Expose AI infrastructure status

README.md

@@ -61,6 +61,8 @@ domain metadata fields in `input`, but the worker only reads chat fields such as
 - `GET /api/v1/stats` returns queue and error counters.
 - `GET /api/v1/providers/status` checks configured AI providers without
   returning secrets.
+- `GET /api/v1/infra/status` returns AI-server sidecar telemetry
+  (GPU, containers, vLLM and WhisperX live metrics) when configured.
 - `GET /healthz` returns process health.
 - `GET /readyz` checks PostgreSQL readiness.
 

internal/config/config.go

@@ -14,12 +14,14 @@ type Config struct {
 	MigrateOnStart bool
 	APIAuthToken   string
 
-	LLMBaseURL      string
+	LLMBaseURL        string
-	LLMAPIKey       string
+	LLMAPIKey         string
-	LLMModel        string
+	LLMModel          string
-	LLMTimeout      time.Duration
+	LLMTimeout        time.Duration
-	WhisperXURL     string
+	WhisperXURL       string
-	WhisperXTimeout time.Duration
+	WhisperXTimeout   time.Duration
+	AIStatsSidecarURL string
+	AIStatsTimeout    time.Duration
 
 	WorkerID            string
 	WorkerPollInterval  time.Duration
@@ -37,12 +39,14 @@ func Load() Config {
 		MigrateOnStart: envBool("MIGRATE_ON_START", true),
 		APIAuthToken:   envString("AI_SERVICE_TOKEN", ""),
 
-		LLMBaseURL:      envString("LLM_BASE_URL", ""),
+		LLMBaseURL:        envString("LLM_BASE_URL", ""),
-		LLMAPIKey:       envString("LLM_API_KEY", ""),
+		LLMAPIKey:         envString("LLM_API_KEY", ""),
-		LLMModel:        envString("LLM_MODEL", "qwen2.5-14b"),
+		LLMModel:          envString("LLM_MODEL", "qwen2.5-14b"),
-		LLMTimeout:      envDuration("LLM_TIMEOUT", 5*time.Minute),
+		LLMTimeout:        envDuration("LLM_TIMEOUT", 5*time.Minute),
-		WhisperXURL:     envString("WHISPERX_URL", ""),
+		WhisperXURL:       envString("WHISPERX_URL", ""),
-		WhisperXTimeout: envDuration("WHISPERX_TIMEOUT", 10*time.Minute),
+		WhisperXTimeout:   envDuration("WHISPERX_TIMEOUT", 10*time.Minute),
+		AIStatsSidecarURL: envString("AI_STATS_SIDECAR_URL", ""),
+		AIStatsTimeout:    envDuration("AI_STATS_TIMEOUT", 8*time.Second),
 
 		WorkerID:            envString("WORKER_ID", hostname()),
 		WorkerPollInterval:  envDuration("WORKER_POLL_INTERVAL", 2*time.Second),

internal/httpapi/infra.go

@@ -0,0 +1,63 @@
+package httpapi
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+)
+
+type infraStatusResponse struct {
+	At           time.Time      `json:"at"`
+	Sidecar      map[string]any `json:"sidecar,omitempty"`
+	SidecarError string         `json:"sidecar_error,omitempty"`
+}
+
+func (s *Server) handleInfraStatus(w http.ResponseWriter, r *http.Request) {
+	resp := infraStatusResponse{At: time.Now().UTC()}
+	baseURL := strings.TrimRight(strings.TrimSpace(s.cfg.AIStatsSidecarURL), "/")
+	if baseURL == "" {
+		resp.SidecarError = "AI stats sidecar is not configured"
+		writeJSON(w, http.StatusOK, resp)
+		return
+	}
+
+	timeout := s.cfg.AIStatsTimeout
+	if timeout <= 0 {
+		timeout = 8 * time.Second
+	}
+	ctx, cancel := contextWithTimeout(r, timeout)
+	defer cancel()
+	sidecar, err := fetchAIStatsSidecar(ctx, baseURL, timeout)
+	if err != nil {
+		resp.SidecarError = err.Error()
+	} else {
+		resp.Sidecar = sidecar
+	}
+	writeJSON(w, http.StatusOK, resp)
+}
+
+func fetchAIStatsSidecar(ctx context.Context, baseURL string, timeout time.Duration) (map[string]any, error) {
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, baseURL+"/stats", nil)
+	if err != nil {
+		return nil, err
+	}
+	client := &http.Client{Timeout: timeout}
+	res, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer res.Body.Close()
+	if res.StatusCode >= 300 {
+		body, _ := io.ReadAll(io.LimitReader(res.Body, 2048))
+		return nil, fmt.Errorf("sidecar HTTP %d: %s", res.StatusCode, strings.TrimSpace(string(body)))
+	}
+	var out map[string]any
+	if err := json.NewDecoder(res.Body).Decode(&out); err != nil {
+		return nil, err
+	}
+	return out, nil
+}

internal/httpapi/server.go

@@ -67,6 +67,8 @@ func (s *Server) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		s.handleStats(w, r)
 	case r.Method == http.MethodGet && path == "/api/v1/providers/status":
 		s.handleProviderStatus(w, r)
+	case r.Method == http.MethodGet && path == "/api/v1/infra/status":
+		s.handleInfraStatus(w, r)
 	default:
 		writeError(w, http.StatusNotFound, "not found")
 	}

k8s/configmap.yaml

@@ -13,6 +13,8 @@ data:
   LLM_TIMEOUT: "5m"
   WHISPERX_URL: "http://10.2.3.5:8001"
   WHISPERX_TIMEOUT: "10m"
+  AI_STATS_SIDECAR_URL: "http://10.2.3.5:9090"
+  AI_STATS_TIMEOUT: "8s"
   WORKER_POLL_INTERVAL: "2s"
   WORKER_CLAIM_LIMIT: "4"
   WORKER_LEASE_TIMEOUT: "15m"