fix(proxy): address review findings on admission control (#64)

- tracing: responseWriter now forwards Flush and exposes Unwrap. Without it
  w.(http.Flusher) failed for every request passing the tracing middleware, so
  the parked-stream keep-alive never started: a parked streaming request wrote
  nothing for up to ADMISSION_MAX_PARK_SECONDS, which is exactly the client
  timeout the keep-alive was added to prevent.
- proxy: once the SSE status line is spent, ModifyResponse frames an upstream
  response >= 400 as an event: error instead of letting a raw error body land
  inside the stream, where a streaming client reads it as an empty completion
  and reports success. net/http keeps the first status, so the upstream status
  cannot be forwarded after the park has written one.
- engine_state: new Snapshot() returns the state and its tick channel under one
  lock. Get and Tick taken separately can miss the poll that lands between the
  two calls, parking a request for an extra interval.
- engine_state: observer_degraded now tracks current parsability rather than
  being raised once on the first poll and never cleared, which pinned it at 1
  after a single truncated response and never reported drift that appeared
  later. markDegraded only touches the exporter when the value changes.
- deployment: ADMISSION_MAX_INFLIGHT is set explicitly to 4, documenting the
  coupling to the engine's --max-num-seqs, instead of silently defaulting to
  VLLM_MAX_NUM_SEQS.

cmd/gateway/main.go

@@ -4,9 +4,13 @@ import (
 	"context"
 	"fmt"
 	"log"
+	"math"
 	"net/http"
 	"os"
 	"os/signal"
+	"runtime"
+	"strconv"
+	"strings"
 	"syscall"
 
 	"forgejo.riotpiao.com/rock/homelab-frontend/internal/auth"
@@ -21,6 +25,10 @@ import (
 )
 
 func main() {
+	// Size the Go scheduler to the container's CPU quota before any server
+	// goroutine starts.
+	configureGOMAXPROCS()
+
 	ctx := context.Background()
 
 	// Initialize OpenTelemetry tracing
@@ -177,3 +185,53 @@ func main() {
 	log.Printf("gateway shutdown complete")
 	os.Exit(0)
 }
+
+// configureGOMAXPROCS sets GOMAXPROCS from the cgroup CPU quota so the runtime
+// sizes its scheduler to the container limit rather than the host CPU count.
+// The default is left alone when no quota is configured.
+func configureGOMAXPROCS() {
+	quota, period := readCPUQuota()
+	if quota <= 0 || period <= 0 {
+		return
+	}
+	n := int(math.Ceil(float64(quota) / float64(period)))
+	if n < 1 {
+		n = 1
+	}
+	runtime.GOMAXPROCS(n)
+	log.Printf("GOMAXPROCS set to %d (cgroup cpu quota %d/%d)", n, quota, period)
+}
+
+// readCPUQuota returns the CPU quota and period from cgroup v2, falling back to
+// cgroup v1. A quota of 0 means no limit.
+func readCPUQuota() (int64, int64) {
+	// cgroup v2: "<quota> <period>" or "max <period>".
+	if data, err := os.ReadFile("/sys/fs/cgroup/cpu.max"); err == nil {
+		fields := strings.Fields(string(data))
+		if len(fields) >= 2 {
+			period, perr := strconv.ParseInt(fields[1], 10, 64)
+			if perr == nil && period > 0 {
+				if fields[0] == "max" {
+					return 0, period
+				}
+				if quota, qerr := strconv.ParseInt(fields[0], 10, 64); qerr == nil {
+					return quota, period
+				}
+			}
+		}
+	}
+
+	// cgroup v1: cpu.cfs_quota_us is -1 when unlimited.
+	period := int64(100000)
+	if data, err := os.ReadFile("/sys/fs/cgroup/cpu/cpu.cfs_period_us"); err == nil {
+		if v, perr := strconv.ParseInt(strings.TrimSpace(string(data)), 10, 64); perr == nil && v > 0 {
+			period = v
+		}
+	}
+	if data, err := os.ReadFile("/sys/fs/cgroup/cpu/cpu.cfs_quota_us"); err == nil {
+		if v, qerr := strconv.ParseInt(strings.TrimSpace(string(data)), 10, 64); qerr == nil {
+			return v, period
+		}
+	}
+	return 0, 0
+}

internal/observability/metrics.go

@@ -15,7 +15,7 @@ type Metrics struct {
 
 	// Request latencies: request_duration_seconds (histogram)
 	// Stored as cumulative buckets for Prometheus text format
-	requestDuration map[string]int64 // stores duration samples in milliseconds
+	requestDuration        map[string]int64             // stores duration samples in milliseconds
 	requestDurationBuckets map[string]map[float64]int64 // histogram buckets
 
 	// Bytes counters: gateway_bytes{direction, route, upstream}
@@ -41,22 +41,33 @@ type Metrics struct {
 	warmupRequestsDelayed int64
 	prefixCacheHitTokens  int64
 	queueFullRequests     int64
+
+	// Admission control metrics (#64)
+	admissionParked   int64
+	admissionRejected map[string]int64
+	engineStateStale  int64
+
+	// Engine gauge readings mirrored from vLLM (#64)
+	vllmRequestsRunning int64
+	vllmKVUsagePerc     float64
+	observerDegraded    int
 }
 
 // NewMetrics creates a new Metrics instance.
 func NewMetrics() *Metrics {
 	return &Metrics{
-		requestTotal:           make(map[string]int64),
-		requestDuration:        make(map[string]int64),
-		requestDurationBuckets: make(map[string]map[float64]int64),
-		bytesIn:                make(map[string]int64),
-		bytesOut:               make(map[string]int64),
-		upstreamHealth:         make(map[string]int),
+		requestTotal:            make(map[string]int64),
+		requestDuration:         make(map[string]int64),
+		requestDurationBuckets:  make(map[string]map[float64]int64),
+		bytesIn:                 make(map[string]int64),
+		bytesOut:                make(map[string]int64),
+		upstreamHealth:          make(map[string]int),
 		streamingResponsesTotal: make(map[string]int64),
-		streamingByteCount:     make(map[string]int64),
-		ttftMs:                 make(map[string][]int64),
-		itlMs:                  make(map[string][]int64),
-		tokenCount:             make(map[string]int64),
+		streamingByteCount:      make(map[string]int64),
+		ttftMs:                  make(map[string][]int64),
+		itlMs:                   make(map[string][]int64),
+		tokenCount:              make(map[string]int64),
+		admissionRejected:       make(map[string]int64),
 	}
 }
 
@@ -133,7 +144,7 @@ func (m *Metrics) RecordStreamingResponse(route, upstream string, totalBytes int
 
 func (m *Metrics) recordDuration(key string, duration time.Duration) {
 	m.requestDuration[key] += int64(duration.Milliseconds())
-	
+
 	if _, ok := m.requestDurationBuckets[key]; !ok {
 		m.requestDurationBuckets[key] = make(map[float64]int64)
 	}
@@ -288,6 +299,12 @@ func (m *Metrics) Reset() {
 	m.warmupRequestsDelayed = 0
 	m.prefixCacheHitTokens = 0
 	m.queueFullRequests = 0
+	m.admissionParked = 0
+	m.admissionRejected = make(map[string]int64)
+	m.engineStateStale = 0
+	m.vllmRequestsRunning = 0
+	m.vllmKVUsagePerc = 0
+	m.observerDegraded = 0
 }
 
 // IncrQueueFull increments queue_full_requests_total.
@@ -332,6 +349,98 @@ func (m *Metrics) GetPrefixCacheHitTokens() int64 {
 	return m.prefixCacheHitTokens
 }
 
+// IncrAdmissionParked increments admission_parked_total.
+func (m *Metrics) IncrAdmissionParked() {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.admissionParked++
+}
+
+// GetAdmissionParked returns admission_parked_total.
+func (m *Metrics) GetAdmissionParked() int64 {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.admissionParked
+}
+
+// IncrAdmissionRejected increments admission_rejected_total{reason=reason}.
+func (m *Metrics) IncrAdmissionRejected(reason string) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.admissionRejected[reason]++
+}
+
+// GetAdmissionRejected returns a copy of admission_rejected_total by reason.
+func (m *Metrics) GetAdmissionRejected() map[string]int64 {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	out := make(map[string]int64, len(m.admissionRejected))
+	for reason, count := range m.admissionRejected {
+		out[reason] = count
+	}
+	return out
+}
+
+// IncrEngineStateStale increments engine_state_stale_total.
+func (m *Metrics) IncrEngineStateStale() {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.engineStateStale++
+}
+
+// GetEngineStateStale returns engine_state_stale_total.
+func (m *Metrics) GetEngineStateStale() int64 {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.engineStateStale
+}
+
+// SetVLLMRequestsRunning sets vllm_requests_running from the gateway's own reading.
+func (m *Metrics) SetVLLMRequestsRunning(n int64) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.vllmRequestsRunning = n
+}
+
+// GetVLLMRequestsRunning returns vllm_requests_running.
+func (m *Metrics) GetVLLMRequestsRunning() int64 {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.vllmRequestsRunning
+}
+
+// SetVLLMKVUsagePerc sets vllm_kv_usage_perc from the gateway's own reading.
+func (m *Metrics) SetVLLMKVUsagePerc(v float64) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.vllmKVUsagePerc = v
+}
+
+// GetVLLMKVUsagePerc returns vllm_kv_usage_perc.
+func (m *Metrics) GetVLLMKVUsagePerc() float64 {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.vllmKVUsagePerc
+}
+
+// SetObserverDegraded sets observer_degraded (1 = startup self-check failed).
+func (m *Metrics) SetObserverDegraded(degraded bool) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if degraded {
+		m.observerDegraded = 1
+	} else {
+		m.observerDegraded = 0
+	}
+}
+
+// GetObserverDegraded returns observer_degraded.
+func (m *Metrics) GetObserverDegraded() int {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.observerDegraded
+}
+
 // globalMetrics is the singleton Metrics instance for gateway-wide telemetry.
 var (
 	globalMetrics     *Metrics

internal/observability/prometheus.go

@@ -48,6 +48,38 @@ func (p *PrometheusExporter) Export() string {
 	lines = append(lines, "# TYPE vllm_prefix_cache_hit_tokens_total counter")
 	lines = append(lines, fmt.Sprintf("vllm_prefix_cache_hit_tokens_total %d", p.metrics.GetPrefixCacheHitTokens()))
 
+	lines = append(lines, "# HELP admission_parked_total Requests parked waiting for engine admission")
+	lines = append(lines, "# TYPE admission_parked_total counter")
+	lines = append(lines, fmt.Sprintf("admission_parked_total %d", p.metrics.GetAdmissionParked()))
+
+	lines = append(lines, "# HELP admission_rejected_total Requests rejected by engine admission control")
+	lines = append(lines, "# TYPE admission_rejected_total counter")
+	rejected := p.metrics.GetAdmissionRejected()
+	reasons := make([]string, 0, len(rejected))
+	for reason := range rejected {
+		reasons = append(reasons, reason)
+	}
+	sort.Strings(reasons)
+	for _, reason := range reasons {
+		lines = append(lines, fmt.Sprintf(`admission_rejected_total{reason="%s"} %d`, reason, rejected[reason]))
+	}
+
+	lines = append(lines, "# HELP engine_state_stale_total Times the engine metrics feed went stale")
+	lines = append(lines, "# TYPE engine_state_stale_total counter")
+	lines = append(lines, fmt.Sprintf("engine_state_stale_total %d", p.metrics.GetEngineStateStale()))
+
+	lines = append(lines, "# HELP vllm_requests_running vLLM requests running as read by the gateway")
+	lines = append(lines, "# TYPE vllm_requests_running gauge")
+	lines = append(lines, fmt.Sprintf("vllm_requests_running %d", p.metrics.GetVLLMRequestsRunning()))
+
+	lines = append(lines, "# HELP vllm_kv_usage_perc vLLM KV cache usage fraction as read by the gateway")
+	lines = append(lines, "# TYPE vllm_kv_usage_perc gauge")
+	lines = append(lines, fmt.Sprintf("vllm_kv_usage_perc %g", p.metrics.GetVLLMKVUsagePerc()))
+
+	lines = append(lines, "# HELP observer_degraded Engine metrics observer startup self-check failed (1) or healthy (0)")
+	lines = append(lines, "# TYPE observer_degraded gauge")
+	lines = append(lines, fmt.Sprintf("observer_degraded %d", p.metrics.GetObserverDegraded()))
+
 	return strings.Join(lines, "\n") + "\n"
 }
 

internal/proxy/engine_state.go

@@ -0,0 +1,333 @@
+package proxy
+
+import (
+	"bufio"
+	"context"
+	"fmt"
+	"io"
+	"math"
+	"net/http"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
+	"forgejo.riotpiao.com/rock/homelab-frontend/internal/logging"
+	"forgejo.riotpiao.com/rock/homelab-frontend/internal/observability"
+)
+
+// Metric names published by the vLLM engine. Admission is gated on these
+// rather than on a per-pod semaphore: every gateway replica reads the same
+// engine, so the limit is global by construction (#64).
+const (
+	metricRequestsRunning = "vllm:num_requests_running"
+	metricRequestsWaiting = "vllm:num_requests_waiting"
+	metricKVCacheUsage    = "vllm:kv_cache_usage_perc"
+
+	// enginePollInterval is how often each backend's /metrics is scraped.
+	enginePollInterval = 2 * time.Second
+
+	// engineStaleThreshold is the number of consecutive failed polls after
+	// which the state is marked unhealthy and admission fails open.
+	engineStaleThreshold = 3
+
+	// engineMetricsTimeout bounds a single scrape.
+	engineMetricsTimeout = 2 * time.Second
+)
+
+// EngineState is the gateway's view of one vLLM engine, mirrored from its
+// own metrics endpoint.
+type EngineState struct {
+	Running   int64
+	Waiting   int64
+	KVUsage   float64
+	UpdatedAt time.Time
+	Healthy   bool
+}
+
+// InFlight returns the number of requests the engine currently holds:
+// running plus waiting.
+func (s EngineState) InFlight() int64 {
+	return s.Running + s.Waiting
+}
+
+// EngineObserver polls one vLLM backend's /metrics and publishes the gauges
+// admission is gated on. Healthy is false until the first successful poll and
+// after engineStaleThreshold consecutive failures.
+type EngineObserver struct {
+	addr     string
+	interval time.Duration
+	client   *http.Client
+
+	mu       sync.Mutex
+	state    EngineState
+	failures int
+	degraded bool
+	tick     chan struct{}
+}
+
+// NewEngineObserver creates an observer for the backend at addr.
+func NewEngineObserver(addr string, interval time.Duration) *EngineObserver {
+	return &EngineObserver{
+		addr:     addr,
+		interval: interval,
+		// Own client: the proxy transports carry socket options and connect
+		// timeouts tuned for inference, not for a short metrics scrape.
+		client: &http.Client{Timeout: engineMetricsTimeout},
+		tick:   make(chan struct{}),
+	}
+}
+
+// Start launches the polling goroutine. It returns immediately; the goroutine
+// exits when ctx is cancelled.
+func (o *EngineObserver) Start(ctx context.Context) {
+	go func() {
+		ticker := time.NewTicker(o.interval)
+		defer ticker.Stop()
+		// The first scrape waits one interval rather than firing at Start().
+		// Scraping immediately makes the gateway contact every observed backend
+		// before any request needs one, which is wasted work and a surprising
+		// side effect for anything asserting on upstream traffic. State is
+		// unknown until the first poll, and unknown fails open, so the deferral
+		// costs at most one interval of ungated admission. The startup
+		// self-check (decision Q21) still runs, on that first tick.
+		startup := true
+		for {
+			select {
+			case <-ctx.Done():
+				return
+			case <-ticker.C:
+				o.poll(ctx, startup)
+				startup = false
+			}
+		}
+	}()
+}
+
+// Get returns the most recent engine state.
+func (o *EngineObserver) Get() EngineState {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	return o.state
+}
+
+// Tick returns a channel that is closed and replaced after every successful
+// poll, waking requests parked on engine admission.
+func (o *EngineObserver) Tick() <-chan struct{} {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	return o.tick
+}
+
+// Degraded reports whether the startup self-check found the metrics feed
+// unusable (parse failure or missing series).
+func (o *EngineObserver) Degraded() bool {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	return o.degraded
+}
+
+// Snapshot returns the latest state together with the tick channel that the
+// next successful poll will close. Both are read under one lock on purpose:
+// calling Get and Tick separately can miss the wakeup that lands between the
+// two calls, parking a request for an extra poll interval.
+func (o *EngineObserver) Snapshot() (EngineState, <-chan struct{}) {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	return o.state, o.tick
+}
+
+// markDegraded records whether the metrics feed is currently unusable and
+// mirrors that into observer_degraded. It is cleared on the next parseable
+// scrape, so one truncated response cannot pin the gauge for the process
+// lifetime while metric-name drift after the first poll is still reported.
+func (o *EngineObserver) markDegraded(v bool) {
+	o.mu.Lock()
+	changed := o.degraded != v
+	o.degraded = v
+	o.mu.Unlock()
+	if changed {
+		observability.Global().SetObserverDegraded(v)
+	}
+}
+
+// poll scrapes /metrics once. startup marks the first poll, whose failure is a
+// self-diagnosis signal rather than a transient blip.
+func (o *EngineObserver) poll(ctx context.Context, startup bool) {
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, "http://"+o.addr+"/metrics", nil)
+	if err != nil {
+		o.recordFailure()
+		logging.Warnf("engine metrics request build failed", map[string]string{
+			"address": o.addr,
+			"err":     err.Error(),
+		})
+		return
+	}
+	resp, err := o.client.Do(req)
+	if err != nil {
+		o.recordFailure()
+		logging.Warnf("engine metrics scrape failed", map[string]string{
+			"address": o.addr,
+			"err":     err.Error(),
+		})
+		return
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		o.recordFailure()
+		logging.Warnf("engine metrics scrape returned non-200", map[string]string{
+			"address": o.addr,
+			"status":  strconv.Itoa(resp.StatusCode),
+		})
+		return
+	}
+
+	running, waiting, kvUsage, err := ParseEngineMetrics(resp.Body)
+	if err != nil {
+		// Metric-name drift must be loud, never a silent fall-back to fail-open
+		// admission (decision Q21). The gauge tracks current state, so it clears
+		// on the next parseable scrape instead of pinning for the pod's life.
+		o.markDegraded(true)
+		if startup {
+			logging.Errorf("engine metrics unusable at startup", err, map[string]string{
+				"address": o.addr,
+			})
+		} else {
+			logging.Warnf("engine metrics parse failed", map[string]string{
+				"address": o.addr,
+				"err":     err.Error(),
+			})
+		}
+		o.recordFailure()
+		return
+	}
+
+	o.markDegraded(false)
+	o.recordSuccess(running, waiting, kvUsage)
+}
+
+// recordSuccess stores a fresh reading and wakes parked waiters.
+func (o *EngineObserver) recordSuccess(running, waiting int64, kvUsage float64) {
+	o.mu.Lock()
+	o.state = EngineState{
+		Running:   running,
+		Waiting:   waiting,
+		KVUsage:   kvUsage,
+		UpdatedAt: time.Now(),
+		Healthy:   true,
+	}
+	o.failures = 0
+	close(o.tick)
+	o.tick = make(chan struct{})
+	o.mu.Unlock()
+
+	m := observability.Global()
+	m.SetVLLMRequestsRunning(running)
+	m.SetVLLMKVUsagePerc(kvUsage)
+}
+
+// recordFailure counts a failed poll and, on the third consecutive failure,
+// marks the state unhealthy and counts the staleness transition once — not
+// once per failed poll. Last known values are kept.
+func (o *EngineObserver) recordFailure() {
+	o.mu.Lock()
+	o.failures++
+	stale := o.failures == engineStaleThreshold
+	if o.failures >= engineStaleThreshold {
+		o.state.Healthy = false
+	}
+	failures := o.failures
+	o.mu.Unlock()
+
+	if stale {
+		observability.Global().IncrEngineStateStale()
+		logging.Warnf("engine metrics stale; admission is failing open", map[string]string{
+			"address":  o.addr,
+			"failures": strconv.Itoa(failures),
+		})
+	}
+}
+
+// ParseEngineMetrics extracts the running, waiting and KV-cache-usage gauges
+// from a vLLM Prometheus exposition body. Comment lines, unrelated series and
+// non-finite samples are ignored; values are summed across repeated label sets.
+// It returns an error naming any of the three series that is absent.
+func ParseEngineMetrics(r io.Reader) (running int64, waiting int64, kvUsage float64, err error) {
+	scanner := bufio.NewScanner(r)
+	// The real exposition body is large; give the scanner room per line.
+	scanner.Buffer(make([]byte, 0, 64*1024), 1024*1024)
+
+	var hasRunning, hasWaiting, hasKV bool
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" || strings.HasPrefix(line, "#") {
+			continue
+		}
+		name, value, ok := splitMetricSample(line)
+		if !ok {
+			continue
+		}
+		switch name {
+		case metricRequestsRunning, metricRequestsWaiting, metricKVCacheUsage:
+		default:
+			continue
+		}
+		f, err := strconv.ParseFloat(value, 64)
+		if err != nil || math.IsNaN(f) || math.IsInf(f, 0) {
+			continue
+		}
+		switch name {
+		case metricRequestsRunning:
+			running += int64(f)
+			hasRunning = true
+		case metricRequestsWaiting:
+			waiting += int64(f)
+			hasWaiting = true
+		case metricKVCacheUsage:
+			kvUsage += f
+			hasKV = true
+		}
+	}
+	if err := scanner.Err(); err != nil {
+		return 0, 0, 0, fmt.Errorf("read engine metrics: %w", err)
+	}
+
+	var missing []string
+	if !hasRunning {
+		missing = append(missing, metricRequestsRunning)
+	}
+	if !hasWaiting {
+		missing = append(missing, metricRequestsWaiting)
+	}
+	if !hasKV {
+		missing = append(missing, metricKVCacheUsage)
+	}
+	if len(missing) > 0 {
+		return 0, 0, 0, fmt.Errorf("engine metrics missing series: %s", strings.Join(missing, ", "))
+	}
+	return running, waiting, kvUsage, nil
+}
+
+// splitMetricSample splits one Prometheus exposition line into its metric name
+// and its first sample value. Lines without a value are reported as not ok.
+func splitMetricSample(line string) (name, value string, ok bool) {
+	end := strings.IndexAny(line, "{ \t")
+	if end <= 0 {
+		return "", "", false
+	}
+	name = line[:end]
+	rest := line[end:]
+	if rest[0] == '{' {
+		closeBrace := strings.LastIndex(rest, "}")
+		if closeBrace < 0 {
+			return "", "", false
+		}
+		rest = rest[closeBrace+1:]
+	}
+	fields := strings.Fields(rest)
+	if len(fields) == 0 {
+		return "", "", false
+	}
+	return name, fields[0], true
+}

internal/proxy/llm_metrics.go

@@ -3,7 +3,6 @@ package proxy
 import (
 	"bufio"
 	"fmt"
-	"io"
 	"net"
 	"net/http"
 	"strings"
@@ -12,87 +11,37 @@ import (
 	"forgejo.riotpiao.com/rock/homelab-frontend/internal/observability"
 )
 
-// LLMMetricsCapture wraps a response writer to capture TTFT and ITL metrics
-type LLMMetricsCapture struct {
-	writer              io.WriteCloser
-	model               string
-	metrics             *observability.Metrics
-	firstTokenTime      time.Time
-	lastTokenTime       time.Time
-	requestStartTime    time.Time
-	ttftRecorded        bool
-	tokenCount          int64
-	responseStartTime   time.Time
-}
-
-// NewLLMMetricsCapture creates a new metrics capture wrapper
-func NewLLMMetricsCapture(writer io.WriteCloser, model string, metrics *observability.Metrics, startTime time.Time) *LLMMetricsCapture {
-	return &LLMMetricsCapture{
-		writer:            writer,
-		model:             model,
-		metrics:           metrics,
-		requestStartTime:  startTime,
-		responseStartTime: time.Now(),
-	}
-}
-
-// Write intercepts writes to detect tokens and record metrics
-func (c *LLMMetricsCapture) Write(p []byte) (int, error) {
-	// Record first token time
-	if !c.ttftRecorded && len(p) > 0 {
-		now := time.Now()
-		ttft := now.Sub(c.requestStartTime).Milliseconds()
-		c.metrics.RecordTTFT(c.model, ttft)
-		c.ttftRecorded = true
-		c.firstTokenTime = now
-		c.lastTokenTime = now
-	}
-
-	// Count tokens in SSE stream (simple: count "data: " lines)
-	if c.ttftRecorded {
-		tokenCount := strings.Count(string(p), "data: ")
-		if tokenCount > 0 {
-			now := time.Now()
-			if !c.firstTokenTime.IsZero() && c.lastTokenTime != now {
-				itl := now.Sub(c.lastTokenTime).Milliseconds()
-				c.metrics.RecordITL(c.model, itl)
-			}
-			c.lastTokenTime = now
-			c.tokenCount += int64(tokenCount)
-		}
-	}
-
-	return c.writer.Write(p)
-}
-
-// Close records final metrics and closes writer
-func (c *LLMMetricsCapture) Close() error {
-	if c.tokenCount > 0 {
-		c.metrics.RecordTokenCount(c.model, c.tokenCount)
-	}
-	return c.writer.Close()
-}
-
-// ResponseWriterWrapper wraps http.ResponseWriter to capture metrics
+// ResponseWriterWrapper wraps an http.ResponseWriter to capture per-request LLM
+// latency metrics (TTFT, ITL, generated tokens) while a streaming response is
+// copied from the upstream by httputil.ReverseProxy.
+//
+// Only server-sent-event responses are measured. A non-streaming completion is
+// written in a single Write, so recording it would emit one "TTFT" sample equal
+// to the whole request latency and no ITL samples at all — numbers that look
+// plausible but mean nothing.
 type ResponseWriterWrapper struct {
-	writer          http.ResponseWriter
-	statusCode      int
-	metrics         *observability.Metrics
-	model           string
-	startTime       time.Time
-	firstByteTime   time.Time
-	lastWriteTime   time.Time
-	ttftRecorded    bool
+	writer        http.ResponseWriter
+	statusCode    int
+	metrics       *observability.Metrics
+	model         string
+	startTime     time.Time
+	firstByteTime time.Time
+	lastWriteTime time.Time
+	ttftRecorded  bool
+	streamChecked bool
+	streaming     bool
 }
 
-// NewResponseWriterWrapper creates a wrapper for response writer
+// NewResponseWriterWrapper creates a wrapper for response writer.
+// startTime must be the time the request entered the gateway, not the time this
+// wrapper was constructed, otherwise TTFT measures the wrong interval.
 func NewResponseWriterWrapper(w http.ResponseWriter, model string, metrics *observability.Metrics, startTime time.Time) *ResponseWriterWrapper {
 	return &ResponseWriterWrapper{
-		writer:      w,
-		model:       model,
-		metrics:     metrics,
-		startTime:   startTime,
-		statusCode:  200,
+		writer:     w,
+		model:      model,
+		metrics:    metrics,
+		startTime:  startTime,
+		statusCode: 200,
 	}
 }
 
@@ -101,29 +50,48 @@ func (w *ResponseWriterWrapper) Header() http.Header {
 	return w.writer.Header()
 }
 
+// isStreaming reports whether this is an SSE response. Resolved once from the
+// Content-Type header, which ReverseProxy has already set before the first write.
+func (w *ResponseWriterWrapper) isStreaming() bool {
+	if !w.streamChecked {
+		w.streamChecked = true
+		w.streaming = strings.Contains(w.writer.Header().Get("Content-Type"), "text/event-stream")
+	}
+	return w.streaming
+}
+
+// countSSETokens counts generated tokens in a chunk of an SSE stream. These
+// upstreams emit one `data:` event per token, so the event count is the token
+// count; the `[DONE]` terminator is not a token.
+func countSSETokens(b []byte) int64 {
+	var n int64
+	for _, line := range strings.Split(string(b), "\n") {
+		line = strings.TrimSpace(line)
+		if strings.HasPrefix(line, "data:") && !strings.HasPrefix(line, "data: [DONE]") {
+			n++
+		}
+	}
+	return n
+}
+
 // Write implements http.ResponseWriter
 func (w *ResponseWriterWrapper) Write(b []byte) (int, error) {
-	// Record TTFT on first write
-	if !w.ttftRecorded && len(b) > 0 {
+	if len(b) > 0 && w.isStreaming() {
 		now := time.Now()
-		ttft := now.Sub(w.startTime).Milliseconds()
-		w.metrics.RecordTTFT(w.model, ttft)
-		w.ttftRecorded = true
-		w.firstByteTime = now
-		w.lastWriteTime = now
-	}
-
-	// Record ITL for subsequent writes (for streaming)
-	if w.ttftRecorded && len(b) > 0 {
-		now := time.Now()
-		if !w.firstByteTime.IsZero() && w.lastWriteTime != now {
-			itl := now.Sub(w.lastWriteTime).Milliseconds()
-			// Only record if ITL > 0 (avoid recording same millisecond twice)
-			if itl > 0 {
-				w.metrics.RecordITL(w.model, itl)
-			}
+		if !w.ttftRecorded {
+			w.metrics.RecordTTFT(w.model, now.Sub(w.startTime).Milliseconds())
+			w.ttftRecorded = true
+			w.firstByteTime = now
+		} else if itl := now.Sub(w.lastWriteTime).Milliseconds(); itl > 0 {
+			// Only record positive ITL: a zero means two chunks landed in the
+			// same millisecond, which carries no timing information.
+			w.metrics.RecordITL(w.model, itl)
 		}
 		w.lastWriteTime = now
+
+		if n := countSSETokens(b); n > 0 {
+			w.metrics.RecordTokenCount(w.model, n)
+		}
 	}
 
 	return w.writer.Write(b)
@@ -150,4 +118,8 @@ func (w *ResponseWriterWrapper) Hijack() (net.Conn, *bufio.ReadWriter, error) {
 	return nil, nil, fmt.Errorf("response writer does not implement Hijacker")
 }
 
-
+// Unwrap exposes the underlying writer to http.ResponseController so optional
+// capabilities (Flush, Hijack, deadlines) keep working through the wrapper.
+func (w *ResponseWriterWrapper) Unwrap() http.ResponseWriter {
+	return w.writer
+}

internal/proxy/llm_metrics_test.go

@@ -0,0 +1,171 @@
+package proxy
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"regexp"
+	"strings"
+	"testing"
+	"time"
+
+	"forgejo.riotpiao.com/rock/homelab-frontend/internal/config"
+	"forgejo.riotpiao.com/rock/homelab-frontend/internal/observability"
+)
+
+// exportMetrics renders the Prometheus text exposition for m.
+func exportMetrics(m *observability.Metrics) string {
+	return observability.NewPrometheusExporter(m).Export()
+}
+
+// requireSeries asserts the exporter emitted a sample line with the exact value.
+func requireSeries(t *testing.T, out, want string) {
+	t.Helper()
+	for _, line := range strings.Split(out, "\n") {
+		if strings.TrimSpace(line) == want {
+			return
+		}
+	}
+	t.Errorf("missing metrics series %q\n--- got ---\n%s", want, out)
+}
+
+// TestResponseWriterWrapperRecordsSSEMetrics guards the regression where the
+// LLM latency wrapper existed but was never constructed, leaving
+// llm_ttft_seconds / llm_itl_seconds / llm_tokens_total as HELP-only stubs.
+func TestResponseWriterWrapperRecordsSSEMetrics(t *testing.T) {
+	m := observability.NewMetrics()
+
+	rec := httptest.NewRecorder()
+	rec.Header().Set("Content-Type", "text/event-stream")
+
+	w := NewResponseWriterWrapper(rec, "reasoning", m, time.Now().Add(-50*time.Millisecond))
+
+	if _, err := w.Write([]byte("data: {\"token\":\"a\"}\n\n")); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	time.Sleep(5 * time.Millisecond)
+	if _, err := w.Write([]byte("data: {\"token\":\"b\"}\n\n")); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	// The terminator is not a generated token.
+	if _, err := w.Write([]byte("data: [DONE]\n\n")); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	out := exportMetrics(m)
+	requireSeries(t, out, `llm_ttft_seconds_count{model="reasoning"} 1`)
+	requireSeries(t, out, `llm_itl_seconds_count{model="reasoning"} 1`)
+	requireSeries(t, out, `llm_tokens_total{model="reasoning"} 2`)
+}
+
+// TestResponseWriterWrapperIgnoresNonStreaming verifies a single-shot JSON
+// response does not produce a bogus TTFT sample equal to full request latency.
+func TestResponseWriterWrapperIgnoresNonStreaming(t *testing.T) {
+	m := observability.NewMetrics()
+
+	rec := httptest.NewRecorder()
+	rec.Header().Set("Content-Type", "application/json")
+
+	w := NewResponseWriterWrapper(rec, "reasoning", m, time.Now().Add(-50*time.Millisecond))
+	if _, err := w.Write([]byte(`{"choices":[]}`)); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+
+	out := exportMetrics(m)
+	if strings.Contains(out, `llm_ttft_seconds_count{model="reasoning"}`) {
+		t.Errorf("non-streaming response recorded TTFT\n--- got ---\n%s", out)
+	}
+	if strings.Contains(out, `llm_tokens_total{model="reasoning"}`) {
+		t.Errorf("non-streaming response recorded tokens\n--- got ---\n%s", out)
+	}
+}
+
+// TestResponseWriterWrapperPreservesStreaming verifies the wrapper stays
+// transparent: bytes reach the client and Flush still works, so wrapping the
+// writer cannot silently break SSE (issue #33).
+func TestResponseWriterWrapperPreservesStreaming(t *testing.T) {
+	m := observability.NewMetrics()
+
+	rec := httptest.NewRecorder()
+	rec.Header().Set("Content-Type", "text/event-stream")
+
+	w := NewResponseWriterWrapper(rec, "reasoning", m, time.Now())
+	payload := "data: {\"token\":\"a\"}\n\n"
+	if _, err := w.Write([]byte(payload)); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	w.Flush()
+
+	if got := rec.Body.String(); got != payload {
+		t.Errorf("body = %q, want %q", got, payload)
+	}
+	if !rec.Flushed {
+		t.Error("Flush did not reach the underlying writer")
+	}
+}
+
+// TestHandlerRecordsLLMMetricsForStreaming is the end-to-end guard: it drives a
+// real streaming request through the Handler, so it fails if the wrapper is ever
+// constructed nowhere (the original defect) rather than merely being broken.
+func TestHandlerRecordsLLMMetricsForStreaming(t *testing.T) {
+	// Unique model name keeps the observations isolated in the global registry.
+	model := fmt.Sprintf("metrics-test-%d", time.Now().UnixNano())
+
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(http.StatusOK)
+		rc := http.NewResponseController(w)
+		for _, tok := range []string{"a", "b", "c"} {
+			time.Sleep(2 * time.Millisecond) // guarantee distinct ITL samples
+			fmt.Fprintf(w, "data: {\"token\":%q}\n\n", tok)
+			if err := rc.Flush(); err != nil {
+				return
+			}
+		}
+		fmt.Fprint(w, "data: [DONE]\n\n")
+		_ = rc.Flush()
+	}))
+	defer upstream.Close()
+
+	cfg := &config.Config{
+		Models: map[string]*config.ModelUpstream{
+			model: {
+				Name:    model,
+				Address: strings.TrimPrefix(upstream.URL, "http://"),
+				Path:    "/v1/chat/completions",
+			},
+		},
+		Routes: make(map[string]*config.Route),
+	}
+
+	handler := New(cfg)
+	defer handler.Close()
+	srv := httptest.NewServer(handler)
+	defer srv.Close()
+
+	body := fmt.Sprintf(`{"model":%q,"stream":true,"messages":[{"role":"user","content":"hi"}]}`, model)
+	resp, err := http.Post(srv.URL+"/v1/chat/completions", "application/json", strings.NewReader(body))
+	if err != nil {
+		t.Fatalf("request failed: %v", err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		t.Fatalf("status = %d, want 200", resp.StatusCode)
+	}
+	if _, err := io.ReadAll(resp.Body); err != nil {
+		t.Fatalf("read body: %v", err)
+	}
+
+	out := exportMetrics(observability.Global())
+	requireSeries(t, out, fmt.Sprintf(`llm_ttft_seconds_count{model=%q} 1`, model))
+	requireSeries(t, out, fmt.Sprintf(`llm_tokens_total{model=%q} 3`, model))
+
+	itlCount := regexp.MustCompile(fmt.Sprintf(`llm_itl_seconds_count\{model=%q\} (\d+)`, model))
+	match := itlCount.FindStringSubmatch(out)
+	if match == nil {
+		t.Errorf("no llm_itl_seconds samples recorded for %s\n--- got ---\n%s", model, out)
+	} else if match[1] == "0" {
+		t.Errorf("llm_itl_seconds_count = 0, want > 0")
+	}
+}

internal/proxy/proxy.go

@@ -2,6 +2,7 @@
 package proxy
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
@@ -47,6 +48,10 @@ type Handler struct {
 	schedulers  map[string]*Scheduler
 	// CAG: delay applied to warmup requests before queue entry (#45)
 	warmupDelay time.Duration
+	// engineObservers polls each model backend's /metrics for admission (#64)
+	engineObservers map[string]*EngineObserver
+	// maxBodyBytes caps synthesized model-route request bodies (MAX_BODY_BYTES)
+	maxBodyBytes int64
 	// transportsMu guards h.transports for goroutine-safe reads/writes.
 	transportsMu sync.RWMutex
 }
@@ -122,18 +127,80 @@ func New(cfg *config.Config) *Handler {
 			maxSlots = n
 		}
 	}
-	maxQueue := 20
+	maxQueue := 8
 	if v := os.Getenv("VLLM_MAX_QUEUE_DEPTH"); v != "" {
 		if n, err := strconv.Atoi(v); err == nil && n > 0 {
 			maxQueue = n
 		}
 	}
+
+	// Issue #64: engine-wide admission thresholds. MaxInFlight defaults to the
+	// same value as VLLM_MAX_NUM_SEQS; the observers below read the engine's
+	// own gauges, which every replica shares.
+	admissionMaxInFlight := maxSlots
+	if v := os.Getenv("ADMISSION_MAX_INFLIGHT"); v != "" {
+		if n, err := strconv.Atoi(v); err == nil && n > 0 {
+			admissionMaxInFlight = n
+		}
+	}
+	admissionMaxKVUsage := 0.90
+	if v := os.Getenv("ADMISSION_MAX_KV_USAGE"); v != "" {
+		if f, err := strconv.ParseFloat(v, 64); err == nil && f > 0 {
+			admissionMaxKVUsage = f
+		}
+	}
+	maxParkSeconds := 120
+	if v := os.Getenv("ADMISSION_MAX_PARK_SECONDS"); v != "" {
+		if n, err := strconv.Atoi(v); err == nil && n > 0 {
+			maxParkSeconds = n
+		}
+	}
+	h.maxBodyBytes = 1048576
+	if v := os.Getenv("MAX_BODY_BYTES"); v != "" {
+		if n, err := strconv.ParseInt(v, 10, 64); err == nil && n > 0 {
+			h.maxBodyBytes = n
+		}
+	}
+
+	limits := AdmissionLimits{
+		MaxInFlight: admissionMaxInFlight,
+		MaxKVUsage:  admissionMaxKVUsage,
+		MaxPark:     time.Duration(maxParkSeconds) * time.Second,
+	}
+
+	// Issue #64: only backends that actually publish the vLLM scheduler gauges
+	// may be observed. `reasoning` is the only vLLM engine here; ornith and
+	// qwen-cpu are Ollama and embeddings/rerank are TEI, none of which export
+	// vllm:* series. Observing them would trip the startup self-check and hold
+	// observer_degraded at 1 forever. Unset (the default) observes nothing and
+	// leaves admission exactly as it was before this change.
+	observeModels := make(map[string]bool)
+	for _, name := range strings.Split(os.Getenv("ADMISSION_OBSERVE_MODELS"), ",") {
+		if name = strings.TrimSpace(name); name != "" {
+			observeModels[name] = true
+		}
+	}
+
+	h.engineObservers = make(map[string]*EngineObserver)
 	h.schedulers = make(map[string]*Scheduler)
 	for modelName, model := range cfg.Models {
 		addrs := model.AllAddresses()
-		if len(addrs) > 0 {
-			h.schedulers[modelName] = NewScheduler(addrs, maxSlots, maxQueue)
+		if len(addrs) == 0 {
+			continue
 		}
+		if observeModels[modelName] {
+			for _, addr := range addrs {
+				if _, ok := h.engineObservers[addr]; !ok {
+					h.engineObservers[addr] = NewEngineObserver(addr, enginePollInterval)
+				}
+			}
+		}
+		h.schedulers[modelName] = NewScheduler(addrs, maxSlots, maxQueue, h.engineObservers, limits)
+	}
+	// The handler has no request-path lifecycle to hang cancellation on, so the
+	// observers poll for the lifetime of the process.
+	for _, o := range h.engineObservers {
+		o.Start(context.Background())
 	}
 	// Pre-populate transports for all scheduler backends at startup (single-threaded).
 	// getOrCreateTransport is not goroutine-safe; pre-populating avoids concurrent writes.
@@ -151,7 +218,7 @@ func New(cfg *config.Config) *Handler {
 		// Create a transport per unique upstream address for connection reuse
 		transport := h.getOrCreateTransport(route.Upstream.Address, &route.Upstream)
 		
-		upstreamURL, _ := url.Parse("http://" + route.Upstream.Address)
+		upstreamURL := &url.URL{Scheme: "http", Host: route.Upstream.Address}
 		
 		r := &Route{
 			Name:      name,
@@ -314,6 +381,10 @@ func writeProblemDetail(w http.ResponseWriter, status int, problemType, title, d
 
 // ServeHTTP implements http.Handler.
 func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	// Captured before any work so TTFT measures from the moment the gateway
+	// accepted the request, not from when the upstream response is copied.
+	requestStart := time.Now()
+
 	if h.handleAuthEndpoint(w, r) {
 		return
 	}
@@ -358,7 +429,14 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	// For streaming responses (SSE, chunked), disable buffering. Issue #33.
 	w.Header().Set("X-Accel-Buffering", "no")
 
-	proxy.ServeHTTP(w, r)
+	// Capture per-request LLM latency metrics (TTFT/ITL/tokens) while the
+	// response is copied from the upstream. Inert for non-SSE responses.
+	var respWriter http.ResponseWriter = w
+	if route.ModelName != "" {
+		respWriter = NewResponseWriterWrapper(w, route.ModelName, observability.Global(), requestStart)
+	}
+
+	proxy.ServeHTTP(respWriter, r)
 }
 
 
@@ -384,27 +462,104 @@ func (h *Handler) applyWarmup(ctx context.Context, r *http.Request) bool {
 // acquireScheduledBackend selects a backend from sched, applies warmup
 // deprioritisation, acquires a concurrency slot, and patches route when the
 // selected backend differs from the config default. It returns the backend
-// to Release (caller must defer), the (possibly patched) route, and ok=true.
-// If the slot cannot be acquired it writes a 503 and returns (nil, nil, false).
-func (h *Handler) acquireScheduledBackend(ctx context.Context, w http.ResponseWriter, r *http.Request, sched *Scheduler, route *Route, cacheKey string) (*Backend, *Route, bool) {
+// to Release (caller must defer), the (possibly patched) route, whether SSE
+// headers were already written while parking, and ok=true.
+// If the slot cannot be acquired it writes a 503 and returns ok=false.
+func (h *Handler) acquireScheduledBackend(ctx context.Context, w http.ResponseWriter, r *http.Request, sched *Scheduler, route *Route, cacheKey string) (*Backend, *Route, bool, bool) {
 	backend := sched.Select(cacheKey)
 	if !h.applyWarmup(ctx, r) {
-		return nil, nil, false
+		return nil, nil, false, false
 	}
 	if backend == nil {
-		return nil, route, true
+		return nil, route, false, true
 	}
-	if err := backend.Acquire(ctx); err != nil {
-		if errors.Is(err, ErrQueueFull) {
+
+	// Decision Q22: a request that actually parks on the engine gate and wants
+	// an SSE stream gets its headers pre-written plus a 15 s keep-alive, so no
+	// intermediary times the connection out before the engine admits it.
+	var (
+		prewritten bool
+		hbStop     chan struct{}
+		hbDone     chan struct{}
+	)
+	onPark := func() {
+		if !requestWantsStream(r) {
+			return
+		}
+		flusher, ok := w.(http.Flusher)
+		if !ok {
+			return
+		}
+		prewritten = true
+		hbStop = make(chan struct{})
+		hbDone = make(chan struct{})
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.Header().Set("Cache-Control", "no-cache")
+		w.Header().Set("Connection", "keep-alive")
+		w.Header().Set("X-Accel-Buffering", "no")
+		w.WriteHeader(http.StatusOK)
+		fmt.Fprint(w, ": keep-alive\n\n")
+		flusher.Flush()
+		go func() {
+			defer close(hbDone)
+			ticker := time.NewTicker(15 * time.Second)
+			defer ticker.Stop()
+			for {
+				select {
+				case <-hbStop:
+					return
+				case <-ctx.Done():
+					return
+				case <-ticker.C:
+					fmt.Fprint(w, ": keep-alive\n\n")
+					flusher.Flush()
+				}
+			}
+		}()
+	}
+
+	err := backend.acquire(ctx, onPark)
+	if prewritten {
+		close(hbStop)
+		<-hbDone
+	}
+	if err != nil {
+		// Once the SSE status line is spent, report the rejection as an event
+		// instead of a second WriteHeader.
+		reject := func(body string) {
+			if prewritten {
+				fmt.Fprintf(w, "event: error\ndata: {\"error\":%q}\n\n", body)
+				if f, ok := w.(http.Flusher); ok {
+					f.Flush()
+				}
+				return
+			}
+			w.WriteHeader(http.StatusServiceUnavailable)
+			fmt.Fprintf(w, "%s", body)
+		}
+		if errors.Is(err, ErrAdmissionTimeout) {
+			observability.Global().IncrAdmissionRejected("park_timeout")
+			logging.Warnf("admission timeout", map[string]string{
+				"backend": backend.Address,
+				"path":    r.URL.Path,
+				"model":   route.ModelName,
+			})
+			if !prewritten {
+				w.Header().Set("Retry-After", "5")
+			}
+			reject("admission timeout")
+		} else if errors.Is(err, ErrQueueFull) {
 			observability.Global().IncrQueueFull()
+			observability.Global().IncrAdmissionRejected("queue_full")
 			logging.Warnf("backend queue full", map[string]string{
 				"backend": backend.Address,
 				"path":    r.URL.Path,
 				"model":   route.ModelName,
 			})
-			w.Header().Set("Retry-After", "5")
-			w.WriteHeader(http.StatusServiceUnavailable)
-			fmt.Fprintf(w, "queue full")
+			if !prewritten {
+				w.Header().Set("Retry-After", "5")
+			}
+			reject("queue full")
 		} else {
 			logging.Warnf("backend acquire failed", map[string]string{
 				"backend": backend.Address,
@@ -412,15 +567,14 @@ func (h *Handler) acquireScheduledBackend(ctx context.Context, w http.ResponseWr
 				"model":   route.ModelName,
 				"err":     err.Error(),
 			})
-			w.WriteHeader(http.StatusServiceUnavailable)
-			fmt.Fprintf(w, "upstream busy")
+			reject("upstream busy")
 		}
-		return nil, nil, false
+		return nil, nil, prewritten, false
 	}
 	if backend.Address != route.Upstream.Address {
 		patchedCfg := *route.Upstream
 		patchedCfg.Address = backend.Address
-		patchedURL, _ := url.Parse("http://" + backend.Address)
+		patchedURL := &url.URL{Scheme: "http", Host: backend.Address}
 		route = &Route{
 			Name:      route.Name,
 			ModelName: route.ModelName,
@@ -431,7 +585,27 @@ func (h *Handler) acquireScheduledBackend(ctx context.Context, w http.ResponseWr
 			},
 		}
 	}
-	return backend, route, true
+	return backend, route, prewritten, true
+}
+
+// requestWantsStream reports whether the JSON request body asks for an SSE
+// stream ("stream": true). The body is restored for the upstream.
+func requestWantsStream(r *http.Request) bool {
+	if r.Body == nil {
+		return false
+	}
+	body, err := io.ReadAll(r.Body)
+	if err != nil {
+		return false
+	}
+	r.Body = io.NopCloser(bytes.NewReader(body))
+	var probe struct {
+		Stream bool `json:"stream"`
+	}
+	if err := json.Unmarshal(body, &probe); err != nil {
+		return false
+	}
+	return probe.Stream
 }
 
 // handleAuthEndpoint routes /auth/* requests and returns true if handled.
@@ -594,9 +768,10 @@ func (h *Handler) enforceBodyLimit(w http.ResponseWriter, r *http.Request, route
 // Returns (nil, nil, false) when the backend slot cannot be acquired (response written).
 func (h *Handler) setupCAGProxy(ctx context.Context, w http.ResponseWriter, r *http.Request, route *Route, sched *Scheduler, cacheKey string) (*httputil.ReverseProxy, *Backend, bool) {
 	var selectedBackend *Backend
+	var headersPrewritten bool
 	if sched != nil {
 		var ok bool
-		selectedBackend, route, ok = h.acquireScheduledBackend(ctx, w, r, sched, route, cacheKey)
+		selectedBackend, route, headersPrewritten, ok = h.acquireScheduledBackend(ctx, w, r, sched, route, cacheKey)
 		if !ok {
 			return nil, nil, false
 		}
@@ -611,12 +786,52 @@ func (h *Handler) setupCAGProxy(ctx context.Context, w http.ResponseWriter, r *h
 
 	capturedBackend := selectedBackend
 	capturedRoute := route
+	if headersPrewritten {
+		// The 200 + text/event-stream status is already on the wire, so an
+		// upstream status can no longer be forwarded: httputil.ReverseProxy
+		// calls WriteHeader again and net/http keeps the first one. Frame a
+		// failure as an SSE event rather than letting a raw error body land
+		// inside the stream, which a streaming client reads as an empty
+		// completion and reports as success.
+		proxy.ModifyResponse = func(resp *http.Response) error {
+			if resp.StatusCode < http.StatusBadRequest {
+				return nil
+			}
+			body, _ := io.ReadAll(io.LimitReader(resp.Body, 4096))
+			resp.Body.Close()
+			logging.Warnf("upstream error after parked stream started", map[string]string{
+				"upstream": capturedRoute.Upstream.Address,
+				"status":   strconv.Itoa(resp.StatusCode),
+			})
+			resp.Body = io.NopCloser(strings.NewReader(fmt.Sprintf(
+				"event: error\ndata: {\"upstream_status\":%d,\"error\":%q}\n\n",
+				resp.StatusCode, strings.TrimSpace(string(body)))))
+			resp.ContentLength = -1
+			resp.Header.Del("Content-Length")
+			resp.Header.Set("Content-Type", "text/event-stream")
+			return nil
+		}
+	}
 	proxy.ErrorHandler = func(w http.ResponseWriter, r *http.Request, err error) {
 		if capturedBackend != nil {
 			if s := h.schedulers[capturedRoute.ModelName]; s != nil {
 				s.MarkFailed(capturedBackend.Address)
 			}
 		}
+		if headersPrewritten {
+			// SSE headers and a 200 are already on the wire: a 502 here would be
+			// an ignored WriteHeader and would tear the event stream. Report the
+			// failure as an event instead.
+			fmt.Fprint(w, "event: error\ndata: {\"error\":\"upstream error\"}\n\n")
+			if f, ok := w.(http.Flusher); ok {
+				f.Flush()
+			}
+			logging.Errorf("upstream error", err, map[string]string{
+				"upstream": capturedRoute.Upstream.Address,
+				"path":     r.URL.Path,
+			})
+			return
+		}
 		w.WriteHeader(http.StatusBadGateway)
 		fmt.Fprintf(w, "upstream error")
 		logging.Errorf("upstream error", err, map[string]string{

internal/proxy/router.go

@@ -69,9 +69,10 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 	// Restore the body so it can be read again by the upstream
 	r.Body = io.NopCloser(bytes.NewReader(bodyBytes))
 
-	// Parse the JSON to find the model field
-	var payload map[string]interface{}
-	if err := json.Unmarshal(bodyBytes, &payload); err != nil {
+	// Extract only the top-level "model" value with a token walk: the common
+	// path must not materialise the whole body (messages and all) as a map.
+	modelRaw, hasModel, err := extractTopLevelModel(bodyBytes)
+	if err != nil {
 		return nil, &modelValidationError{
 			Kind:    "invalid_json",
 			Message: "request body is not valid JSON",
@@ -79,7 +80,6 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 	}
 
 	// Extract the model name
-	modelVal, hasModel := payload["model"]
 	if !hasModel {
 		return nil, &modelValidationError{
 			Kind:    "missing_model",
@@ -88,7 +88,7 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 	}
 
 	// Handle null model
-	if modelVal == nil {
+	if bytes.Equal(bytes.TrimSpace(modelRaw), []byte("null")) {
 		return nil, &modelValidationError{
 			Kind:    "null_model",
 			Message: "'model' field is null",
@@ -96,8 +96,8 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 	}
 
 	// Extract as string
-	modelName, ok := modelVal.(string)
-	if !ok {
+	var modelName string
+	if err := json.Unmarshal(modelRaw, &modelName); err != nil {
 		return nil, &modelValidationError{
 			Kind:    "missing_model",
 			Message: "'model' field must be a string",
@@ -122,10 +122,26 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 		}
 	}
 
-	// If upstream expects a different model name, rewrite the body
+	// If upstream expects a different model name, rewrite the body. This path is
+	// uncommon, so it may decode the body as raw fields; the common path above
+	// does not build a full map.
 	if modelUpstream.UpstreamModel != "" && modelUpstream.UpstreamModel != modelName {
-		payload["model"] = modelUpstream.UpstreamModel
-		newBody, err := json.Marshal(payload)
+		var fields map[string]json.RawMessage
+		if err := json.Unmarshal(bodyBytes, &fields); err != nil {
+			return nil, &modelValidationError{
+				Kind:    "invalid_request",
+				Message: fmt.Sprintf("failed to rewrite model name: %v", err),
+			}
+		}
+		newModel, err := json.Marshal(modelUpstream.UpstreamModel)
+		if err != nil {
+			return nil, &modelValidationError{
+				Kind:    "invalid_request",
+				Message: fmt.Sprintf("failed to rewrite model name: %v", err),
+			}
+		}
+		fields["model"] = newModel
+		newBody, err := json.Marshal(fields)
 		if err != nil {
 			return nil, &modelValidationError{
 				Kind:    "invalid_request",
@@ -151,11 +167,11 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 		ConnectTimeout: h.defaultConnectTimeout,
 		ReadTimeout:    h.defaultReadTimeout,
 		WriteTimeout:   h.defaultWriteTimeout,
-		MaxBodySize:    h.defaultMaxBodySize,
+		MaxBodySize:    h.maxBodyBytes,
 		AuthRequired:   false,
 	}
 
-	targetURL, _ := url.Parse("http://" + modelUpstream.Address)
+	targetURL := &url.URL{Scheme: "http", Host: modelUpstream.Address}
 
 	route := &Route{
 		Name:      "v1-chat-" + modelName,
@@ -169,3 +185,87 @@ func (h *Handler) routeByModel(r *http.Request, path string) (*Route, error) {
 
 	return route, nil
 }
+
+// extractTopLevelModel walks a JSON object body with a streaming decoder and
+// returns the raw value of the top-level "model" field. It never builds the
+// full document as a map, so a large "messages" array is skipped by token.
+// hasModel is false when the key is absent or the document is a bare null.
+func extractTopLevelModel(body []byte) (model json.RawMessage, hasModel bool, err error) {
+	dec := json.NewDecoder(bytes.NewReader(body))
+	tok, err := dec.Token()
+	if err != nil {
+		return nil, false, err
+	}
+	if tok == nil {
+		// A bare `null` body decoded to a nil map before, i.e. no model field.
+		if _, err := dec.Token(); err != io.EOF {
+			if err == nil {
+				return nil, false, fmt.Errorf("unexpected data after JSON null")
+			}
+			return nil, false, err
+		}
+		return nil, false, nil
+	}
+	if delim, ok := tok.(json.Delim); !ok || delim != '{' {
+		return nil, false, fmt.Errorf("request body is not a JSON object")
+	}
+	for dec.More() {
+		keyTok, err := dec.Token()
+		if err != nil {
+			return nil, false, err
+		}
+		key, _ := keyTok.(string)
+		if key == "model" {
+			if err := dec.Decode(&model); err != nil {
+				return nil, false, err
+			}
+			hasModel = true
+			continue
+		}
+		if err := skipJSONValue(dec); err != nil {
+			return nil, false, err
+		}
+	}
+	// Consume the closing brace and reject trailing data, preserving the
+	// invalid_json answer the old whole-body decode produced.
+	if _, err := dec.Token(); err != nil {
+		return nil, false, err
+	}
+	if _, err := dec.Token(); err != io.EOF {
+		if err == nil {
+			return nil, false, fmt.Errorf("unexpected data after JSON object")
+		}
+		return nil, false, err
+	}
+	return model, hasModel, nil
+}
+
+// skipJSONValue advances dec past the next JSON value without materialising it.
+func skipJSONValue(dec *json.Decoder) error {
+	tok, err := dec.Token()
+	if err != nil {
+		return err
+	}
+	delim, ok := tok.(json.Delim)
+	if !ok {
+		return nil // scalar
+	}
+	if delim != '{' && delim != '[' {
+		return fmt.Errorf("unexpected delimiter %v", delim)
+	}
+	depth := 1
+	for depth > 0 {
+		tok, err := dec.Token()
+		if err != nil {
+			return err
+		}
+		if d, ok := tok.(json.Delim); ok {
+			if d == '{' || d == '[' {
+				depth++
+			} else {
+				depth--
+			}
+		}
+	}
+	return nil
+}

internal/proxy/scheduler.go

@@ -8,20 +8,41 @@ import (
 	"sync"
 	"sync/atomic"
 	"time"
+
+	"forgejo.riotpiao.com/rock/homelab-frontend/internal/observability"
 )
 
 // ErrQueueFull is returned when the in-memory queue is at capacity.
 var ErrQueueFull = errors.New("backend queue full")
 
+// ErrAdmissionTimeout is returned when a request parked on engine admission
+// exceeds the configured maximum park duration.
+var ErrAdmissionTimeout = errors.New("admission timeout")
+
+// AdmissionLimits are the engine-wide admission thresholds applied to every
+// backend of a scheduler (#64).
+type AdmissionLimits struct {
+	MaxInFlight int
+	MaxKVUsage  float64
+	MaxPark     time.Duration
+}
+
 // Backend represents a single vLLM upstream with a bounded in-memory queue.
 // Two channels model the queue:
 //
 //	queue  — waiting requests (bounded by VLLM_MAX_QUEUE_DEPTH)
 //	slots  — concurrent in-service requests (bounded by VLLM_MAX_NUM_SEQS)
+//
+// The engine gate is applied before the queue: admission reads the engine's
+// own published gauges via observer, which is global across replicas.
 type Backend struct {
-	Address string
-	queue   chan struct{}
-	slots   chan struct{}
+	Address     string
+	queue       chan struct{}
+	slots       chan struct{}
+	observer    *EngineObserver
+	maxInFlight int
+	maxKVUsage  float64
+	maxPark     time.Duration
 }
 
 // newBackend creates a Backend with given queue depth and concurrency limit.
@@ -40,7 +61,48 @@ func newBackend(addr string, maxSlots, maxQueue int) *Backend {
 // Acquire enters the in-memory queue then waits for a vLLM concurrency slot.
 // Returns ErrQueueFull immediately if the queue channel is at capacity (no goroutine queued).
 // Returns ctx.Err() if the context is cancelled while waiting for a slot.
+// Returns ErrAdmissionTimeout if the engine gate parks the request past maxPark.
 func (b *Backend) Acquire(ctx context.Context) error {
+	return b.acquire(ctx, nil)
+}
+
+// acquire is Acquire with an optional onPark callback. onPark is invoked once,
+// on the calling goroutine, the first time the request has to wait for engine
+// admission; the parked streaming path uses it to start an SSE heartbeat.
+func (b *Backend) acquire(ctx context.Context, onPark func()) error {
+	// Engine gate (#64): admit on the engine's own gauges, not on this pod's
+	// counters. Failing OPEN on !Healthy is deliberate — a metrics outage must
+	// not become an inference outage; the three-replica gate is only a fairness
+	// improvement on top of the engine's own queueing.
+	if b.observer != nil {
+		deadline := time.Now().Add(b.maxPark)
+		timer := time.NewTimer(time.Until(deadline))
+		defer timer.Stop()
+		parked := false
+		for {
+			// State and tick channel are read together under one lock: taking
+			// them separately can miss a poll that lands between the two calls.
+			st, tick := b.observer.Snapshot()
+			if !st.Healthy || (st.InFlight() < int64(b.maxInFlight) && st.KVUsage < b.maxKVUsage) {
+				break
+			}
+			if !parked {
+				observability.Global().IncrAdmissionParked()
+				parked = true
+				if onPark != nil {
+					onPark()
+				}
+			}
+			select {
+			case <-tick:
+			case <-ctx.Done():
+				return ctx.Err()
+			case <-timer.C:
+				return ErrAdmissionTimeout
+			}
+		}
+	}
+
 	// Non-blocking queue entry: reject immediately when queue is full.
 	select {
 	case b.queue <- struct{}{}:
@@ -76,10 +138,15 @@ type Scheduler struct {
 
 // NewScheduler creates a Scheduler for the given addresses.
 // maxSlots = VLLM_MAX_NUM_SEQS, maxQueue = VLLM_MAX_QUEUE_DEPTH.
-func NewScheduler(addresses []string, maxSlots, maxQueue int) *Scheduler {
+// observers holds one engine observer per backend address (may be nil).
+func NewScheduler(addresses []string, maxSlots, maxQueue int, observers map[string]*EngineObserver, limits AdmissionLimits) *Scheduler {
 	backends := make([]*Backend, len(addresses))
 	for i, addr := range addresses {
 		backends[i] = newBackend(addr, maxSlots, maxQueue)
+		backends[i].observer = observers[addr]
+		backends[i].maxInFlight = limits.MaxInFlight
+		backends[i].maxKVUsage = limits.MaxKVUsage
+		backends[i].maxPark = limits.MaxPark
 	}
 	return &Scheduler{
 		backends:    backends,

internal/proxy/scheduler_test.go

@@ -67,7 +67,7 @@ func TestBackendAcquireReturnsErrQueueFullWhenQueueFull(t *testing.T) {
 // always selects the same backend (rendezvous / highest-random-weight hashing).
 func TestSchedulerSelectRendezvousHashDeterministic(t *testing.T) {
 	addrs := []string{"a:80", "b:80", "c:80"}
-	s := NewScheduler(addrs, 10, 10)
+	s := NewScheduler(addrs, 10, 10, nil, AdmissionLimits{})
 
 	first := s.Select("same-cache-key")
 	if first == nil {
@@ -85,7 +85,7 @@ func TestSchedulerSelectRendezvousHashDeterministic(t *testing.T) {
 // requests evenly across all backends when no cache key is given.
 func TestSchedulerSelectNoKeyRoundRobin(t *testing.T) {
 	addrs := []string{"a:80", "b:80", "c:80"}
-	s := NewScheduler(addrs, 10, 10)
+	s := NewScheduler(addrs, 10, 10, nil, AdmissionLimits{})
 
 	counts := make(map[string]int)
 	for i := 0; i < 9; i++ {
@@ -160,7 +160,7 @@ func TestBackendAcquireCancelledContext(t *testing.T) {
 // selected within the fail window (30 s default).
 func TestSchedulerMarkFailedExcludesBackend(t *testing.T) {
 	addrs := []string{"a:80", "b:80"}
-	s := NewScheduler(addrs, 10, 10)
+	s := NewScheduler(addrs, 10, 10, nil, AdmissionLimits{})
 
 	s.MarkFailed("a:80")
 

internal/serviceadapter/session_adapter.go

@@ -102,6 +102,22 @@ func GetSessionSpec() *Spec {
 					},
 				},
 			},
+			// sandbox — per-session k8s Pod+PVC+Service lifecycle (§4.15.4)
+			{
+				Name: "sandbox",
+				Methods: []Method{
+					{
+						Verb:         "POST",
+						UpstreamPath: "/v1/sessions/sandbox",
+						Auth:         &Auth{Required: true, Capability: "sessions:write"},
+					},
+					{
+						Verb:         "DELETE",
+						UpstreamPath: "/v1/sessions/sandbox",
+						Auth:         &Auth{Required: true, Capability: "sessions:write"},
+					},
+				},
+			},
 		},
 	}
 }

internal/sessions/fleet_client.go

@@ -0,0 +1,97 @@
+// Package sessions — FleetClient calls the SandboxFleet controller REST API
+// to assign sessions to fleet pods. This is the primary sandbox path;
+// the raw-k8s SandboxManager is the fallback when the fleet controller
+// is unavailable (e.g. during fleet rollout).
+package sessions
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"time"
+)
+
+const (
+	defaultFleetControllerURL = "http://tool-definition-controller.poimen.svc.cluster.local:8788"
+)
+
+// FleetClient calls the SandboxFleet controller REST API.
+type FleetClient struct {
+	baseURL string
+	client  *http.Client
+}
+
+// FleetAssignResult is returned by the controller assign endpoint.
+type FleetAssignResult struct {
+	PodSvcDns string `json:"podSvcDns"`
+	PodName   string `json:"podName"`
+}
+
+// NewFleetClient creates a FleetClient from env or default URL.
+// FLEET_CONTROLLER_URL — default: defaultFleetControllerURL
+func NewFleetClient() *FleetClient {
+	url := os.Getenv("FLEET_CONTROLLER_URL")
+	if url == "" {
+		url = defaultFleetControllerURL
+	}
+	return &FleetClient{
+		baseURL: url,
+		client:  &http.Client{Timeout: 35 * time.Second},
+	}
+}
+
+// Assign asks the fleet controller to assign sessionID to a fleet pod.
+// Returns the pod's ClusterIP service DNS and pod name.
+func (f *FleetClient) Assign(ctx context.Context, sessionID string) (*FleetAssignResult, error) {
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost,
+		f.baseURL+"/sessions/"+sessionID+"/assign", nil)
+	if err != nil {
+		return nil, fmt.Errorf("build request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := f.client.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("fleet controller unreachable: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode >= 300 {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
+		return nil, fmt.Errorf("fleet controller returned %d: %s", resp.StatusCode, body)
+	}
+
+	var result FleetAssignResult
+	if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
+		return nil, fmt.Errorf("decode: %w", err)
+	}
+	if result.PodSvcDns == "" {
+		return nil, fmt.Errorf("fleet controller returned empty podSvcDns")
+	}
+	return &result, nil
+}
+
+// Release notifies the fleet controller that a session has ended.
+// Cleans up the per-session ServiceAccount (audit trail SA).
+// Non-fatal: fleet controller handles this async.
+func (f *FleetClient) Release(ctx context.Context, sessionID string) error {
+	req, err := http.NewRequestWithContext(ctx, http.MethodDelete,
+		f.baseURL+"/sessions/"+sessionID, bytes.NewReader(nil))
+	if err != nil {
+		return fmt.Errorf("build request: %w", err)
+	}
+	resp, err := f.client.Do(req)
+	if err != nil {
+		return fmt.Errorf("fleet controller unreachable: %w", err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode >= 300 {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
+		return fmt.Errorf("fleet controller returned %d: %s", resp.StatusCode, body)
+	}
+	return nil
+}

internal/sessions/handler.go

@@ -12,9 +12,11 @@ package sessions
 
 import (
 	"bytes"
+	"context"
 	"crypto/sha256"
 	"encoding/json"
 	"fmt"
+	"log"
 	"net/http"
 	"os"
 	"time"
@@ -30,9 +32,13 @@ const (
 
 // Handler handles session REST endpoints via internal service calls.
 // All traffic (reads and writes) goes to graphqlURL (poimen-graphql).
+// Sandbox creates/deletes pi-session pods via the SandboxFleet controller
+// (primary) with SandboxManager as direct-k8s fallback.
 type Handler struct {
 	graphqlURL string
 	client     *http.Client
+	sandbox    *SandboxManager
+	fleet      *FleetClient
 }
 
 // NewHandler creates a Handler reading upstream URL from env var.
@@ -42,10 +48,34 @@ func NewHandler() *Handler {
 	if gql == "" {
 		gql = defaultGraphQLURL
 	}
-	return &Handler{
+	h := &Handler{
 		graphqlURL: gql,
 		client:     &http.Client{Timeout: 30 * time.Second, Transport: tracing.NewTransport(nil)},
+		sandbox:    NewSandboxManager(),
+		fleet:      NewFleetClient(),
 	}
+	h.startIdleReaper()
+	return h
+}
+
+// startIdleReaper launches a background goroutine that deletes exited pi-session
+// pods (bridge.js idle-timeout triggered) while preserving their PVCs.
+func (h *Handler) startIdleReaper() {
+	intervalStr := os.Getenv("SANDBOX_REAPER_INTERVAL_MINUTES")
+	interval := defaultReaperInterval
+	if intervalStr != "" {
+		var min int
+		if n, _ := fmt.Sscanf(intervalStr, "%d", &min); n == 1 && min > 0 {
+			interval = time.Duration(min) * time.Minute
+		}
+	}
+	go func() {
+		ticker := time.NewTicker(interval)
+		defer ticker.Stop()
+		for range ticker.C {
+			h.sandbox.ReapIdlePods(context.Background())
+		}
+	}()
 }
 
 // ServeHTTP routes by X-Upstream-Path set by the dispatcher.
@@ -66,6 +96,8 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		h.handleListEpisodes(w, r)
 	case "/v1/sessions/episodes/get":
 		h.handleGetEpisode(w, r)
+	case "/v1/sessions/sandbox":
+		h.handleSandbox(w, r)
 	default:
 		writeJSON(w, http.StatusNotFound, map[string]string{"error": "sessions endpoint not found"})
 	}
@@ -180,9 +212,12 @@ func (h *Handler) handleList(w http.ResponseWriter, r *http.Request) {
 		searchQuery = fmt.Sprintf("%s cwd:%s", searchQuery, cwd)
 	}
 
+	// entities query (not hybridSearch) — project arg is optional, so it queries
+	// across ALL projects. Required because each session lives in its own
+	// session-<uuid> project; there is no single project to hybridSearch against.
 	gqlQuery := `query($q:String!,$limit:Int!){
-		hybridSearch(query:$q,limit:$limit,entityType:SESSION){
-			entities{id name summary tCreated}
+		entities(query:$q,limit:$limit,entityType:SESSION){
+			entities{id name summary projectId tCreated}
 			searchMode
 		}
 	}`
@@ -495,6 +530,94 @@ func (h *Handler) gql(r *http.Request, query string, variables map[string]any) (
 	return result, nil
 }
 
+// handleSandbox dispatches POST (create/resume) and DELETE (teardown).
+// Both take ?session_id= as a query param.
+func (h *Handler) handleSandbox(w http.ResponseWriter, r *http.Request) {
+	sessionID := r.URL.Query().Get("session_id")
+	if sessionID == "" {
+		writeJSON(w, http.StatusBadRequest, map[string]string{"error": "session_id query param required"})
+		return
+	}
+
+	switch r.Method {
+	case http.MethodPost:
+		h.handleSandboxCreate(w, r, sessionID)
+	case http.MethodDelete:
+		h.handleSandboxDelete(w, r, sessionID)
+	default:
+		writeJSON(w, http.StatusMethodNotAllowed, map[string]string{"error": "method not allowed"})
+	}
+}
+
+// POST ?session_id= — assign session to SandboxFleet pod (primary) or
+// create a dedicated pod via SandboxManager (fallback).
+func (h *Handler) handleSandboxCreate(w http.ResponseWriter, r *http.Request, sessionID string) {
+	// Primary: fleet controller assigns session to a pre-warmed pod.
+	if assign, err := h.fleet.Assign(r.Context(), sessionID); err == nil {
+		svcDns := assign.PodSvcDns + fmt.Sprintf(":%d", sandboxPort)
+		log.Printf("sandbox: fleet assigned session=%s pod=%s dns=%s", sessionID, assign.PodName, svcDns)
+		go h.updateSandboxSummary(r, sessionID, assign.PodName, "", svcDns)
+		writeJSON(w, http.StatusOK, map[string]any{
+			"session_id": sessionID,
+			"pod_name":   assign.PodName,
+			"svc_dns":    svcDns,
+			"resumed":    false,
+			"fleet":      true,
+		})
+		return
+	} else {
+		log.Printf("sandbox: fleet unavailable for session=%s (%v), falling back to direct k8s", sessionID, err)
+	}
+
+	// Fallback: create a dedicated pod via SandboxManager.
+	result, err := h.sandbox.CreateOrResume(r.Context(), sessionID)
+	if err != nil {
+		log.Printf("sandbox create/resume session=%s: %v", sessionID, err)
+		writeJSON(w, http.StatusBadGateway, map[string]string{"error": "sandbox error: " + err.Error()})
+		return
+	}
+	go h.updateSandboxSummary(r, sessionID, result.PodName, result.PVCName, result.SvcDNS)
+	log.Printf("sandbox: direct pod session=%s pod=%s resumed=%v dns=%s", sessionID, result.PodName, result.Resumed, result.SvcDNS)
+	writeJSON(w, http.StatusOK, result)
+}
+
+// updateSandboxSummary merges pod/svc metadata into the session memory_entity (fire-and-forget).
+// Uses context.Background() because r.Context() is cancelled once the HTTP handler returns.
+func (h *Handler) updateSandboxSummary(r *http.Request, sessionID, podName, pvcName, svcDns string) {
+	// Build a detached request carrying just the auth header.
+	detached, _ := http.NewRequestWithContext(context.Background(), r.Method, r.URL.String(), nil)
+	if auth := r.Header.Get("Authorization"); auth != "" {
+		detached.Header.Set("Authorization", auth)
+	}
+	current, _ := h.fetchEntitySummary(detached, "session-"+sessionID, sessionID, "SESSION")
+	if current == nil {
+		current = map[string]any{}
+	}
+	current["pod_name"] = podName
+	current["svc_dns"] = svcDns
+	current["last_active_at"] = time.Now().UTC().Format(time.RFC3339)
+	if pvcName != "" {
+		current["pvc_name"] = pvcName
+	}
+	if b, err := json.Marshal(current); err == nil {
+		h.upsertEntity(r, "session-"+sessionID, sessionID, "SESSION", string(b)) //nolint:errcheck
+	}
+}
+
+// DELETE ?session_id= — release from fleet + delete dedicated pod+svc+pvc if any.
+func (h *Handler) handleSandboxDelete(w http.ResponseWriter, r *http.Request, sessionID string) {
+	// Release from fleet (cleans up per-session SA; non-fatal if session was fleet-based).
+	if err := h.fleet.Release(r.Context(), sessionID); err != nil {
+		log.Printf("sandbox: fleet release session=%s: %v (continuing)", sessionID, err)
+	}
+	// Also attempt direct k8s cleanup (no-ops if resources don't exist).
+	if err := h.sandbox.Delete(r.Context(), sessionID); err != nil {
+		log.Printf("sandbox delete session=%s: %v", sessionID, err)
+	}
+	log.Printf("sandbox: released session=%s", sessionID)
+	writeJSON(w, http.StatusOK, map[string]string{"session_id": sessionID, "status": "deleted"})
+}
+
 // EpisodeID computes the deterministic episode ID per §4.14 Q6:
 // sha256(session_id + milestone_id + attempt).
 // Exported so callers can verify the value the server would produce.

internal/sessions/handler_test.go

@@ -18,7 +18,8 @@ import (
 // "query" field and records captured request bodies.
 //
 //   - "upsertEntity" → mutation response
-//   - "hybridSearch"  → search response
+//   - "hybridSearch"  → search response (get, patch, episodes handlers)
+//   - "entities"      → list response (sessions list handler)
 //   - "entity"        → single entity response
 func newGQLServer(t *testing.T, captured *[]byte) *httptest.Server {
 	t.Helper()
@@ -33,7 +34,11 @@ func newGQLServer(t *testing.T, captured *[]byte) *httptest.Server {
 		case strings.Contains(bs, "upsertEntity"):
 			fmt.Fprint(w, `{"data":{"upsertEntity":{"id":"test-entity-id","created":true}}}`)
 		case strings.Contains(bs, "hybridSearch"):
+			// get, patch, episodes handlers still use hybridSearch
 			fmt.Fprint(w, `{"data":{"hybridSearch":{"entities":[]}}}`)
+		case strings.Contains(bs, "entities"):
+			// sessions list handler uses entities query
+			fmt.Fprint(w, `{"data":{"entities":{"entities":[]}}}`)
 		case strings.Contains(bs, "entity"):
 			fmt.Fprint(w, `{"data":{"entity":null}}`)
 		default:

internal/sessions/sandbox.go

@@ -0,0 +1,492 @@
+// Package sessions — SandboxManager provisions per-session k8s resources:
+//
+//	PVC   workspace-session-<uuid>   Longhorn 1Gi
+//	Pod   pi-session-<uuid>           single container, /workspace <- PVC
+//	Svc   pi-session-<uuid>           ClusterIP :8787
+//	NetPol pi-session-sandbox         shared for all pi-session pods (created once)
+//
+// All k8s API calls use raw HTTP + in-cluster SA token (no client-go).
+// Same pattern as internal/webhook/promoter.go.
+package sessions
+
+import (
+	"bytes"
+	"context"
+	"crypto/tls"
+	"crypto/x509"
+	"encoding/json"
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"os"
+	"strings"
+	"time"
+)
+
+const (
+	sandboxNamespace    = "poimen"
+	sandboxImage        = "forgejo.riotpiao.com/riotpiao-poimen/poimen-pi-session-runtime:latest"
+	sandboxPort         = 8787
+	defaultPVCSize      = "1Gi"
+	defaultStorageClass = "longhorn"
+	defaultReadyTimeout  = 30 * time.Second
+	defaultReaperInterval = 5 * time.Minute
+	sandboxSATokenPath  = "/var/run/secrets/kubernetes.io/serviceaccount/token"
+	sandboxCACertPath   = "/var/run/secrets/kubernetes.io/serviceaccount/ca.crt"
+)
+
+// SandboxManager manages the k8s lifecycle of pi-session Pods, PVCs, Services.
+type SandboxManager struct {
+	k8sHost      string
+	k8sClient    *http.Client
+	pvcSize      string
+	readyTimeout time.Duration
+}
+
+// SandboxResult is returned by CreateOrResume.
+type SandboxResult struct {
+	SessionID string `json:"session_id"`
+	PodName   string `json:"pod_name"`
+	PVCName   string `json:"pvc_name"`
+	SvcDNS    string `json:"svc_dns"`
+	Resumed   bool   `json:"resumed"` // true = pod already existed
+}
+
+// NewSandboxManager creates a SandboxManager from env + in-cluster SA files.
+func NewSandboxManager() *SandboxManager {
+	host := os.Getenv("K8S_API_HOST")
+	if host == "" {
+		host = "https://kubernetes.default.svc"
+	}
+	pvcSize := os.Getenv("SANDBOX_PVC_SIZE")
+	if pvcSize == "" {
+		pvcSize = defaultPVCSize
+	}
+	caCert, _ := os.ReadFile(sandboxCACertPath)
+	return &SandboxManager{
+		k8sHost:      host,
+		k8sClient:    buildSandboxK8sClient(caCert),
+		pvcSize:      pvcSize,
+		readyTimeout: defaultReadyTimeout,
+	}
+}
+
+func buildSandboxK8sClient(caCert []byte) *http.Client {
+	transport := &http.Transport{}
+	if len(caCert) > 0 {
+		pool := x509.NewCertPool()
+		pool.AppendCertsFromPEM(caCert)
+		transport.TLSClientConfig = &tls.Config{RootCAs: pool}
+	} else {
+		log.Printf("sandbox: WARNING no SA CA cert at %s — using InsecureSkipVerify", sandboxCACertPath)
+		transport.TLSClientConfig = &tls.Config{InsecureSkipVerify: true} //nolint:gosec
+	}
+	return &http.Client{Timeout: 30 * time.Second, Transport: transport}
+}
+
+func sandboxSAToken() string {
+	b, err := os.ReadFile(sandboxSATokenPath)
+	if err != nil {
+		log.Printf("sandbox: WARNING failed to read SA token %s: %v", sandboxSATokenPath, err)
+		return ""
+	}
+	return strings.TrimSpace(string(b))
+}
+
+func pvcName(sessionID string) string { return "workspace-session-" + sessionID }
+func podName(sessionID string) string { return "pi-session-" + sessionID }
+func svcName(sessionID string) string { return "pi-session-" + sessionID }
+func svcDNS(sessionID string) string {
+	return fmt.Sprintf("%s.%s.svc.cluster.local:%d", svcName(sessionID), sandboxNamespace, sandboxPort)
+}
+
+// ── public API ────────────────────────────────────────────────────────────
+
+// CreateOrResume provisions or resumes a sandbox for sessionID.
+// Idempotent: if the pod is already Running, returns its DNS immediately.
+func (s *SandboxManager) CreateOrResume(ctx context.Context, sessionID string) (*SandboxResult, error) {
+	res := &SandboxResult{
+		SessionID: sessionID,
+		PodName:   podName(sessionID),
+		PVCName:   pvcName(sessionID),
+		SvcDNS:    svcDNS(sessionID),
+	}
+
+	// 1. Ensure PVC exists
+	if err := s.ensurePVC(ctx, sessionID); err != nil {
+		return nil, fmt.Errorf("ensure PVC: %w", err)
+	}
+
+	// 2. Check if pod already running (resume case)
+	phase, err := s.getPodPhase(ctx, sessionID)
+	if err != nil {
+		return nil, fmt.Errorf("check pod: %w", err)
+	}
+	if phase == "Running" {
+		res.Resumed = true
+		return res, nil
+	}
+
+	// 3. Ensure NetworkPolicy for all pi-session pods exists
+	if err := s.ensureNetworkPolicy(ctx); err != nil {
+		log.Printf("sandbox: WARNING failed to ensure network policy: %v", err)
+	}
+
+	// 4. Create Pod + Service
+	if err := s.createPod(ctx, sessionID); err != nil {
+		return nil, fmt.Errorf("create pod: %w", err)
+	}
+	if err := s.ensureService(ctx, sessionID); err != nil {
+		return nil, fmt.Errorf("ensure service: %w", err)
+	}
+
+	// 5. Wait for Ready
+	if err := s.waitPodReady(ctx, sessionID); err != nil {
+		return nil, fmt.Errorf("wait ready: %w", err)
+	}
+
+	return res, nil
+}
+
+// Delete removes the Pod, Service, and PVC for sessionID.
+func (s *SandboxManager) Delete(ctx context.Context, sessionID string) error {
+	var errs []string
+	for _, fn := range []func(context.Context, string) error{
+		s.deletePod, s.deleteService, s.deletePVC,
+	} {
+		if err := fn(ctx, sessionID); err != nil && !isNotFound(err) {
+			errs = append(errs, err.Error())
+		}
+	}
+	if len(errs) > 0 {
+		return fmt.Errorf("delete sandbox: %s", strings.Join(errs, "; "))
+	}
+	return nil
+}
+
+// ReapIdlePods lists all pi-session-* pods and deletes those that have exited
+// (Succeeded/Failed) — bridge.js self-exits after idle timeout.
+// The PVC is preserved so the next CreateOrResume reattaches the same workspace.
+func (s *SandboxManager) ReapIdlePods(ctx context.Context) {
+	pods, err := s.listSandboxPods(ctx)
+	if err != nil {
+		log.Printf("sandbox reaper: list pods: %v", err)
+		return
+	}
+	for _, podAny := range pods {
+		pod, _ := podAny.(map[string]any)
+		phase, _ := pod["status"].(map[string]any)["phase"].(string)
+		name, _ := pod["metadata"].(map[string]any)["name"].(string)
+		if phase == "Succeeded" || phase == "Failed" {
+			log.Printf("sandbox reaper: deleting exited pod %s (phase=%s)", name, phase)
+			if err := s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+name); err != nil && !isNotFound(err) {
+				log.Printf("sandbox reaper: delete pod %s: %v", name, err)
+			}
+		}
+	}
+}
+
+// ── k8s resource helpers ─────────────────────────────────────────────────
+
+func (s *SandboxManager) ensurePVC(ctx context.Context, sessionID string) error {
+	name := pvcName(sessionID)
+	_, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/persistentvolumeclaims/"+name)
+	if err == nil {
+		return nil // already exists
+	}
+	if !isNotFound(err) {
+		return err
+	}
+	// Create
+	body := map[string]any{
+		"apiVersion": "v1",
+		"kind":       "PersistentVolumeClaim",
+		"metadata": map[string]any{
+			"name": name, "namespace": sandboxNamespace,
+			"labels": map[string]string{"app": "pi-session", "session-id": sessionID},
+		},
+		"spec": map[string]any{
+			"accessModes":      []string{"ReadWriteOnce"},
+			"storageClassName": defaultStorageClass,
+			"resources": map[string]any{
+				"requests": map[string]string{"storage": s.pvcSize},
+			},
+		},
+	}
+	return s.k8sCreate(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/persistentvolumeclaims", body)
+}
+
+func (s *SandboxManager) getPodPhase(ctx context.Context, sessionID string) (string, error) {
+	pod, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+podName(sessionID))
+	if isNotFound(err) {
+		return "", nil
+	}
+	if err != nil {
+		return "", err
+	}
+	phase, _ := pod["status"].(map[string]any)["phase"].(string)
+	return phase, nil
+}
+
+func (s *SandboxManager) createPod(ctx context.Context, sessionID string) error {
+	body := map[string]any{
+		"apiVersion": "v1",
+		"kind":       "Pod",
+		"metadata": map[string]any{
+			"name": podName(sessionID), "namespace": sandboxNamespace,
+			"labels": map[string]string{"app": "pi-session", "session-id": sessionID},
+		},
+		"spec": map[string]any{
+			"restartPolicy":                "Never",
+			"automountServiceAccountToken": false,
+			"securityContext": map[string]any{
+				"runAsNonRoot": true,
+				"runAsUser":    int64(1000),
+				"seccompProfile": map[string]string{"type": "RuntimeDefault"},
+			},
+			"containers": []map[string]any{{
+				"name":  "pi-session",
+				"image": sandboxImage,
+				"ports": []map[string]any{{"containerPort": sandboxPort}},
+				"securityContext": map[string]any{
+					"readOnlyRootFilesystem":   true,
+					"allowPrivilegeEscalation": false,
+					"capabilities": map[string]any{"drop": []string{"ALL"}},
+				},
+				"resources": map[string]any{
+					"requests": map[string]string{"cpu": "500m", "memory": "512Mi"},
+					"limits":   map[string]string{"cpu": "1", "memory": "1Gi"},
+				},
+				"readinessProbe": map[string]any{
+					"tcpSocket":           map[string]any{"port": sandboxPort},
+					"initialDelaySeconds": 3,
+					"periodSeconds":       2,
+					"failureThreshold":    15,
+				},
+				"volumeMounts": []map[string]any{
+					{"name": "workspace", "mountPath": "/workspace"},
+					{"name": "tmp", "mountPath": "/tmp"},
+				},
+			}},
+			"volumes": []map[string]any{
+				{"name": "workspace", "persistentVolumeClaim": map[string]string{"claimName": pvcName(sessionID)}},
+				{"name": "tmp", "emptyDir": map[string]any{}},
+			},
+		},
+	}
+	return s.k8sCreate(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods", body)
+}
+
+func (s *SandboxManager) ensureService(ctx context.Context, sessionID string) error {
+	name := svcName(sessionID)
+	_, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/services/"+name)
+	if err == nil {
+		return nil
+	}
+	if !isNotFound(err) {
+		return err
+	}
+	body := map[string]any{
+		"apiVersion": "v1",
+		"kind":       "Service",
+		"metadata": map[string]any{
+			"name": name, "namespace": sandboxNamespace,
+			"labels": map[string]string{"app": "pi-session", "session-id": sessionID},
+		},
+		"spec": map[string]any{
+			"type":     "ClusterIP",
+			"selector": map[string]string{"app": "pi-session", "session-id": sessionID},
+			"ports": []map[string]any{{
+				"port": sandboxPort, "targetPort": sandboxPort, "protocol": "TCP",
+			}},
+		},
+	}
+	return s.k8sCreate(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/services", body)
+}
+
+func (s *SandboxManager) ensureNetworkPolicy(ctx context.Context) error {
+	const npName = "pi-session-sandbox"
+	_, err := s.k8sGet(ctx, "/apis/networking.k8s.io/v1/namespaces/"+sandboxNamespace+"/networkpolicies/"+npName)
+	if err == nil {
+		return nil // already exists
+	}
+	if !isNotFound(err) {
+		return err
+	}
+	body := map[string]any{
+		"apiVersion": "networking.k8s.io/v1",
+		"kind":       "NetworkPolicy",
+		"metadata": map[string]any{
+			"name": npName, "namespace": sandboxNamespace,
+			"labels": map[string]string{"app": "pi-session"},
+		},
+		"spec": map[string]any{
+			"podSelector": map[string]any{"matchLabels": map[string]string{"app": "pi-session"}},
+			"policyTypes": []string{"Ingress", "Egress"},
+			"ingress": []map[string]any{{
+				"from": []map[string]any{{
+					"podSelector": map[string]any{"matchLabels": map[string]string{"app": "poimen-frontend"}},
+				}},
+				"ports": []map[string]any{{"protocol": "TCP", "port": sandboxPort}},
+			}},
+			"egress": []map[string]any{
+				// DNS
+				{
+					"to": []map[string]any{{"namespaceSelector": map[string]any{"matchLabels": map[string]string{"kubernetes.io/metadata.name": "kube-system"}}}},
+					"ports": []map[string]any{{"protocol": "UDP", "port": 53}, {"protocol": "TCP", "port": 53}},
+				},
+				// api-gateway (LLM inference)
+				{
+					"to": []map[string]any{{"namespaceSelector": map[string]any{"matchLabels": map[string]string{"kubernetes.io/metadata.name": "api"}}}},
+					"ports": []map[string]any{{"protocol": "TCP", "port": 8080}},
+				},
+			},
+		},
+	}
+	return s.k8sCreate(ctx, "/apis/networking.k8s.io/v1/namespaces/"+sandboxNamespace+"/networkpolicies", body)
+}
+
+func (s *SandboxManager) waitPodReady(ctx context.Context, sessionID string) error {
+	deadline := time.Now().Add(s.readyTimeout)
+	for time.Now().Before(deadline) {
+		pod, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+podName(sessionID))
+		if err != nil {
+			return fmt.Errorf("poll pod: %w", err)
+		}
+		status, _ := pod["status"].(map[string]any)
+		phase, _ := status["phase"].(string)
+		if phase == "Failed" || phase == "Succeeded" {
+			return fmt.Errorf("pod entered terminal phase %s before ready", phase)
+		}
+		if phase == "Running" {
+			for _, cond := range toSlice(status["conditions"]) {
+				if c, ok := cond.(map[string]any); ok {
+					if c["type"] == "Ready" && c["status"] == "True" {
+						return nil
+					}
+				}
+			}
+		}
+		select {
+		case <-time.After(2 * time.Second):
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
+	return fmt.Errorf("pod not ready after %v", s.readyTimeout)
+}
+
+func (s *SandboxManager) deletePod(ctx context.Context, sessionID string) error {
+	return s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+podName(sessionID))
+}
+func (s *SandboxManager) deleteService(ctx context.Context, sessionID string) error {
+	return s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/services/"+svcName(sessionID))
+}
+func (s *SandboxManager) deletePVC(ctx context.Context, sessionID string) error {
+	return s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/persistentvolumeclaims/"+pvcName(sessionID))
+}
+
+func (s *SandboxManager) listSandboxPods(ctx context.Context) ([]any, error) {
+	body, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods?labelSelector=app%3Dpi-session")
+	if err != nil {
+		return nil, err
+	}
+	items, _ := body["items"].([]any)
+	return items, nil
+}
+
+// ── raw k8s HTTP ─────────────────────────────────────────────────────────
+
+func (s *SandboxManager) k8sGet(ctx context.Context, path string) (map[string]any, error) {
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, s.k8sHost+path, nil)
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Authorization", "Bearer "+sandboxSAToken())
+	resp, err := s.k8sClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("k8s GET %s: %w", path, err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode == 404 {
+		return nil, &k8sNotFound{path: path}
+	}
+	if resp.StatusCode >= 400 {
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
+		return nil, fmt.Errorf("k8s GET %s returned %d: %s", path, resp.StatusCode, b)
+	}
+	var result map[string]any
+	if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
+		return nil, fmt.Errorf("decode: %w", err)
+	}
+	return result, nil
+}
+
+func (s *SandboxManager) k8sCreate(ctx context.Context, path string, body any) error {
+	payload, err := json.Marshal(body)
+	if err != nil {
+		return fmt.Errorf("marshal: %w", err)
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, s.k8sHost+path, bytes.NewReader(payload))
+	if err != nil {
+		return err
+	}
+	req.Header.Set("Authorization", "Bearer "+sandboxSAToken())
+	req.Header.Set("Content-Type", "application/json")
+	resp, err := s.k8sClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("k8s POST %s: %w", path, err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode == 409 { // AlreadyExists — idempotent
+		return nil
+	}
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
+		return fmt.Errorf("k8s POST %s returned %d: %s", path, resp.StatusCode, b)
+	}
+	return nil
+}
+
+func (s *SandboxManager) k8sDelete(ctx context.Context, path string) error {
+	req, err := http.NewRequestWithContext(ctx, http.MethodDelete, s.k8sHost+path, nil)
+	if err != nil {
+		return err
+	}
+	req.Header.Set("Authorization", "Bearer "+sandboxSAToken())
+	resp, err := s.k8sClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("k8s DELETE %s: %w", path, err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode == 404 {
+		return &k8sNotFound{path: path}
+	}
+	if resp.StatusCode >= 400 {
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
+		return fmt.Errorf("k8s DELETE %s returned %d: %s", path, resp.StatusCode, b)
+	}
+	return nil
+}
+
+// ── error helpers ─────────────────────────────────────────────────────────
+
+type k8sNotFound struct{ path string }
+
+func (e *k8sNotFound) Error() string { return "not found: " + e.path }
+
+func isNotFound(err error) bool {
+	if err == nil {
+		return false
+	}
+	_, ok := err.(*k8sNotFound)
+	return ok
+}
+
+func toSlice(v any) []any {
+	if s, ok := v.([]any); ok {
+		return s
+	}
+	return nil
+}

internal/sessions/sandbox_test.go

@@ -0,0 +1,271 @@
+package sessions
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+// ── helpers ───────────────────────────────────────────────────────────────────
+
+// runningReadyPod returns minimal k8s pod JSON: phase Running + Ready condition.
+func runningReadyPod(name string) map[string]any {
+	return map[string]any{
+		"metadata": map[string]any{"name": name},
+		"status": map[string]any{
+			"phase": "Running",
+			"conditions": []any{
+				map[string]any{"type": "Ready", "status": "True"},
+			},
+		},
+	}
+}
+
+// respondK8sJSON writes a JSON body with the given status code.
+func respondK8sJSON(w http.ResponseWriter, status int, body any) {
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(status)
+	json.NewEncoder(w).Encode(body) //nolint:errcheck
+}
+
+// newTestSM builds a SandboxManager pointed at the given test server URL.
+func newTestSM(url string) *SandboxManager {
+	return &SandboxManager{
+		k8sHost:      url,
+		k8sClient:    &http.Client{Timeout: 5 * time.Second},
+		pvcSize:      "1Gi",
+		readyTimeout: 3 * time.Second,
+	}
+}
+
+// ── TestSandboxCreateOrResume_CreateNew ───────────────────────────────────────
+
+// No existing resources → creates PVC, NetworkPolicy, Pod, Service; Resumed=false.
+func TestSandboxCreateOrResume_CreateNew(t *testing.T) {
+	id := "newid"
+	var podGetCount int32
+
+	mux := http.NewServeMux()
+
+	// PVC: GET → 404, POST → 201
+	mux.HandleFunc("GET /api/v1/namespaces/poimen/persistentvolumeclaims/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	})
+	mux.HandleFunc("POST /api/v1/namespaces/poimen/persistentvolumeclaims", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusCreated)
+	})
+
+	// NetworkPolicy (fixed name "pi-session-sandbox"): GET → 404, POST → 201
+	mux.HandleFunc("GET /apis/networking.k8s.io/v1/namespaces/poimen/networkpolicies/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	})
+	mux.HandleFunc("POST /apis/networking.k8s.io/v1/namespaces/poimen/networkpolicies", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusCreated)
+	})
+
+	// Pod: first GET (getPodPhase) → 404; second GET (waitPodReady) → Running+Ready; POST → 201
+	mux.HandleFunc("GET /api/v1/namespaces/poimen/pods/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		if atomic.AddInt32(&podGetCount, 1) == 1 {
+			w.WriteHeader(http.StatusNotFound)
+			return
+		}
+		respondK8sJSON(w, http.StatusOK, runningReadyPod(podName(id)))
+	})
+	mux.HandleFunc("POST /api/v1/namespaces/poimen/pods", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusCreated)
+	})
+
+	// Service: GET → 404, POST → 201
+	mux.HandleFunc("GET /api/v1/namespaces/poimen/services/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	})
+	mux.HandleFunc("POST /api/v1/namespaces/poimen/services", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusCreated)
+	})
+
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	result, err := newTestSM(srv.URL).CreateOrResume(context.Background(), id)
+	if err != nil {
+		t.Fatalf("CreateOrResume: %v", err)
+	}
+	if want := svcDNS(id); result.SvcDNS != want {
+		t.Errorf("SvcDNS: want %q, got %q", want, result.SvcDNS)
+	}
+	if result.Resumed {
+		t.Error("want Resumed=false, got true")
+	}
+}
+
+// ── TestSandboxCreateOrResume_Resume ─────────────────────────────────────────
+
+// PVC exists + pod Running → Resumed=true, no POST calls.
+func TestSandboxCreateOrResume_Resume(t *testing.T) {
+	id := "resumeid"
+
+	mux := http.NewServeMux()
+
+	// PVC exists
+	mux.HandleFunc("GET /api/v1/namespaces/poimen/persistentvolumeclaims/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		respondK8sJSON(w, http.StatusOK, map[string]any{"metadata": map[string]any{"name": pvcName(id)}})
+	})
+
+	// Pod exists and Running (getPodPhase uses GET /{name})
+	mux.HandleFunc("GET /api/v1/namespaces/poimen/pods/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		respondK8sJSON(w, http.StatusOK, map[string]any{
+			"metadata": map[string]any{"name": podName(id)},
+			"status":   map[string]any{"phase": "Running"},
+		})
+	})
+
+	// Any POST → 500 to assert no creates are issued
+	mux.HandleFunc("POST /", func(w http.ResponseWriter, r *http.Request) {
+		t.Errorf("unexpected POST to %s", r.URL.Path)
+		w.WriteHeader(http.StatusInternalServerError)
+	})
+
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	result, err := newTestSM(srv.URL).CreateOrResume(context.Background(), id)
+	if err != nil {
+		t.Fatalf("CreateOrResume: %v", err)
+	}
+	if !result.Resumed {
+		t.Error("want Resumed=true, got false")
+	}
+	if want := svcDNS(id); result.SvcDNS != want {
+		t.Errorf("SvcDNS: want %q, got %q", want, result.SvcDNS)
+	}
+}
+
+// ── TestSandboxDelete ─────────────────────────────────────────────────────────
+
+// All DELETEs succeed → no error returned.
+func TestSandboxDelete(t *testing.T) {
+	id := "deleteid"
+
+	mux := http.NewServeMux()
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/pods/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusOK)
+	})
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/services/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusOK)
+	})
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/persistentvolumeclaims/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusOK)
+	})
+
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	if err := newTestSM(srv.URL).Delete(context.Background(), id); err != nil {
+		t.Errorf("Delete: %v", err)
+	}
+}
+
+// ── TestSandboxDelete_NotFound ────────────────────────────────────────────────
+
+// All DELETEs return 404 → isNotFound errors swallowed → no error returned.
+func TestSandboxDelete_NotFound(t *testing.T) {
+	id := "ghostid"
+
+	mux := http.NewServeMux()
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/pods/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	})
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/services/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	})
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/persistentvolumeclaims/{name}", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	})
+
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	if err := newTestSM(srv.URL).Delete(context.Background(), id); err != nil {
+		t.Errorf("Delete_NotFound: want nil, got %v", err)
+	}
+}
+
+// ── TestSandboxReapIdlePods ───────────────────────────────────────────────────
+
+// pod-a=Succeeded → deleted; pod-b=Running → skipped.
+func TestSandboxReapIdlePods(t *testing.T) {
+	var deletedA int32
+
+	mux := http.NewServeMux()
+
+	// List all pi-session pods (query string not part of mux path matching)
+	mux.HandleFunc("GET /api/v1/namespaces/poimen/pods", func(w http.ResponseWriter, _ *http.Request) {
+		respondK8sJSON(w, http.StatusOK, map[string]any{
+			"items": []any{
+				map[string]any{
+					"metadata": map[string]any{"name": "pod-a"},
+					"status":   map[string]any{"phase": "Succeeded"},
+				},
+				map[string]any{
+					"metadata": map[string]any{"name": "pod-b"},
+					"status":   map[string]any{"phase": "Running"},
+				},
+			},
+		})
+	})
+
+	mux.HandleFunc("DELETE /api/v1/namespaces/poimen/pods/{name}", func(w http.ResponseWriter, r *http.Request) {
+		name := r.PathValue("name")
+		if name == "pod-b" {
+			t.Errorf("pod-b (Running) must NOT be deleted")
+		}
+		if name == "pod-a" {
+			atomic.AddInt32(&deletedA, 1)
+		}
+		w.WriteHeader(http.StatusOK)
+	})
+
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	newTestSM(srv.URL).ReapIdlePods(context.Background())
+
+	if got := atomic.LoadInt32(&deletedA); got != 1 {
+		t.Errorf("pod-a delete count: want 1, got %d", got)
+	}
+}
+
+// ── TestSandboxNameHelpers ────────────────────────────────────────────────────
+
+func TestSandboxNameHelpers(t *testing.T) {
+	cases := []struct{ fn, want, got string }{
+		{"pvcName", "workspace-session-abc", pvcName("abc")},
+		{"podName", "pi-session-abc", podName("abc")},
+		{"svcName", "pi-session-abc", svcName("abc")},
+		{"svcDNS", "pi-session-abc.poimen.svc.cluster.local:8787", svcDNS("abc")},
+	}
+	for _, c := range cases {
+		if c.got != c.want {
+			t.Errorf("%s: want %q, got %q", c.fn, c.want, c.got)
+		}
+	}
+}
+
+// ── TestSandboxIsNotFound ─────────────────────────────────────────────────────
+
+func TestSandboxIsNotFound(t *testing.T) {
+	if !isNotFound(&k8sNotFound{}) {
+		t.Error("isNotFound(&k8sNotFound{}) must be true")
+	}
+	if isNotFound(fmt.Errorf("other")) {
+		t.Error("isNotFound(fmt.Errorf) must be false")
+	}
+	if isNotFound(nil) {
+		t.Error("isNotFound(nil) must be false")
+	}
+}

internal/tracing/middleware.go

@@ -36,6 +36,21 @@ func (rw *responseWriter) Write(b []byte) (int, error) {
 	return n, err
 }
 
+// Flush forwards to the wrapped writer so that streaming handlers can still
+// detect http.Flusher through this wrapper. Without it a bare
+// `w.(http.Flusher)` assertion fails for every request that passes through
+// this middleware, which silently disables SSE flushing and keep-alives.
+func (rw *responseWriter) Flush() {
+	if f, ok := rw.ResponseWriter.(http.Flusher); ok {
+		f.Flush()
+	}
+}
+
+// Unwrap exposes the wrapped writer to http.ResponseController.
+func (rw *responseWriter) Unwrap() http.ResponseWriter {
+	return rw.ResponseWriter
+}
+
 // Middleware returns an HTTP middleware that adds tracing to requests.
 func Middleware(next http.Handler) http.Handler {
 	tracer := otel.Tracer(tracerName)

k8s/deployment.yaml

@@ -21,9 +21,10 @@ spec:
       labels:
         app: api-gateway
         component: gateway
-        # Required by the llm-serving-default-deny NetworkPolicy, which admits
-        # only pods labelled llm-client=true (from any namespace) on port 8080.
-        # Without it every upstream dial times out and dispatch returns 502.
+        # The llm-serving-ingress NetworkPolicy in the llm-serving namespace now
+        # selects by namespace (kubernetes.io/metadata.name: api), not by this
+        # label. Kept for older NetworkPolicy revisions that still admit on
+        # llm-client=true.
         llm-client: "true"
       annotations:
         reloader.stakater.com/auto: "true"
@@ -67,6 +68,24 @@ spec:
           value: "5m"
         - name: LOG_LEVEL
           value: "info"
+        - name: ADMISSION_MAX_PARK_SECONDS
+          value: "120"
+        # Must track the engine's --max-num-seqs (homelab: reasoning.yaml).
+        # The gauge this compares against is engine-wide, so this number is a
+        # ceiling on concurrent requests across all gateway replicas, not a
+        # per-pod slot count. Drift here is silent; startup logs the value.
+        - name: ADMISSION_MAX_INFLIGHT
+          value: "4"
+        # Only backends that publish the vLLM scheduler gauges may be observed.
+        # `reasoning` is the only vLLM engine; ornith/qwen-cpu are Ollama and
+        # embeddings/rerank are TEI, so observing them would hold
+        # observer_degraded at 1 forever.
+        - name: ADMISSION_OBSERVE_MODELS
+          value: "reasoning"
+        - name: VLLM_MAX_QUEUE_DEPTH
+          value: "8"
+        - name: MAX_BODY_BYTES
+          value: "1048576"
         # OpenTelemetry tracing configuration
         - name: OTEL_EXPORTER_OTLP_ENDPOINT
           value: "otel-collector.tracing.svc.cluster.local:4317"

k8s/network-policy.yaml

@@ -21,11 +21,17 @@ spec:
     ports:
     - protocol: TCP
       port: 8080
-  # Allow from Prometheus scraping (if in monitoring namespace)
+  # Allow from Prometheus scraping (monitoring namespace)
+  # MUST select on kubernetes.io/metadata.name, not a plain `name:` label.
+  # Only ingress-nginx carries a plain `name` label in this cluster; every
+  # other namespace has just kubernetes.io/metadata.name, so `name: monitoring`
+  # matched nothing and silently denied every Prometheus scrape of :8080
+  # (surfaced as context-deadline-exceeded on all 3 gateway replicas while the
+  # ServiceMonitor itself was discovered and healthy).
   - from:
     - namespaceSelector:
         matchLabels:
-          name: monitoring
+          kubernetes.io/metadata.name: monitoring
     ports:
     - protocol: TCP
       port: 8080
@@ -122,7 +128,7 @@ spec:
     ports:
     - protocol: TCP
       port: 7233
-  # Allow to Poimen (memory:8080 + graphql:8090 + redis:6379)
+  # Allow to Poimen (memory:8080 + graphql:8090 + redis:6379 + fleet-controller:8788)
   - to:
     - namespaceSelector:
         matchLabels:
@@ -134,6 +140,8 @@ spec:
       port: 8090
     - protocol: TCP
       port: 6379
+    - protocol: TCP
+      port: 8788
   # Allow to Gotify (push notifications)
   - to:
     - namespaceSelector:

k8s/rbac.yaml

@@ -26,3 +26,37 @@ roleRef:
   kind: ClusterRole
   name: image-promoter
   apiGroup: rbac.authorization.k8s.io
+---
+# Role + RoleBinding for pi-session sandbox manager (§4.15.4).
+# api-gateway SA needs get/create/delete for pods, pvcs, services,
+# and networkpolicies in the poimen namespace.
+apiVersion: rbac.authorization.k8s.io/v1
+kind: Role
+metadata:
+  name: session-sandbox-manager
+  namespace: poimen
+  labels:
+    app: api-gateway
+rules:
+- apiGroups: [""]
+  resources: ["pods", "persistentvolumeclaims", "services"]
+  verbs: ["get", "list", "create", "delete"]
+- apiGroups: ["networking.k8s.io"]
+  resources: ["networkpolicies"]
+  verbs: ["get", "create", "delete"]
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: RoleBinding
+metadata:
+  name: session-sandbox-manager-api-gateway
+  namespace: poimen
+  labels:
+    app: api-gateway
+subjects:
+- kind: ServiceAccount
+  name: api-gateway
+  namespace: api
+roleRef:
+  kind: Role
+  name: session-sandbox-manager
+  apiGroup: rbac.authorization.k8s.io