fix(api): address copilot review feedback on probe cache key and test stability

web/backend/api/model_status.go

@@ -4,9 +4,11 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"hash/fnv"
 	"net"
 	"net/http"
 	"net/url"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -192,13 +194,17 @@ func modelProbeCacheKey(m *config.ModelConfig) string {
 	protocol, modelID := splitModel(m.Model)
 
 	modelName := strings.ToLower(strings.TrimSpace(m.ModelName))
-	apiBase := strings.ToLower(strings.TrimSpace(modelProbeAPIBase(m)))
+	apiBaseRaw := modelProbeAPIBase(m)
+	apiBase := strings.ToLower(strings.TrimRight(strings.TrimSpace(apiBaseRaw), "/"))
 	authMethod := strings.ToLower(strings.TrimSpace(m.AuthMethod))
 	connectMode := strings.ToLower(strings.TrimSpace(m.ConnectMode))
-	hasAPIKey := strings.TrimSpace(m.APIKey()) != ""
+	apiKeyFingerprint := modelProbeAPIKeyFingerprint(m.APIKey())
 
 	var b strings.Builder
-	b.Grow(len(modelName) + len(protocol) + len(modelID) + len(apiBase) + len(authMethod) + len(connectMode) + 8)
+	b.Grow(
+		len(modelName) + len(protocol) + len(modelID) + len(apiBase) + len(authMethod) +
+			len(connectMode) + len(apiKeyFingerprint) + 8,
+	)
 	b.WriteString(modelName)
 	b.WriteByte('|')
 	b.WriteString(protocol)
@@ -211,15 +217,22 @@ func modelProbeCacheKey(m *config.ModelConfig) string {
 	b.WriteByte('|')
 	b.WriteString(connectMode)
 	b.WriteByte('|')
-	if hasAPIKey {
+	b.WriteString(apiKeyFingerprint)
-		b.WriteByte('1')
-	} else {
-		b.WriteByte('0')
-	}
 
 	return b.String()
 }
 
+func modelProbeAPIKeyFingerprint(raw string) string {
+	apiKey := strings.TrimSpace(raw)
+	if apiKey == "" {
+		return "none"
+	}
+
+	h := fnv.New64a()
+	_, _ = h.Write([]byte(apiKey))
+	return strconv.FormatUint(h.Sum64(), 36)
+}
+
 func (s *modelProbeCacheState) getCachedResult(cacheKey string, now time.Time) (bool, bool) {
 	s.mu.RLock()
 	defer s.mu.RUnlock()

web/backend/api/model_status_test.go

@@ -89,6 +89,46 @@ func TestProbeLocalModelAvailability_LMStudioUsesOpenAICompatibleProbe(t *testin
 	}
 }
 
+func TestModelProbeCacheKey_DifferentAPIKeysProduceDifferentKeys(t *testing.T) {
+	base := &config.ModelConfig{
+		ModelName:   "local-vllm",
+		Model:       "vllm/custom-model",
+		APIBase:     "http://127.0.0.1:8000/v1",
+		AuthMethod:  "local",
+		ConnectMode: "",
+	}
+
+	m1 := *base
+	m1.SetAPIKey("key-a")
+	m2 := *base
+	m2.SetAPIKey("key-b")
+
+	k1 := modelProbeCacheKey(&m1)
+	k2 := modelProbeCacheKey(&m2)
+	if k1 == k2 {
+		t.Fatal("modelProbeCacheKey() should differ when api key changes")
+	}
+}
+
+func TestModelProbeCacheKey_NormalizesTrailingSlashInAPIBase(t *testing.T) {
+	m1 := &config.ModelConfig{
+		ModelName: "local-vllm",
+		Model:     "vllm/custom-model",
+		APIBase:   "http://127.0.0.1:8000/v1",
+	}
+	m2 := &config.ModelConfig{
+		ModelName: "local-vllm",
+		Model:     "vllm/custom-model",
+		APIBase:   "http://127.0.0.1:8000/v1/",
+	}
+
+	k1 := modelProbeCacheKey(m1)
+	k2 := modelProbeCacheKey(m2)
+	if k1 != k2 {
+		t.Fatalf("modelProbeCacheKey() mismatch for equivalent api_base values: %q vs %q", k1, k2)
+	}
+}
+
 func TestProbeLocalModelAvailability_SuccessBackoff(t *testing.T) {
 	resetModelProbeHooks(t)
 
@@ -272,23 +312,27 @@ func TestProbeLocalModelAvailability_DeduplicatesInflightProbe(t *testing.T) {
 	const workers = 8
 	var wg sync.WaitGroup
 	results := make(chan bool, workers)
+	workerStarted := make(chan struct{}, workers)
 
 	for range workers {
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
+			workerStarted <- struct{}{}
 			results <- probeLocalModelAvailability(model)
 		}()
 	}
 
+	for range workers {
+		<-workerStarted
+	}
+
 	select {
 	case <-probeStarted:
 	case <-time.After(200 * time.Millisecond):
 		t.Fatal("probe did not start in time")
 	}
 
-	// Give waiting goroutines a short window to join the same inflight call.
-	time.Sleep(50 * time.Millisecond)
 	if got := atomic.LoadInt32(&calls); got != 1 {
 		t.Fatalf("concurrent probe calls = %d, want 1", got)
 	}
@@ -302,6 +346,10 @@ func TestProbeLocalModelAvailability_DeduplicatesInflightProbe(t *testing.T) {
 			t.Fatal("deduplicated probe result = false, want true")
 		}
 	}
+
+	if got := atomic.LoadInt32(&calls); got != 1 {
+		t.Fatalf("final probe calls = %d, want 1", got)
+	}
 }
 
 func TestOllamaModelMatches_WithTagRequiresExactTag(t *testing.T) {