fix: materialize egress checks before basic tasks

internal/controller/health/scheduler_runner.go

@@ -295,7 +295,10 @@ func (supervisor *ConfiguredSchedulerSupervisor) upstreamDueGroups(
 	egressURLs []string,
 	targets []configuredTargetProfile,
 ) ([]DueSource, error) {
-	groups := []DueSource{basic}
+	// EGRESS and TARGET due reads lazily materialize references from BASIC due
+	// entries. Run them before BASIC so a small batch cannot consume the only
+	// source reference before these independent checks are established.
+	groups := make([]DueSource, 0, 1+len(egressURLs)+len(targets))
 	if egress, supported := supervisor.source.(EgressUpstreamTaskSource); supported {
 		for _, targetURL := range egressURLs {
 			source, err := newEgressUpstreamDueSource(egress, upstreamID, targetURL)
@@ -314,6 +317,7 @@ func (supervisor *ConfiguredSchedulerSupervisor) upstreamDueGroups(
 			groups = append(groups, source)
 		}
 	}
+	groups = append(groups, basic)
 	return groups, nil
 }
 

internal/controller/health/scheduler_runner_test.go

@@ -188,9 +188,50 @@ func TestConfiguredSchedulerSupervisorSchedulesBoundedEgressGroups(t *testing.T)
 		t.Fatalf("NewConfiguredSchedulerSupervisor() = %v", err)
 	}
 	result, err := supervisor.Tick(context.Background())
-	if err != nil || result != (TickResult{Planned: 3, Offered: 3}) || len(sink.tasks) != 1 ||
+	if err != nil || result != (TickResult{Planned: 3, Offered: 3}) || !sink.offeredLevel(healthDomain.LevelEgress) ||
-		sink.tasks[0].Candidate.Level != healthDomain.LevelEgress || source.sources["provider-a"].egressCalls != 2 {
+		source.sources["provider-a"].egressCalls != 2 {
-		t.Fatalf("Tick() = (%+v, %v); source=%+v sink=%+v", result, err, source, sink.tasks)
+		t.Fatalf("Tick() = (%+v, %v); source=%+v sink=%+v", result, err, source, sink.batches)
+	}
+}
+
+func TestConfiguredSchedulerSupervisorMaterializesEveryEgressGroupBeforeBasic(t *testing.T) {
+	now := time.Date(2026, 8, 2, 11, 15, 0, 0, time.UTC)
+	firstURL := "https://egress-one.example/identity"
+	secondURL := "https://egress-two.example/identity"
+	configuration := &config.Config{
+		Defaults: config.Defaults{Check: config.Check{
+			Interval: config.Duration(time.Minute), MaxInFlight: 1, Timeout: config.Duration(time.Second), MaxAttempts: 1,
+			URLs: []string{firstURL, secondURL},
+		}},
+		Upstreams: map[string]config.Upstream{"provider-a": {Enabled: true}},
+	}
+	source := &upstreamTaskSourceStub{sources: map[string]*dueSourceStub{
+		"provider-a": {candidates: []Candidate{{
+			ProxyID: "basic", State: proxyDomain.StateFetched, Level: healthDomain.LevelBasic, DueAt: now,
+		}}, egressCandidates: map[string][]Candidate{
+			firstURL:  {{ProxyID: "egress-one", State: proxyDomain.StateFetched, Level: healthDomain.LevelEgress, TargetURL: firstURL, DueAt: now}},
+			secondURL: {{ProxyID: "egress-two", State: proxyDomain.StateFetched, Level: healthDomain.LevelEgress, TargetURL: secondURL, DueAt: now}},
+		}},
+	}}
+	sink := &taskSinkStub{}
+	supervisor, err := NewConfiguredSchedulerSupervisor(&configurationSourceStub{configuration: configuration}, source, sink,
+		SchedulerRunnerOptions{PollInterval: time.Second, BatchSize: 1, Now: func() time.Time { return now }})
+	if err != nil {
+		t.Fatalf("NewConfiguredSchedulerSupervisor() = %v", err)
+	}
+	for tick := 0; tick < 2; tick++ {
+		if result, err := supervisor.Tick(context.Background()); err != nil || result != (TickResult{Planned: 1, Offered: 1}) {
+			t.Fatalf("Tick(%d) = (%+v, %v)", tick, result, err)
+		}
+	}
+	item := source.sources["provider-a"]
+	if item.egressCalls != 2 || item.dueCalls != 0 || len(sink.batches) != 2 ||
+		sink.batches[0][0].Candidate.Level != healthDomain.LevelEgress || sink.batches[1][0].Candidate.Level != healthDomain.LevelEgress {
+		t.Fatalf("first two ticks did not establish egress groups: source=%+v batches=%+v", item, sink.batches)
+	}
+	if result, err := supervisor.Tick(context.Background()); err != nil || result != (TickResult{Planned: 1, Offered: 1}) ||
+		item.dueCalls != 1 || len(sink.batches) != 3 || sink.batches[2][0].Candidate.Level != healthDomain.LevelBasic {
+		t.Fatalf("basic Tick() = (%+v, %v), source=%+v batches=%+v", result, err, item, sink.batches)
 	}
 }
 
@@ -225,10 +266,9 @@ func TestConfiguredSchedulerSupervisorSchedulesRoutingTargetProfiles(t *testing.
 		t.Fatalf("NewConfiguredSchedulerSupervisor() = %v", err)
 	}
 	result, err := supervisor.Tick(context.Background())
-	if err != nil || result != (TickResult{Planned: 2, Offered: 2}) || len(sink.tasks) != 1 ||
+	if err != nil || result != (TickResult{Planned: 2, Offered: 2}) || !sink.offeredTarget(routingName, targetURL) ||
-		sink.tasks[0].Candidate.Level != healthDomain.LevelTarget || sink.tasks[0].Candidate.RoutingName != routingName ||
 		source.sources["provider-a"].targetCalls != 1 {
-		t.Fatalf("Tick() = (%+v, %v); source=%+v sink=%+v", result, err, source, sink.tasks)
+		t.Fatalf("Tick() = (%+v, %v); source=%+v sink=%+v", result, err, source, sink.batches)
 	}
 }
 
@@ -359,14 +399,39 @@ func targetCandidateKey(routingName, targetURL string) string {
 }
 
 type taskSinkStub struct {
-	tasks []PlannedTask
+	tasks   []PlannedTask
-	offer int
+	batches [][]PlannedTask
+	offer   int
 }
 
 func (sink *taskSinkStub) Offer(_ context.Context, tasks []PlannedTask) (int, error) {
 	sink.tasks = append([]PlannedTask(nil), tasks...)
+	sink.batches = append(sink.batches, append([]PlannedTask(nil), tasks...))
 	if sink.offer != 0 {
 		return sink.offer, nil
 	}
 	return len(tasks), nil
 }
+
+func (sink *taskSinkStub) offeredLevel(level healthDomain.Level) bool {
+	for _, batch := range sink.batches {
+		for _, task := range batch {
+			if task.Candidate.Level == level {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+func (sink *taskSinkStub) offeredTarget(routingName, targetURL string) bool {
+	for _, batch := range sink.batches {
+		for _, task := range batch {
+			if task.Candidate.Level == healthDomain.LevelTarget && task.Candidate.RoutingName == routingName &&
+				task.Candidate.TargetURL == targetURL {
+				return true
+			}
+		}
+	}
+	return false
+}

progress.md

@@ -19,6 +19,8 @@
   前重建并下发递增版本，Gateway 在同一长连接内原子替换视图；构建失败或流中断时才由
   SessionSupervisor 按原有退避重连。该机制复用 `RefreshingSnapshotSource`，不降低
   Delta 的 fail-closed 校验，也不把自动 Drain/ACK 记为已完成。
+- Health Scheduler 现在先建立 EGRESS 与 Routing TARGET 的独立 due 引用，再投递
+  BASIC 任务；在单项小批次下，短 TTL Proxy 不会因 BASIC 先出队而错过首次出口/目标探测。
 
 ## 2026-07-30