ID root CA: %w", err)
	}
	config.serverCert, config.serverKey, err = kpki.Certificate(ctx, pki.ControllerManager)
	if err != nil {
		return nil, fmt.Errorf("failed to get controller-manager serving certificate: %w", err)
	}
	config.serviceAccountPrivKey, err = kpki.ServiceAccountKey(ctx)
	if err != nil {
		return nil, fmt.Errorf("failed to get serviceaccount privkey: %w", err)
	}
	config.kubeConfig, err = kpki.Kubeconfig(ctx, pki.ControllerManagerClient, pki.KubernetesAPIEndpointForController)
	if err != nil {
		return nil, fmt.Errorf("failed to get controller-manager kubeconfig: %w", err)
	}
	return &config, nil
}

func runControllerManager(config controllerManagerConfig) supervisor.Runnable {
	return func(ctx context.Context) error {
		args, err := fileargs.New()
		if err != nil {
			panic(err) // If this fails, something is very wrong. Just crash.
		}
		defer args.Close()

		cmd := exec.CommandContext(ctx, "/kubernetes/bin/kube", "kube-controller-manager",
			args.FileOpt("--kubeconfig", "kubeconfig", config.kubeConfig),
			args.FileOpt("--service-account-private-key-file", "service-account-privkey.pem",
				pem.EncodeToMemory(&pem.Block{Type: "PRIVATE KEY", Bytes: config.serviceAccountPrivKey})),
			args.FileOpt("--root-ca-file", "root-ca.pem",
				pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: config.rootCA})),
			args.FileOpt("--client-ca-file", "root-ca.pem",
				pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: config.rootCA})),
			"--use-service-account-credentials=true",
			fmt.Sprintf("--cluster-cidr=%v", config.clusterNet.String()),
			fmt.Sprintf("--service-cluster-ip-range=%v", config.serviceNet.String()),
			args.FileOpt("--tls-cert-file", "server-cert.pem",
				pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: config.serverCert})),
			args.FileOpt("--tls-private-key-file", "server-key.pem",
				pem.EncodeToMemory(&pem.Block{Type: "PRIVATE KEY", Bytes: config.serverKey})),
			"--allocate-node-cidrs",
			// Disables unused cloud control loops and prevents warnings.
			"--cloud-provider=external",
			"--controllers=*,-certificatesigningrequest-signing-controller",
			// This is intentionally empty, but if unset it tries to mkdir it
			// in the usual place, generating an error.
			"--flex-volume-plugin-dir=/kubernetes/conf/flexvolume-plugins",
			extraFeatureGates.AsFlag(),
		)

		if args.Error() != nil {
			return fmt.Errorf("failed to use fileargs: %w", err)
		}
		return supervisor.RunCommand(ctx, cmd, supervisor.ParseKLog())
	}
}