fix(ratelimit): no longer stall after rate limit reset

internal/controller/shared.go

@@ -35,8 +35,10 @@ func handleRequeueError(ctx context.Context, err error) (controllerruntime.Resul
 		resetTime = &ownRateLimitErr.ResetTime
 	}
 	if resetTime != nil {
-		logPkg.FromContext(ctx).Info("GitHub API rate limit reached, requeuing after reset time", "resetTime", *resetTime)
-		return controllerruntime.Result{RequeueAfter: time.Until(*resetTime)}, nil
+		// requeue after at least 30 seconds
+		delay := max(time.Until(*resetTime), 30*time.Second)
+		logPkg.FromContext(ctx).Info("GitHub API rate limit reached, requeuing after reset time", "resetTime", *resetTime, "requeueAfter", delay)
+		return controllerruntime.Result{RequeueAfter: delay}, nil
 	}
 
 	// requeue because spreading requires it

internal/controller/shared_test.go

@@ -215,7 +215,7 @@ var _ = Describe("Shared Controller Functions", func() {
 		})
 
 		Context("rate limit error with past reset time", func() {
-			It("should handle reset time in the past gracefully", func() {
+			It("should apply minimum backoff when reset time is in the past", func() {
 				resetTime := time.Now().Add(-1 * time.Minute)
 				err := &ghclient.RateLimitedError{
 					ResetTime: resetTime,
@@ -224,8 +224,8 @@ var _ = Describe("Shared Controller Functions", func() {
 				result, returnErr := handleRequeueError(ctx, err)
 
 				Expect(returnErr).ToNot(HaveOccurred())
-				// RequeueAfter will be negative, which controller-runtime interprets as immediate
-				Expect(result.RequeueAfter).To(BeNumerically("<", 0))
+				// Minimum backoff of 30s applied to avoid tight-loop
+				Expect(result.RequeueAfter).To(Equal(30 * time.Second))
 			})
 		})
 	})

internal/ratelimit/org_registry.go

@@ -221,8 +221,14 @@ func (r *OrgRateLimitRegistry) ShouldStall(orgLogin string, categories ...Catego
 			continue
 		}
 		if catState.Remaining < threshold {
+			// If the reset time (plus grace period) has already passed, the rate limit window
+			// has renewed and the cached Remaining value is stale. Do NOT stall — let the
+			// request through so the rateLimitTrackerTransport can record fresh headers.
+			delay := time.Until(catState.ResetTime.Add(r.config.ResetGracePeriod))
+			if delay <= 0 {
+				continue
+			}
 			stalled = true
-			delay := max(time.Until(catState.ResetTime.Add(r.config.ResetGracePeriod)), 0)
 			if delay > maxDelay {
 				maxDelay = delay
 			}

internal/ratelimit/org_registry_test.go

@@ -131,12 +131,12 @@ var _ = Describe("OrgRateLimitRegistry", func() {
 		})
 
 		Context("when reset time is already in the past", func() {
-			It("returns stalled=true but with zero delay", func() {
+			It("returns stalled=false because the rate limit window has renewed", func() {
 				past := time.Now().Add(-5 * time.Minute)
 				registry.Update("my-org", ratelimit.CategoryCore, 5, 5000, past, 1)
 
 				stalled, delay := registry.ShouldStall("my-org", ratelimit.CategoryCore)
-				Expect(stalled).To(BeTrue())
+				Expect(stalled).To(BeFalse())
 				Expect(delay).To(Equal(time.Duration(0)))
 			})
 		})