edgeguard-native/cmd/edgeguard-scheduler/main.go

// edgeguard-scheduler runs background jobs that don't belong on the
// API request path:
//
//   - ACME cert renewal (every 6h, re-issues anything < 30d to expiry)
//
// Future jobs (cluster heartbeat, backup, audit-log retention)
// hang off the same Tick loop. Stays single-process — no leader
// election yet (Phase 3).
package main

import (
	"context"
	"encoding/json"
	"fmt"
	"log/slog"
	"os"
	"strconv"
	"time"

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

	"git.netcell-it.de/projekte/edgeguard-native/internal/cluster"
	"git.netcell-it.de/projekte/edgeguard-native/internal/database"
	"git.netcell-it.de/projekte/edgeguard-native/internal/license"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/acme"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/alerts"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/backup"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/certrenewer"
	licsvc "git.netcell-it.de/projekte/edgeguard-native/internal/services/license"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/setup"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/tlscerts"
)

var version = "1.0.74"

const (
	// renewTickInterval — how often we re-evaluate expiring certs.
	// 6h is enough: LE renewal window is 30 days; missing one tick
	// makes no difference. Hourly would log too much.
	renewTickInterval = 6 * time.Hour

	// certDir matches handlers.NewTLSCertsHandler default — HAProxy
	// reads from this directory.
	certDir = "/etc/edgeguard/tls"

	// licenseTickInterval — daily re-verify against
	// license.netcell-it.com. Result lands in the licenses table.
	licenseTickInterval = 24 * time.Hour

	// backupTickInterval — daily scheduled backup at ~03:00 (Tick
	// alignment ist approximativ, weil time.Ticker bei Boot startet).
	// Retention: 14 erfolgreiche Backups (default in backup.Service).
	backupTickInterval = 24 * time.Hour

	// configHashTickInterval — alle 5 min config_hash neu berechnen
	// und in ha_nodes der eigenen Row schreiben. Cluster-UI nutzt
	// das fürs Drift-Banner — pro-Mutation-Refresh wäre teuer.
	configHashTickInterval = 5 * time.Minute
)

func main() {
	slog.SetDefault(slog.New(slog.NewTextHandler(os.Stderr, &slog.HandlerOptions{Level: slog.LevelInfo})))
	slog.Info("edgeguard-scheduler starting", "version", version)

	ctx := context.Background()

	pool, err := database.Open(ctx, database.ConnStringFromEnv())
	if err != nil {
		slog.Error("scheduler: DB open failed — sleeping forever", "error", err)
		select {}
	}
	defer pool.Close()

	tlsRepo := tlscerts.New(pool)
	setupStore := setup.NewStore(setup.DefaultDir)
	st, _ := setupStore.Load()

	var renewer *certrenewer.Service
	if st != nil && st.ACMEEmail != "" {
		issuer := acme.New(st.ACMEEmail)
		renewer = certrenewer.New(tlsRepo, issuer, certDir, 30*24*time.Hour)
		slog.Info("scheduler: ACME renewer enabled",
			"email", st.ACMEEmail, "tick", renewTickInterval, "threshold", "30d")
	} else {
		slog.Warn("scheduler: setup.acme_email empty — ACME renewal disabled until setup wizard ran")
	}

	licRepo := licsvc.New(pool)
	licClient := license.NewClient()
	licKeyStore := license.NewKeyStore()
	nodeID := os.Getenv("EDGEGUARD_NODE_ID")
	slog.Info("scheduler: license re-verify enabled", "tick", licenseTickInterval)

	backupSvc := backup.New(pool)
	slog.Info("scheduler: daily backup enabled", "tick", backupTickInterval,
		"dir", backupSvc.BackupDir, "keep_n", backup.DefaultKeepN)

	alertSvc := alerts.New(pool)
	alertDedupe := newDedupe(12 * time.Hour)

	if renewer != nil {
		runRenewer(ctx, renewer, alertSvc, alertDedupe)
	}
	runLicenseVerify(ctx, licClient, licKeyStore, licRepo, nodeID, alertSvc, alertDedupe)

	// Lokale Node-ID für config-hash-refresh. EnsureNodeID liefert
	// dieselbe ID die die API hat (gleiches /var/lib/edgeguard/node-id).
	localID, _ := cluster.EnsureNodeID("")
	slog.Info("scheduler: config-hash refresh enabled", "tick", configHashTickInterval, "node_id", localID)
	// Initial-Refresh damit /cluster/status nach API+Scheduler-Boot
	// nicht 5min auf den ersten Wert wartet.
	runConfigHash(ctx, pool, localID)

	renewTick := time.NewTicker(renewTickInterval)
	defer renewTick.Stop()
	licTick := time.NewTicker(licenseTickInterval)
	defer licTick.Stop()
	backupTick := time.NewTicker(backupTickInterval)
	defer backupTick.Stop()
	hashTick := time.NewTicker(configHashTickInterval)
	defer hashTick.Stop()

	for {
		select {
		case <-renewTick.C:
			if renewer != nil {
				runRenewer(ctx, renewer, alertSvc, alertDedupe)
			}
			runCertExpiryCheck(ctx, tlsRepo, alertSvc, alertDedupe)
		case <-licTick.C:
			runLicenseVerify(ctx, licClient, licKeyStore, licRepo, nodeID, alertSvc, alertDedupe)
		case <-backupTick.C:
			runBackup(ctx, backupSvc, version, alertSvc)
		case <-hashTick.C:
			runConfigHash(ctx, pool, localID)
		}
	}
}

// dedupe verhindert dass derselbe Alert-Key (z.B. "cert.expiring:utm-1.netcell-it.de")
// öfter als alle 12h gefeuert wird. In-memory — Scheduler-Restart
// resettet, was OK ist (Operator soll bei restart wieder einen kennen-
// lernen-Event sehen können).
type dedupe struct {
	ttl  time.Duration
	last map[string]time.Time
}

func newDedupe(ttl time.Duration) *dedupe { return &dedupe{ttl: ttl, last: map[string]time.Time{}} }

func (d *dedupe) shouldFire(key string) bool {
	now := time.Now()
	if last, ok := d.last[key]; ok && now.Sub(last) < d.ttl {
		return false
	}
	d.last[key] = now
	return true
}

// runCertExpiryCheck prüft tls_certs auf bevorstehende Expiry. Warning
// bei <14d Restzeit. Dedupe pro Cert-Name 12h damit der scheduler
// nicht alle 6h dieselbe Warnung feuert.
func runCertExpiryCheck(ctx context.Context, repo *tlscerts.Repo,
	a *alerts.Service, d *dedupe) {
	if repo == nil || a == nil {
		return
	}
	certs, err := repo.List(ctx)
	if err != nil {
		slog.Warn("scheduler: cert-expiry list failed", "error", err)
		return
	}
	threshold := 14 * 24 * time.Hour
	now := time.Now()
	for _, c := range certs {
		if c.NotAfter == nil {
			continue
		}
		remain := c.NotAfter.Sub(now)
		if remain > threshold || remain < -90*24*time.Hour {
			continue
		}
		key := "cert.expiring:" + c.Domain
		if !d.shouldFire(key) {
			continue
		}
		days := int(remain.Hours() / 24)
		sev := alerts.SeverityWarning
		if days < 3 {
			sev = alerts.SeverityError
		}
		_, err := a.Fire(ctx, "cert.expiring", sev,
			"TLS-Zertifikat läuft ab: "+c.Domain,
			"Cert für "+c.Domain+" läuft in "+strconv.Itoa(days)+" Tagen ab ("+c.NotAfter.Format(time.RFC3339)+"). Renewer-Status: "+c.Status)
		if err != nil {
			slog.Warn("scheduler: alert fire failed", "error", err)
		}
	}
}

// runConfigHash berechnet den Hash und schreibt ihn in ha_nodes.
// Pool kann nil sein (scheduler-pool-fail beim boot) — dann no-op.
func runConfigHash(ctx context.Context, pool *pgxpoolPool, localID string) {
	if pool == nil || localID == "" {
		return
	}
	hash, err := cluster.RefreshLocalHash(ctx, pool, localID)
	if err != nil {
		slog.Warn("scheduler: config-hash refresh failed", "error", err)
		return
	}
	slog.Debug("scheduler: config-hash refreshed", "hash", hash)
}

// pgxpoolPool ist ein lokaler Alias damit die Signatur stabil bleibt
// wenn wir später den pool austauschen wollen (z.B. read-only-replica).
type pgxpoolPool = pgxpool.Pool

// runBackup führt einen scheduled Backup aus + prunet alte. Failures
// loggen wir + alarmieren — verlorene Backups sind kritisch.
func runBackup(ctx context.Context, svc *backup.Service, version string, a *alerts.Service) {
	res, err := svc.Run(ctx, backup.KindScheduled, version)
	if err != nil {
		slog.Warn("scheduler: backup failed", "error", err, "file", res.File)
		if a != nil {
			_, _ = a.Fire(ctx, "backup.failed", alerts.SeverityError,
				"Backup fehlgeschlagen",
				"Scheduled Backup konnte nicht erstellt werden: "+err.Error())
		}
		return
	}
	slog.Info("scheduler: backup done",
		"file", res.File, "size", res.SizeBytes,
		"db_bytes", res.DBDumpBytes, "files_bytes", res.FilesBytes,
		"sha256", res.SHA256)
	if err := svc.Prune(ctx, backup.DefaultKeepN); err != nil {
		slog.Warn("scheduler: backup prune failed", "error", err)
	}
}

// runLicenseVerify performs a single re-verify pass. Empty key = no-op
// (box stays in trial), so this is safe to call on every tick.
// Bei valid:false-Antwort + Stand >7d alt → Warnung an Alerts.
func runLicenseVerify(ctx context.Context, c *license.Client, ks *license.KeyStore,
	repo *licsvc.Repo, nodeID string, a *alerts.Service, d *dedupe) {
	key := ks.Get()
	if key == "" {
		slog.Debug("scheduler: license verify skipped — no key")
		return
	}
	res, err := c.Verify(key)
	if err != nil {
		_ = repo.MarkError(ctx, key, err.Error())
		slog.Warn("scheduler: license verify failed", "error", err)
		return
	}
	payload, _ := json.Marshal(res)
	status := "active"
	if !res.Valid {
		status = "expired"
		if res.Status == "revoked" {
			status = "invalid"
		}
	}
	if err := repo.Upsert(ctx, key, status, res.ExpiresAt, nodeID, 0, payload, ""); err != nil {
		slog.Warn("scheduler: license db upsert failed", "error", err)
		return
	}
	slog.Info("scheduler: license verified",
		"status", status, "valid", res.Valid, "expires_at", res.ExpiresAt)

	// Alarm bei ungültiger Lizenz (revoked, expired) — dedupe 12h damit
	// der Operator nicht alle 24h denselben Alert bekommt.
	if a != nil && d != nil && !res.Valid {
		if d.shouldFire("license.invalid") {
			_, _ = a.Fire(ctx, "license.invalid", alerts.SeverityError,
				"License "+status,
				"License-Server liefert valid=false. Reason: "+res.Reason)
		}
	}
}

func runRenewer(ctx context.Context, r *certrenewer.Service, a *alerts.Service, d *dedupe) {
	res, err := r.Run(ctx)
	if err != nil {
		slog.Error("scheduler: renewer run failed", "error", err)
		if a != nil && d != nil && d.shouldFire("cert.renewer.run_failed") {
			_, _ = a.Fire(ctx, "cert.renewer.run_failed", alerts.SeverityError,
				"ACME-Renewer-Lauf fehlgeschlagen",
				"Certrenewer-Cycle abgebrochen: "+err.Error())
		}
		return
	}
	slog.Info("scheduler: renewer pass complete",
		"checked", res.Checked, "renewed", res.Renewed,
		"failed", res.Failed, "skipped", res.Skipped)
	if a != nil && res.Failed > 0 && d != nil && d.shouldFire("cert.renew_failed") {
		_, _ = a.Fire(ctx, "cert.renew_failed", alerts.SeverityError,
			"Cert-Renewal teilweise fehlgeschlagen",
			fmt.Sprintf("Renewer-Cycle: %d checked, %d renewed, %d failed, %d skipped",
				res.Checked, res.Renewed, res.Failed, res.Skipped))
	}
}