feat: per-Backend timeout server + Alerts-Deeplink + Retention + Cert-Prune — v1.3.4

- feat(backends): per-Backend `server_timeout_seconds` (nullable, Default 60s). Rendert `timeout server <N>s` im HAProxy-Backend-Block — für langsam antwortende Upstreams (KI-/Inferenz-Server mit gepufferter Antwort). Migration 0044 (+CHECK 1..86400), Model/Repo/Template/UI + Render-Test. - fix(ui): Dashboard-Alert-Karte verlinkt auf /alerts?tab=events; Alerts-Seite respektiert ?tab= (Deeplink landete bisher auf leerem Channels-Tab). - feat(scheduler): alert_events-Retention (90d) im täglichen Cleanup-Tick — Schutz vor unbounded growth der node-lokalen Health-Event-History. - fix(cluster): Cert-Sync prunt jetzt lokale .pem die der Primary nicht mehr hat (Waisen gelöschter Domains); schützt _default.pem + eigenen Node-Cert. - fix(security): x/text v0.37→v0.39 (GO-2026-5970, Infinite-Loop; via ACME+goose aktiv aufgerufen — govulncheck-Release-Gate). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 15:57:35 +02:00
parent 32ab2c7f47
commit 0ac91a7c59
17 changed files with 227 additions and 41 deletions
--- a/2
+++ b/2
@@ -1 +1 @@
-1.3.3
+1.3.4
--- a/cmd/edgeguard-scheduler/main.go
+++ b/cmd/edgeguard-scheduler/main.go
@@ -104,6 +104,12 @@ const (
 	auditCleanupInterval = 24 * time.Hour
 	auditRetentionDays   = 90

+	// alertRetentionDays — alert_events wächst sonst unbegrenzt (node-lokale
+	// Health-Events: backend.down, mem.high, cert.expiring …). Läuft im
+	// selben täglichen Tick wie der Audit-Cleanup. Fester Default, kein
+	// Setup-Override (Events sind reine Diagnose-History).
+	alertRetentionDays = 90
+
 	// backendDownCheckInterval — alle 2 Minuten HAProxy-Stats lesen und
 	// prüfen ob ein Backend komplett ausgefallen ist (alle Server DOWN).
 	// Dedupe 12h pro Backend → kein Alert-Spam. Frischer Alert wenn das
@@ -272,6 +278,7 @@ func main() {
 			runDiskCheck(ctx, alertSvc, alertDedupe)
 		case <-auditTick.C:
 			runAuditCleanup(ctx, auditRepo, setupStore)
+			runAlertCleanup(ctx, alertSvc)
 		case <-backendDownTick.C:
 			runBackendDownCheck(ctx, pool, alertSvc, alertDedupe)
 		case <-memTick.C:
@@ -317,6 +324,29 @@ func runAuditCleanup(ctx context.Context, r *audit.Repo, setupStore *setup.Store
 	}
 }

+// runAlertCleanup löscht alert_events älter als alertRetentionDays.
+// Schutz vor unbounded growth — auf einer aktiven Box feuern backend.down/
+// mem.high/cert.expiring über Monate tausende Rows (die Tabelle ist
+// node-lokal, wird also nirgends sonst abgeräumt). Best-effort: Fehler
+// werden nur geloggt.
+func runAlertCleanup(ctx context.Context, a *alerts.Service) {
+	if a == nil {
+		return
+	}
+	cctx, cancel := context.WithTimeout(ctx, 30*time.Second)
+	defer cancel()
+	n, err := a.Cleanup(cctx, alertRetentionDays)
+	if err != nil {
+		slog.Warn("scheduler: alert cleanup failed",
+			"keep_days", alertRetentionDays, "error", err)
+		return
+	}
+	if n > 0 {
+		slog.Info("scheduler: alert cleanup",
+			"deleted", n, "keep_days", alertRetentionDays)
+	}
+}
+
 // runDiskCheck prüft die Belegung von / via statfs. Fire-Schwellen:
 //   - >= 90% → Critical (error). Box ist akut gefährdet — beim
 //     nächsten Backup-Run oder größeren apt-Update droht "no space
--- a/go.mod
+++ b/go.mod
@@ -16,7 +16,7 @@ require (
 	github.com/pquerna/otp v1.5.0
 	github.com/pressly/goose/v3 v3.27.1
 	github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e
-	golang.org/x/crypto v0.52.0
+	golang.org/x/crypto v0.53.0
 	golang.org/x/oauth2 v0.36.0
 )

@@ -78,12 +78,12 @@ require (
 	go.uber.org/multierr v1.11.0 // indirect
 	go.yaml.in/yaml/v3 v3.0.4 // indirect
 	golang.org/x/arch v0.8.0 // indirect
-	golang.org/x/mod v0.35.0 // indirect
-	golang.org/x/net v0.55.0 // indirect
-	golang.org/x/sync v0.20.0 // indirect
-	golang.org/x/sys v0.45.0 // indirect
-	golang.org/x/text v0.37.0 // indirect
-	golang.org/x/tools v0.44.0 // indirect
+	golang.org/x/mod v0.37.0 // indirect
+	golang.org/x/net v0.56.0 // indirect
+	golang.org/x/sync v0.21.0 // indirect
+	golang.org/x/sys v0.46.0 // indirect
+	golang.org/x/text v0.39.0 // indirect
+	golang.org/x/tools v0.47.0 // indirect
 	google.golang.org/protobuf v1.36.11 // indirect
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 	rsc.io/binaryregexp v0.2.0 // indirect
--- a/go.sum
+++ b/go.sum
@@ -186,24 +186,24 @@ go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
 golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
 golang.org/x/arch v0.8.0 h1:3wRIsP3pM4yUptoR96otTUOXI367OS0+c9eeRi9doIc=
 golang.org/x/arch v0.8.0/go.mod h1:FEVrYAQjsQXMVJ1nsMoVVXPZg6p2JE2mx8psSWTDQys=
-golang.org/x/crypto v0.52.0 h1:RMs7fP2rXdep0CftQlK8Uf+kibLm7qkCcradZWYz988=
-golang.org/x/crypto v0.52.0/go.mod h1:1QgfPxDqh0T2M/elOJtp9RvuR95kVjir0e6/BvEmGbc=
-golang.org/x/mod v0.35.0 h1:Ww1D637e6Pg+Zb2KrWfHQUnH2dQRLBQyAtpr/haaJeM=
-golang.org/x/mod v0.35.0/go.mod h1:+GwiRhIInF8wPm+4AoT6L0FA1QWAad3OMdTRx4tFYlU=
-golang.org/x/net v0.55.0 h1:bcvxaJn3e1U6InsFWt1JUq1aSjnRxLzT2rtD2KfkDF8=
-golang.org/x/net v0.55.0/go.mod h1:L5U2KuzuOe1lY7Z+aWVIKK6qEeJXnXV9yzGA+WCHJww=
+golang.org/x/crypto v0.53.0 h1:QZ4Muo8THX6CizN2vPPd5fBGHyogrdK9fG4wLPFUsto=
+golang.org/x/crypto v0.53.0/go.mod h1:DNLU434OwVakk9PzuwV8w62mAJpRJL3vsgcfp4Qnsio=
+golang.org/x/mod v0.37.0 h1:vF1DjpVEshcIqoEaauuHebaLk1O1forxjxBaVn884JQ=
+golang.org/x/mod v0.37.0/go.mod h1:m8S8VeM9r4dzDwjrKO0a1sZP3YjeMamRRlD+fmR2Q/0=
+golang.org/x/net v0.56.0 h1:Rw8j/hFzGvJUZwNBXnAtf5sVDVt+65SK2C7IxCxZt5o=
+golang.org/x/net v0.56.0/go.mod h1:D3Ku6r+V6JROoZK144D2XfMHFcMq/0zSfLelVTCFKec=
 golang.org/x/oauth2 v0.36.0 h1:peZ/1z27fi9hUOFCAZaHyrpWG5lwe0RJEEEeH0ThlIs=
 golang.org/x/oauth2 v0.36.0/go.mod h1:YDBUJMTkDnJS+A4BP4eZBjCqtokkg1hODuPjwiGPO7Q=
-golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
-golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
-golang.org/x/sys v0.45.0 h1:dO4czNzziLiiXplLQgBCEpCvXQ3dnkn0SdaZSYdQ+FY=
-golang.org/x/sys v0.45.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
-golang.org/x/term v0.43.0 h1:S4RLU2sB31O/NCl+zFN9Aru9A/Cq2aqKpTZJ6B+DwT4=
-golang.org/x/term v0.43.0/go.mod h1:lrhlHNdQJHO+1qVYiHfFKVuVioJIheAc3fBSMFYEIsk=
-golang.org/x/text v0.37.0 h1:Cqjiwd9eSg8e0QAkyCaQTNHFIIzWtidPahFWR83rTrc=
-golang.org/x/text v0.37.0/go.mod h1:a5sjxXGs9hsn/AJVwuElvCAo9v8QYLzvavO5z2PiM38=
-golang.org/x/tools v0.44.0 h1:UP4ajHPIcuMjT1GqzDWRlalUEoY+uzoZKnhOjbIPD2c=
-golang.org/x/tools v0.44.0/go.mod h1:KA0AfVErSdxRZIsOVipbv3rQhVXTnlU6UhKxHd1seDI=
+golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
+golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sys v0.46.0 h1:noSf2Fq6F8DBgS+LysIkx7rIExoNHJsxOAtPp4rthXw=
+golang.org/x/sys v0.46.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/term v0.44.0 h1:0rLvDRCtNj0gZkyIXhCyOb2OAzEhLVqc4B+hrsBhrmc=
+golang.org/x/term v0.44.0/go.mod h1:7ze4MdzUzLXpSAoFP1H0bOI9aXDqveSvatT5vKcFh2Y=
+golang.org/x/text v0.39.0 h1:UbZz4pLOvn600D6Oh6GGEI6VAmndrEBLv8/6BEXzyus=
+golang.org/x/text v0.39.0/go.mod h1:3UwRclnC2g0TU9x8PZiyfOajCd1zaUNHF9cvqcQZ+ZM=
+golang.org/x/tools v0.47.0 h1:7Kn5x/d1svx/PzryTsqeoZN4TZwqeH5pGWjefhLi/1Q=
+golang.org/x/tools v0.47.0/go.mod h1:dFHnyTvFWY212G+h7ZY4Vsp/K3U4/7W9TyVaAul8uCA=
 google.golang.org/protobuf v1.36.11 h1:fV6ZwhNocDyBLK0dj+fg8ektcVegBBuEolpbTQyBNVE=
 google.golang.org/protobuf v1.36.11/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
--- a/internal/database/migrations/0044_backend_server_timeout.sql
+++ b/internal/database/migrations/0044_backend_server_timeout.sql
@@ -0,0 +1,27 @@
+-- +goose Up
+-- +goose StatementBegin
+
+-- Per-Backend `timeout server` (Sekunden). NULL = defaults-Timeout (60s,
+-- siehe haproxy.cfg.tpl). Gedacht für Upstreams die LANGE für die Antwort
+-- brauchen und dabei NICHT streamen — z. B. KI-/Inferenz-Server, die eine
+-- gepufferte Antwort erst nach Minuten schicken. Ohne Override kappt der
+-- 60s-defaults-Timeout diese Requests.
+--
+-- Bewusst NULL-per-default: Backends ohne Langläufer-Workload behalten den
+-- kurzen Timeout (Connection-Hygiene / Slowloris-Schutz, vgl. v1.3.2).
+-- Der Renderer setzt `timeout server <N>s` NUR wenn ein Wert gesetzt ist.
+--
+-- CHECK 1..86400: mind. 1s, max. 24h — verhindert 0/negativ (würde HAProxy-
+-- Config sprengen bzw. „unendlich" bedeuten) und absurd hohe Werte.
+ALTER TABLE backends
+    ADD COLUMN IF NOT EXISTS server_timeout_seconds INTEGER
+        CONSTRAINT backends_server_timeout_range
+        CHECK (server_timeout_seconds IS NULL
+               OR (server_timeout_seconds BETWEEN 1 AND 86400));
+
+-- +goose StatementEnd
+
+-- +goose Down
+-- +goose StatementBegin
+ALTER TABLE backends DROP COLUMN IF EXISTS server_timeout_seconds;
+-- +goose StatementEnd
--- a/internal/handlers/cluster_certsync.go
+++ b/internal/handlers/cluster_certsync.go
@@ -109,6 +109,40 @@ func SyncTLSCertsFromPrimary(ctx context.Context, pool *pgxpool.Pool, agg *aggre
 		slog.Info("cert-sync: updated", "file", name)
 	}

+	// Prune: lokale .pem entfernen, die der Primary NICHT (mehr) hat.
+	// Ohne diesen Schritt bleiben Zertifikate gelöschter Domains auf dem
+	// Secondary als Waisen liegen — der Sync oben ist write-only, „nicht
+	// mitgeschickt" ≠ „gelöscht". Geschützt bleiben:
+	//   _default.pem       — Self-Signed-Fallback
+	//   <lokaler-FQDN>.pem  — eigener Node-Cert (steht NICHT im Primary-Payload)
+	// Nur prunen wenn der Payload nicht leer ist — Schutz gegen ein
+	// versehentliches Leerräumen bei unvollständiger Primary-Antwort.
+	if len(payload.Certs) > 0 {
+		protected := map[string]bool{"_default.pem": true}
+		var localFQDN string
+		if err := pool.QueryRow(ctx,
+			`SELECT fqdn FROM ha_nodes WHERE id = $1`, localID).Scan(&localFQDN); err == nil && localFQDN != "" {
+			protected[localFQDN+".pem"] = true
+		}
+		if entries, err := os.ReadDir(tlsCertDir); err == nil {
+			for _, e := range entries {
+				name := e.Name()
+				if e.IsDir() || !strings.HasSuffix(name, ".pem") || protected[name] {
+					continue
+				}
+				if _, ok := payload.Certs[name]; ok {
+					continue // vom Primary gepflegt — behalten
+				}
+				if err := os.Remove(filepath.Join(tlsCertDir, name)); err != nil {
+					slog.Warn("cert-sync: prune failed", "file", name, "error", err)
+					continue
+				}
+				changed = true
+				slog.Info("cert-sync: pruned orphan", "file", name)
+			}
+		}
+	}
+
 	if changed {
 		if err := exec.Command("sudo", "-n", "/usr/bin/systemctl", "reload", "haproxy.service").Run(); err != nil {
 			slog.Warn("cert-sync: haproxy reload failed", "error", err)
--- a/internal/haproxy/haproxy.cfg.tpl
+++ b/internal/haproxy/haproxy.cfg.tpl
@@ -226,6 +226,11 @@ backend eg_backend_{{$b.ID}}
    {{- if $b.WebSocket}}
    timeout tunnel 1h
    {{- end}}
+    {{- if $b.ServerTimeoutSeconds}}
+    # Override des defaults-`timeout server 60s` für langsame Upstreams
+    # (z. B. KI-Server mit gepufferter Antwort). Wert per Backend gepflegt.
+    timeout server {{$b.ServerTimeoutSeconds}}s
+    {{- end}}
    {{- if $b.HealthCheckPath}}
    option httpchk
    http-check send meth GET uri {{$b.HealthCheckPath}}
--- a/internal/haproxy/haproxy_test.go
+++ b/internal/haproxy/haproxy_test.go
@@ -554,6 +554,44 @@ func TestRender_WebSocketEmitsTunnelTimeout(t *testing.T) {
 	}
 }

+func TestRender_ServerTimeoutOverride(t *testing.T) {
+	tmo := 300
+	v := View{
+		Backends: []BackendView{
+			{
+				Backend: models.Backend{ID: 11, Name: "ai", Scheme: "http",
+					LBAlgorithm: "roundrobin", ServerTimeoutSeconds: &tmo, Active: true},
+				Servers: []models.BackendServer{
+					{BackendID: 11, Name: "ai-1", Address: "10.0.5.30", Port: 8000, Weight: 100, Active: true},
+				},
+			},
+			{
+				Backend: models.Backend{ID: 12, Name: "web", Scheme: "http",
+					LBAlgorithm: "roundrobin", Active: true},
+				Servers: []models.BackendServer{
+					{BackendID: 12, Name: "web-1", Address: "10.0.5.31", Port: 8080, Weight: 100, Active: true},
+				},
+			},
+		},
+	}
+	out := renderView(t, v)
+	idxAI := strings.Index(out, "backend eg_backend_11")
+	idxWeb := strings.Index(out, "backend eg_backend_12")
+	if idxAI < 0 || idxWeb < 0 {
+		t.Fatalf("backend sections missing in output:\n%s", out)
+	}
+	aiBlock := out[idxAI:idxWeb]
+	webBlock := out[idxWeb:]
+	// ai (nil-Override gesetzt) soll `timeout server 300s` bekommen …
+	if !strings.Contains(aiBlock, "timeout server 300s") {
+		t.Errorf("ai-Block sollte `timeout server 300s` enthalten:\n%s", aiBlock)
+	}
+	// … web (kein Override) soll KEINE eigene timeout-server-Zeile bekommen.
+	if strings.Contains(webBlock, "timeout server") {
+		t.Errorf("web-Block soll KEIN eigenes `timeout server` enthalten:\n%s", webBlock)
+	}
+}
+
 func TestRender_MultiServerPool(t *testing.T) {
 	v := View{
 		Backends: []BackendView{
--- a/internal/models/backend.go
+++ b/internal/models/backend.go
@@ -12,6 +12,10 @@ type Backend struct {
 	LBAlgorithm     string    `gorm:"column:lb_algorithm"        json:"lb_algorithm"`
 	WebSocket       bool      `gorm:"column:websocket"           json:"websocket"`
 	ForceHTTP1      bool      `gorm:"column:force_http1"         json:"force_http1"`
+	// ServerTimeoutSeconds überschreibt `timeout server` für dieses
+	// Backend (Sekunden). nil = defaults-Timeout (60s). Für langsam
+	// antwortende Upstreams (KI-/Inferenz-Server ohne Streaming).
+	ServerTimeoutSeconds *int  `gorm:"column:server_timeout_seconds" json:"server_timeout_seconds,omitempty"`
 	Active          bool      `gorm:"column:active"              json:"active"`
 	CreatedAt       time.Time `gorm:"column:created_at"          json:"created_at"`
 	UpdatedAt       time.Time `gorm:"column:updated_at"          json:"updated_at"`
--- a/internal/services/alerts/alerts.go
+++ b/internal/services/alerts/alerts.go
@@ -194,6 +194,24 @@ FROM alert_events ORDER BY fired_at DESC, id DESC LIMIT $1`, limit)
 	return out, rows.Err()
 }

+// Cleanup löscht alert_events älter als keepDays und liefert die Anzahl
+// gelöschter Rows. make_interval(days => $1) nimmt $1 sauber als int —
+// der frühere ($1 || ' days')::interval-Ansatz erzwang text und scheiterte
+// unter pgx mit einem Encode-Fehler (vgl. waf PurgeAlerts, v1.3.3).
+func (s *Service) Cleanup(ctx context.Context, keepDays int) (int64, error) {
+	if keepDays <= 0 {
+		return 0, nil
+	}
+	tag, err := s.Pool.Exec(ctx,
+		`DELETE FROM alert_events WHERE fired_at < NOW() - make_interval(days => $1)`,
+		keepDays,
+	)
+	if err != nil {
+		return 0, err
+	}
+	return tag.RowsAffected(), nil
+}
+
 // Fire dispatch'ed einen Event an alle aktiven Channels und persistiert
 // das Ergebnis. Non-fatal — Send-Failures werden im sent_to-JSON
 // dokumentiert, der Event selbst landet in jedem Fall in der History.
--- a/internal/services/backends/backends.go
+++ b/internal/services/backends/backends.go
@@ -26,8 +26,8 @@ type Repo struct {
 func New(pool *pgxpool.Pool) *Repo { return &Repo{Pool: pool} }

 const baseSelect = `
-SELECT id, name, scheme, health_check_path, lb_algorithm, websocket, force_http1, active,
-       created_at, updated_at
+SELECT id, name, scheme, health_check_path, lb_algorithm, websocket, force_http1,
+       server_timeout_seconds, active, created_at, updated_at
 FROM backends
 `

@@ -65,11 +65,13 @@ func (r *Repo) Create(ctx context.Context, b models.Backend) (*models.Backend, e
 		b.LBAlgorithm = "roundrobin"
 	}
 	row := r.Pool.QueryRow(ctx, `
-INSERT INTO backends (name, scheme, health_check_path, lb_algorithm, websocket, force_http1, active)
-VALUES ($1, $2, $3, $4, $5, $6, $7)
-RETURNING id, name, scheme, health_check_path, lb_algorithm, websocket, force_http1, active,
-          created_at, updated_at`,
-		b.Name, b.Scheme, b.HealthCheckPath, b.LBAlgorithm, b.WebSocket, b.ForceHTTP1, b.Active)
+INSERT INTO backends (name, scheme, health_check_path, lb_algorithm, websocket, force_http1,
+                      server_timeout_seconds, active)
+VALUES ($1, $2, $3, $4, $5, $6, $7, $8)
+RETURNING id, name, scheme, health_check_path, lb_algorithm, websocket, force_http1,
+          server_timeout_seconds, active, created_at, updated_at`,
+		b.Name, b.Scheme, b.HealthCheckPath, b.LBAlgorithm, b.WebSocket, b.ForceHTTP1,
+		b.ServerTimeoutSeconds, b.Active)
 	return scanBackend(row)
 }

@@ -85,12 +87,14 @@ UPDATE backends SET
   lb_algorithm = $4,
   websocket = $5,
   force_http1 = $6,
-   active = $7,
+   server_timeout_seconds = $7,
+   active = $8,
   updated_at = NOW()
-WHERE id = $8
-RETURNING id, name, scheme, health_check_path, lb_algorithm, websocket, force_http1, active,
-          created_at, updated_at`,
-		b.Name, b.Scheme, b.HealthCheckPath, b.LBAlgorithm, b.WebSocket, b.ForceHTTP1, b.Active, id)
+WHERE id = $9
+RETURNING id, name, scheme, health_check_path, lb_algorithm, websocket, force_http1,
+          server_timeout_seconds, active, created_at, updated_at`,
+		b.Name, b.Scheme, b.HealthCheckPath, b.LBAlgorithm, b.WebSocket, b.ForceHTTP1,
+		b.ServerTimeoutSeconds, b.Active, id)
 	out, err := scanBackend(row)
 	if err != nil {
 		if errors.Is(err, pgx.ErrNoRows) {
@@ -125,7 +129,8 @@ func scanBackend(row interface{ Scan(...any) error }) (*models.Backend, error) {
 	var b models.Backend
 	if err := row.Scan(
 		&b.ID, &b.Name, &b.Scheme,
-		&b.HealthCheckPath, &b.LBAlgorithm, &b.WebSocket, &b.ForceHTTP1, &b.Active,
+		&b.HealthCheckPath, &b.LBAlgorithm, &b.WebSocket, &b.ForceHTTP1,
+		&b.ServerTimeoutSeconds, &b.Active,
 		&b.CreatedAt, &b.UpdatedAt,
 	); err != nil {
 		return nil, err
--- a/management-ui/src/i18n/locales/de/common.json
+++ b/management-ui/src/i18n/locales/de/common.json
@@ -592,6 +592,8 @@
    "websocketHint": "An: erlaubt langlebige WebSocket-/Long-Poll-Verbindungen (z. B. Proxmox-Console, SSH-WS, AsyncAPI) — Tunnel-Idle 1h statt 60s. Aus: strikte HTTP-Timeouts.",
    "forceHttp1": "HTTP/1.1 erzwingen",
    "forceHttp1Hint": "Deaktiviert HTTP/2 (h2) auf der Backend-Verbindung — HAProxy handelt nur noch HTTP/1.1 aus. Nötig für Backends die kein h2 unterstützen (z. B. ältere nginx-Konfigurationen ohne h2-Modul, Legacy-Apps).",
+    "serverTimeout": "Antwort-Timeout (timeout server)",
+    "serverTimeoutHint": "Wie lange HAProxy auf die Antwort dieses Backends wartet, bevor es abbricht. Leer = Default (60s). Höher setzen für langsame Upstreams die NICHT streamen (z. B. KI-/Inferenz-Server mit gepufferter Antwort). Achtung: gilt als Inaktivitäts-Timeout — streamende Backends (SSE/chunked) brauchen das meist nicht.",
    "servers": "Server",
    "noServers": "kein Server",
    "noServersWarning": "Ein oder mehrere aktive Backends haben keine Server konfiguriert.",
--- a/management-ui/src/i18n/locales/en/common.json
+++ b/management-ui/src/i18n/locales/en/common.json
@@ -592,6 +592,8 @@
    "websocketHint": "On: allow long-lived WebSocket / long-poll connections (Proxmox console, SSH-over-WS, AsyncAPI) — tunnel idle 1h instead of 60s. Off: strict HTTP timeouts.",
    "forceHttp1": "Force HTTP/1.1",
    "forceHttp1Hint": "Disables HTTP/2 (h2) on the backend connection — HAProxy negotiates HTTP/1.1 only. Required for backends that don't support h2 (e.g. older nginx configs without the h2 module, legacy apps).",
+    "serverTimeout": "Response timeout (timeout server)",
+    "serverTimeoutHint": "How long HAProxy waits for this backend's response before aborting. Empty = default (60s). Raise it for slow upstreams that do NOT stream (e.g. AI/inference servers with a buffered response). Note: this is an inactivity timeout — streaming backends (SSE/chunked) usually don't need it.",
    "servers": "Servers",
    "noServers": "no server",
    "noServersWarning": "One or more active backends have no servers configured.",
--- a/management-ui/src/pages/Alerts/index.tsx
+++ b/management-ui/src/pages/Alerts/index.tsx
@@ -8,6 +8,7 @@ import {
 } from '@ant-design/icons'
 import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
 import { useTranslation } from 'react-i18next'
+import { useSearchParams } from 'react-router-dom'
 import dayjs from 'dayjs'

 import apiClient, { isEnvelope } from '../../api/client'
@@ -69,6 +70,14 @@ export default function AlertsPage() {
  const qc = useQueryClient()
  const isViewer = useAuthStore((s) => s.user?.role) === 'viewer'

+  // Tab über ?tab= steuerbar, damit der Dashboard-Deeplink
+  // (/alerts?tab=events) direkt auf der Event-History landet statt auf
+  // dem Channels-Tab. Default bleibt 'channels' für den Direktaufruf.
+  const [searchParams, setSearchParams] = useSearchParams()
+  const activeTab = searchParams.get('tab') === 'events' ? 'events' : 'channels'
+  const setActiveTab = (key: string) =>
+    setSearchParams(key === 'events' ? { tab: 'events' } : {}, { replace: true })
+
  const channels = useQuery({
    queryKey: ['alerts', 'channels'],
    queryFn: async () => {
@@ -304,7 +313,7 @@ export default function AlertsPage() {
        description={t('alerts.scopeDesc')}
      />

-      <Tabs items={[
+      <Tabs activeKey={activeTab} onChange={setActiveTab} items={[
        {
          key: 'channels',
          label: t('alerts.tabs.channels'),
--- a/management-ui/src/pages/Backends/Detail.tsx
+++ b/management-ui/src/pages/Backends/Detail.tsx
@@ -20,13 +20,17 @@ interface Backend {
  id: number; name: string; scheme: string
  health_check_path?: string | null
  lb_algorithm: 'roundrobin' | 'leastconn' | 'source'
-  websocket: boolean; force_http1: boolean; active: boolean
+  websocket: boolean; force_http1: boolean
+  server_timeout_seconds?: number | null
+  active: boolean
 }
 interface BackendFormValues {
  name: string; scheme: 'http' | 'https'
  health_check_path?: string
  lb_algorithm: 'roundrobin' | 'leastconn' | 'source'
-  websocket: boolean; force_http1: boolean; active: boolean
+  websocket: boolean; force_http1: boolean
+  server_timeout_seconds?: number | null
+  active: boolean
  domain_ids?: number[]
 }
 interface BackendServer {
@@ -174,6 +178,7 @@ export default function BackendDetailPage() {
                lb_algorithm: backend.lb_algorithm,
                websocket: backend.websocket,
                force_http1: backend.force_http1,
+                server_timeout_seconds: backend.server_timeout_seconds ?? undefined,
                active: backend.active,
                domain_ids: attached.map(d => d.id),
              }}
@@ -205,6 +210,11 @@ export default function BackendDetailPage() {
                extra={t('backends.forceHttp1Hint')}>
                <Switch />
              </Form.Item>
+              <Form.Item label={t('backends.serverTimeout')} name="server_timeout_seconds"
+                extra={t('backends.serverTimeoutHint')}>
+                <InputNumber min={1} max={86400} step={30}
+                  style={{ width: '100%' }} addonAfter="s" placeholder="60 (default)" />
+              </Form.Item>
              <Form.Item label={t('backends.active')} name="active" valuePropName="checked">
                <Switch />
              </Form.Item>
--- a/management-ui/src/pages/Backends/index.tsx
+++ b/management-ui/src/pages/Backends/index.tsx
@@ -26,6 +26,7 @@ interface Backend {
  lb_algorithm: 'roundrobin' | 'leastconn' | 'source'
  websocket: boolean
  force_http1: boolean
+  server_timeout_seconds?: number | null
  active: boolean
  created_at: string
  updated_at: string
@@ -266,6 +267,7 @@ export default function BackendsPage() {
        <Space size={4}>
          <Tag>{v}</Tag>
          {row.websocket && <Tag color="cyan">WS</Tag>}
+          {row.server_timeout_seconds ? <Tag color="gold">{row.server_timeout_seconds}s</Tag> : null}
        </Space>
      ),
    },
--- a/management-ui/src/pages/Dashboard/index.tsx
+++ b/management-ui/src/pages/Dashboard/index.tsx
@@ -361,7 +361,7 @@ export default function DashboardPage() {
              </span>
            </Space>
          }
-          action={<Link to="/alerts" style={{ fontSize: 12 }}>{t('dashboard.alertsCard.viewAll')} →</Link>}
+          action={<Link to="/alerts?tab=events" style={{ fontSize: 12 }}>{t('dashboard.alertsCard.viewAll')} →</Link>}
        />
      )}
@@ -1 +1 @@
 .3.3
 .3.4