edgeguard-native/internal/keepalived/keepalived.go

// Package keepalived rendert /etc/keepalived/keepalived.conf aus
// cluster_settings (VIP/VRRP-Config) und ha_nodes (local vs. peer).
//
// Split-Brain-Strategie: kein Auto-Promote. notify_master loggt nur
// und sendet einen internen Alert. Promotion ist immer manuell via
// "edgeguard-ctl promote" — das ist die einzig sichere Option ohne
// externes Quorum in einem 2-Node-Cluster.
package keepalived

import (
	"bytes"
	"context"
	_ "embed"
	"fmt"
	"os"
	"os/exec"
	"strings"
	"text/template"

	"github.com/jackc/pgx/v5/pgxpool"

	"git.netcell-it.de/projekte/edgeguard-native/internal/configgen"
	"git.netcell-it.de/projekte/edgeguard-native/internal/models"
)

const ConfPath = "/etc/keepalived/keepalived.conf"

//go:embed keepalived.conf.tpl
var cfgTpl string

var tpl = template.Must(template.New("keepalived").Parse(cfgTpl))

// VIPEntry ist eine einzelne VIP-Adresse die keepalived verwaltet.
type VIPEntry struct {
	Address string // z.B. 89.163.205.100
	Prefix  int    // z.B. 24
	Device  string // z.B. eth0
}

// View ist der Template-Kontext.
type View struct {
	State     string // MASTER | BACKUP
	Interface string // Interface für VRRP-Advertisements (VI_1)
	RouterID  int
	Priority  int    // MASTER=200, BACKUP=100
	SrcIP     string // eigene Public-IP (unicast_src_ip)
	PeerIP    string // Peer-Public-IP (unicast_peer)
	AuthPass  string
	VIPs      []VIPEntry // alle is_vip=true Einträge aus ip_addresses
	// Dual-path VRRP (Split-Brain-Schutz, Migration 0033)
	HBInterface string
	HBSrcIP     string
	HBPeerIP    string
	HBRouterID  int
	// GW-Tracking
	GWCheckIP string
}

type generator struct {
	pool    *pgxpool.Pool
	localID string
}

func New(pool *pgxpool.Pool, localID string) configgen.Generator {
	return &generator{pool: pool, localID: localID}
}

func (g *generator) Name() string { return "keepalived" }

func (g *generator) Render(ctx context.Context) error {
	cs, vips, local, peer, err := g.loadData(ctx)
	if err != nil {
		return fmt.Errorf("keepalived: load: %w", err)
	}
	if len(vips) == 0 {
		// Keine VIPs konfiguriert → keepalived.conf nicht schreiben.
		return nil
	}
	v := g.buildView(cs, vips, local, peer)
	var buf bytes.Buffer
	if err := tpl.Execute(&buf, v); err != nil {
		return fmt.Errorf("keepalived: template: %w", err)
	}
	if err := configgen.AtomicWrite(ConfPath, buf.Bytes(), 0o640); err != nil {
		return fmt.Errorf("keepalived: write: %w", err)
	}
	if err := reloadKeepalived(); err != nil {
		return fmt.Errorf("keepalived: reload: %w", err)
	}
	return nil
}

func (g *generator) loadData(ctx context.Context) (*models.ClusterSettings, []VIPEntry, *models.HANode, *models.HANode, error) {
	var cs models.ClusterSettings
	row := g.pool.QueryRow(ctx, `
		SELECT id, vip_address, vip_interface, vip_auth_pass, vrrp_router_id,
		       hb_interface, hb_src_ip, hb_peer_ip, hb_router_id, gw_check_ip
		FROM cluster_settings WHERE id = 1`)
	if err := row.Scan(&cs.ID, &cs.VIPAddress, &cs.VIPInterface, &cs.VIPAuthPass, &cs.VRRPRouterID,
		&cs.HBInterface, &cs.HBSrcIP, &cs.HBPeerIP, &cs.HBRouterID, &cs.GWCheckIP); err != nil {
		return nil, nil, nil, nil, fmt.Errorf("cluster_settings: %w", err)
	}

	// Alle VIPs aus ip_addresses (is_vip=true, active=true) inkl. Interface-Name.
	vipRows, err := g.pool.Query(ctx, `
		SELECT ia.address, ia.prefix, ni.name
		FROM ip_addresses ia
		JOIN network_interfaces ni ON ia.interface_id = ni.id
		WHERE ia.is_vip = true AND ia.active = true
		ORDER BY ni.name, ia.address`)
	if err != nil {
		return nil, nil, nil, nil, fmt.Errorf("ip_addresses: %w", err)
	}
	defer vipRows.Close()
	var vips []VIPEntry
	for vipRows.Next() {
		var v VIPEntry
		if err := vipRows.Scan(&v.Address, &v.Prefix, &v.Device); err != nil {
			continue
		}
		vips = append(vips, v)
	}

	nodeRows, err := g.pool.Query(ctx, `SELECT id, fqdn, role, pg_role, public_ip, status FROM ha_nodes ORDER BY joined_at`)
	if err != nil {
		return nil, nil, nil, nil, fmt.Errorf("ha_nodes: %w", err)
	}
	defer nodeRows.Close()

	var local, peer *models.HANode
	for nodeRows.Next() {
		n := &models.HANode{}
		if err := nodeRows.Scan(&n.ID, &n.FQDN, &n.Role, &n.PGRole, &n.PublicIP, &n.Status); err != nil {
			continue
		}
		if n.ID == g.localID {
			local = n
		} else {
			peer = n
		}
	}
	if local == nil {
		return nil, nil, nil, nil, fmt.Errorf("local node %s not in ha_nodes", g.localID)
	}
	return &cs, vips, local, peer, nil
}

func (g *generator) buildView(cs *models.ClusterSettings, vips []VIPEntry, local, peer *models.HANode) View {
	v := View{
		RouterID:    cs.VRRPRouterID,
		VIPs:        vips,
		Interface:   deref(cs.VIPInterface),
		AuthPass:    deref(cs.VIPAuthPass),
		HBInterface: deref(cs.HBInterface),
		HBSrcIP:     deref(cs.HBSrcIP),
		HBPeerIP:    deref(cs.HBPeerIP),
		HBRouterID:  cs.HBRouterID,
		GWCheckIP:   deref(cs.GWCheckIP),
	}
	if v.Interface == "" {
		v.Interface = "eth0"
	}
	if v.AuthPass == "" {
		v.AuthPass = "edgeguard"
	}
	if v.HBRouterID == 0 {
		v.HBRouterID = 52
	}

	// State IMMER BACKUP: das Template setzt `nopreempt`, und nopreempt wirkt
	// in keepalived NUR, wenn die Instanz im BACKUP-Zustand startet (bei state
	// MASTER wird nopreempt ignoriert). Die Priorität entscheidet weiterhin die
	// Initial-Election (primary=200 gewinnt), aber ein erholter Node reißt die
	// VIP NICHT mehr zurück → kein Flap-Back / Split-Brain. Deckt sich mit der
	// "kein Auto-Promote"-Philosophie: Promotion bleibt manuell.
	// pg_role=standby ist das härtere Signal (Standby ist nie bevorzugter Node).
	v.State = "BACKUP"
	if local.PGRole == "standby" {
		v.Priority = 100
	} else if local.PGRole == "primary" || local.Role == "primary" {
		v.Priority = 200
	} else {
		v.Priority = 100
	}

	if local.PublicIP != nil {
		v.SrcIP = *local.PublicIP
	}
	if peer != nil && peer.PublicIP != nil {
		v.PeerIP = *peer.PublicIP
	}
	return v
}

func reloadKeepalived() error {
	if _, err := os.Stat("/run/keepalived.pid"); os.IsNotExist(err) {
		// keepalived läuft noch nicht — erster Render beim Start.
		return nil
	}
	cmd := exec.Command("sudo", "-n", "/usr/bin/systemctl", "reload-or-restart", "keepalived.service")
	if out, err := cmd.CombinedOutput(); err != nil {
		return fmt.Errorf("sudo systemctl reload-or-restart keepalived.service: %w (output: %s)", err, strings.TrimSpace(string(out)))
	}
	return nil
}

func deref(s *string) string {
	if s == nil {
		return ""
	}
	return *s
}