package zset

import (
	"context"
	"encoding/json"
	"errors"
	"strings"
	"sync"
	"sync/atomic"
	"testing"
	"time"

	"github.com/gogf/gf/v2/container/gvar"

	queue "git.5iot.com/iotlcm-public/iotlcm-worker-svr/internal/queue/zset"
)

type redisCall struct {
	command string
	args    []any
}

type fakeRedis struct {
	mu        sync.Mutex
	calls     []redisCall
	responses []*gvar.Var
}

func (f *fakeRedis) Do(_ context.Context, command string, args ...any) (*gvar.Var, error) {
	f.mu.Lock()
	defer f.mu.Unlock()
	f.calls = append(f.calls, redisCall{command: command, args: args})
	if len(f.responses) == 0 {
		return gvar.New(int64(1)), nil
	}
	response := f.responses[0]
	f.responses = f.responses[1:]
	return response, nil
}

func (f *fakeRedis) Eval(_ context.Context, script string, numKeys int64, keys []string, args []any) (*gvar.Var, error) {
	allArgs := []any{script, numKeys}
	for _, key := range keys {
		allArgs = append(allArgs, key)
	}
	allArgs = append(allArgs, args...)
	return f.Do(context.Background(), "EVAL", allArgs...)
}

type testHandler struct {
	query func(context.Context, queue.Task) Result
	apply func(context.Context, []Result) error
}

func (h *testHandler) Query(ctx context.Context, task queue.Task) Result {
	if h.query != nil {
		return h.query(ctx, task)
	}
	return Result{Task: task, Success: true, UsageMB: float64(task.CardID)}
}

func (h *testHandler) Apply(ctx context.Context, results []Result) error {
	if h.apply != nil {
		return h.apply(ctx, results)
	}
	return nil
}

func taskMember(task queue.Task) string {
	value, _ := json.Marshal(task)
	return string(value)
}

func claimedTask(task queue.Task) queue.ClaimedTask {
	return queue.ClaimedTask{Task: task, Member: taskMember(task), LeaseToken: "token-" + task.TaskID}
}

func newTestConsumer(t *testing.T, fake *fakeRedis, handler Handler, config Config) *Consumer {
	t.Helper()
	store, err := queue.New(fake, "flow:")
	if err != nil {
		t.Fatal(err)
	}
	consumer, err := New(store, handler, config)
	if err != nil {
		t.Fatal(err)
	}
	return consumer
}

func TestRoundSummaryAccumulatesAndFlushes(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New([]any{"runId", "run-1"})}}
	consumer := newTestConsumer(t, fake, &testHandler{}, Config{RoundRetryEnabled: true})
	started := time.Date(2026, 9, 22, 3, 0, 0, 0, time.UTC)
	consumer.recordRoundBatch("run-1", 1, []queue.ClaimedTask{
		claimedTask(queue.Task{RunID: "run-1", TaskID: "a"}),
		claimedTask(queue.Task{RunID: "run-1", TaskID: "b"}),
	}, []queue.TaskOutcome{
		{Success: true},
		{Retry: true},
	}, started)
	consumer.recordRoundBatch("run-1", 1, []queue.ClaimedTask{
		claimedTask(queue.Task{RunID: "run-1", TaskID: "c"}),
	}, []queue.TaskOutcome{{Success: false}}, started.Add(2*time.Second))

	key := "run-1\x001"
	consumer.summaryMu.Lock()
	summary := *consumer.roundSummaries[key]
	consumer.summaryMu.Unlock()
	if summary.inputCount != 3 || summary.successCount != 1 || summary.failedCount != 2 || summary.deferredCount != 1 || summary.terminalFailedCount != 1 {
		t.Fatalf("unexpected summary: %+v", summary)
	}
	if !summary.startedAt.Equal(started) {
		t.Fatalf("startedAt = %v, want %v", summary.startedAt, started)
	}

	consumer.flushRoundSummary(context.Background(), "run-1", 1)
	consumer.summaryMu.Lock()
	_, exists := consumer.roundSummaries[key]
	consumer.summaryMu.Unlock()
	if exists {
		t.Fatal("round summary was not removed after flush")
	}
}

func TestHandleClaimsAppliesSuccessOnceAndAcknowledgesBatch(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New(int64(2))}}
	applyCalls := 0
	var applied []Result
	handler := &testHandler{apply: func(_ context.Context, results []Result) error {
		applyCalls++
		applied = append(applied, results...)
		return nil
	}}
	consumer := newTestConsumer(t, fake, handler, Config{ClaimBatch: 2, MobileConcurrency: 2, TelecomConcurrency: 1})
	claimed := []queue.ClaimedTask{
		claimedTask(queue.Task{TaskID: "1", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1, CardID: 1}),
		claimedTask(queue.Task{TaskID: "2", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1, CardID: 2}),
	}
	if err := consumer.handleClaims(context.Background(), claimed); err != nil {
		t.Fatal(err)
	}
	if applyCalls != 1 || len(applied) != 2 {
		t.Fatalf("apply calls = %d, results = %d", applyCalls, len(applied))
	}
	if len(fake.calls) != 1 || fake.calls[0].command != "EVAL" {
		t.Fatalf("Redis calls = %+v, want one batch ack", fake.calls)
	}
}

func TestDatabaseFailureLeavesEntireBatchLeased(t *testing.T) {
	fake := &fakeRedis{}
	handler := &testHandler{
		query: func(_ context.Context, task queue.Task) Result {
			if task.CardID == 2 {
				return Result{Task: task, ErrorClass: "business", Code: 11011, Message: "invalid card"}
			}
			return Result{Task: task, Success: true, UsageMB: 1.5}
		},
		apply: func(context.Context, []Result) error { return errors.New("database unavailable") },
	}
	consumer := newTestConsumer(t, fake, handler, Config{ClaimBatch: 2, MobileConcurrency: 2, TelecomConcurrency: 1})
	claimed := []queue.ClaimedTask{
		claimedTask(queue.Task{TaskID: "success", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1, CardID: 1}),
		claimedTask(queue.Task{TaskID: "failed", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1, CardID: 2}),
	}
	err := consumer.handleClaims(context.Background(), claimed)
	if err == nil || err.Error() != "apply monthly usage result batch: database unavailable" {
		t.Fatalf("handleClaims() error = %v", err)
	}
	if len(fake.calls) != 0 {
		t.Fatalf("Redis calls = %+v, want no ACK when database update fails", fake.calls)
	}
}

func TestBusinessFailuresAreTerminalWithoutRetry(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New(int64(2))}}
	failureCount := 0
	handler := &testHandler{query: func(_ context.Context, task queue.Task) Result {
		return Result{Task: task, ErrorClass: "business", Code: 13000, Message: "busy"}
	}}
	consumer := newTestConsumer(t, fake, handler, Config{
		MobileConcurrency:  2,
		TelecomConcurrency: 1,
		OnFailure: func(context.Context, Result) {
			failureCount++
		},
	})
	claimed := []queue.ClaimedTask{
		claimedTask(queue.Task{TaskID: "1", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1, CardID: 1}),
		claimedTask(queue.Task{TaskID: "2", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1, CardID: 2}),
	}
	if err := consumer.handleClaims(context.Background(), claimed); err != nil {
		t.Fatal(err)
	}
	if failureCount != 2 {
		t.Fatalf("failure count = %d, want 2", failureCount)
	}
	if len(fake.calls) != 1 || fake.calls[0].command != "EVAL" {
		t.Fatalf("Redis calls = %+v; business failures must only be acknowledged", fake.calls)
	}
}

func TestRetryableFailuresRemainLeasedWhileSuccessesAreAcknowledged(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New(int64(1))}}
	var applied []Result
	handler := &testHandler{
		query: func(_ context.Context, task queue.Task) Result {
			if task.CardID == 1 {
				return Result{Task: task, ErrorClass: "network", Retryable: true, Message: "temporary failure"}
			}
			return Result{Task: task, Success: true, UsageMB: 2}
		},
		apply: func(_ context.Context, results []Result) error {
			applied = append(applied, results...)
			return nil
		},
	}
	consumer := newTestConsumer(t, fake, handler, Config{MobileConcurrency: 2, TelecomConcurrency: 1})
	claimed := []queue.ClaimedTask{
		claimedTask(queue.Task{TaskID: "retry", RunID: "run", ChanKey: "mobile-a", Operator: 1, CardID: 1}),
		claimedTask(queue.Task{TaskID: "success", RunID: "run", ChanKey: "mobile-a", Operator: 1, CardID: 2}),
	}
	if err := consumer.handleClaims(context.Background(), claimed); err != nil {
		t.Fatal(err)
	}
	if len(applied) != 2 {
		t.Fatalf("applied results = %d, want both result rows for handler", len(applied))
	}
	if len(fake.calls) != 1 || fake.calls[0].command != "EVAL" {
		t.Fatalf("Redis calls = %+v, want one ACK for successful/terminal result", fake.calls)
	}
	if len(fake.calls[0].args) == 0 {
		t.Fatal("ACK call did not include arguments")
	}
}

func TestRunRejectsAnotherActiveConsumer(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New(nil)}}
	consumer := newTestConsumer(t, fake, &testHandler{}, Config{})
	if err := consumer.Run(context.Background()); !errors.Is(err, ErrAlreadyRunning) {
		t.Fatalf("Run error = %v, want ErrAlreadyRunning", err)
	}
}

func TestProcessOnceStartsEveryRegisteredChannel(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{
		gvar.New([]any{"mobile-a", "telecom-a", "mobile-b"}),
		gvar.New([]any{}),
		gvar.New([]any{}),
		gvar.New([]any{}),
	}}
	consumer := newTestConsumer(t, fake, &testHandler{}, Config{})
	processed, err := consumer.processOnce(context.Background())
	if err != nil {
		t.Fatal(err)
	}
	if !processed {
		t.Fatal("processOnce() did not start registered channels")
	}
	deadline := time.Now().Add(time.Second)
	for {
		fake.mu.Lock()
		callCount := len(fake.calls)
		fake.mu.Unlock()
		if callCount == 4 {
			break
		}
		if time.Now().After(deadline) {
			t.Fatalf("Redis calls = %d, want channels + 3 claims", callCount)
		}
		time.Sleep(time.Millisecond)
	}
	consumer.channelGroup.Wait()
	if len(fake.calls) != 4 {
		t.Fatalf("Redis calls = %d, want channels + 3 claims", len(fake.calls))
	}
}

type channelRedis struct {
	mu          sync.Mutex
	claims      map[string][]queue.Task
	claimCounts map[string]int
}

func (r *channelRedis) Do(_ context.Context, command string, _ ...any) (*gvar.Var, error) {
	return nil, errors.New("unexpected command " + command)
}

func (r *channelRedis) Eval(_ context.Context, script string, _ int64, keys []string, args []any) (*gvar.Var, error) {
	r.mu.Lock()
	defer r.mu.Unlock()
	if strings.Contains(script, "ZRANGEBYSCORE") {
		chanKey := strings.TrimPrefix(keys[0], "flow:")
		r.claimCounts[chanKey]++
		tasks := r.claims[chanKey]
		if len(tasks) == 0 {
			return gvar.New([]any{}), nil
		}
		task := tasks[0]
		r.claims[chanKey] = tasks[1:]
		return gvar.New([]any{taskMember(task), "token-" + task.TaskID}), nil
	}
	if len(keys) == 2 && len(args) == 2 {
		return gvar.New(int64(1)), nil
	}
	return nil, errors.New("unexpected eval")
}

func TestSlowChannelDoesNotBlockAnotherChannelFromClaimingAgain(t *testing.T) {
	redis := &channelRedis{
		claims: map[string][]queue.Task{
			"telecom-a": {{TaskID: "t1", RunID: "run", ChanKey: "telecom-a", TempKey: telecomTempKey, Operator: 2}},
			"mobile-a": {
				{TaskID: "m1", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1},
				{TaskID: "m2", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1},
			},
		},
		claimCounts: make(map[string]int),
	}
	telecomStarted := make(chan struct{})
	releaseTelecom := make(chan struct{})
	handler := &testHandler{query: func(_ context.Context, task queue.Task) Result {
		if task.ChanKey == "telecom-a" {
			close(telecomStarted)
			<-releaseTelecom
		}
		return Result{Task: task, Success: true}
	}}
	store, err := queue.New(redis, "flow:")
	if err != nil {
		t.Fatal(err)
	}
	consumer, err := New(store, handler, Config{ClaimBatch: 1, MobileConcurrency: 1, TelecomConcurrency: 1, LeaseTTL: time.Minute})
	if err != nil {
		t.Fatal(err)
	}
	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	consumer.startChannel(ctx, "telecom-a")
	consumer.startChannel(ctx, "mobile-a")
	select {
	case <-telecomStarted:
	case <-time.After(time.Second):
		t.Fatal("telecom channel did not start")
	}
	if consumer.startChannel(ctx, "telecom-a") {
		t.Fatal("channel executor started twice while first executor is running")
	}
	deadline := time.Now().Add(time.Second)
	for {
		redis.mu.Lock()
		mobileClaims := redis.claimCounts["mobile-a"]
		redis.mu.Unlock()
		if mobileClaims >= 2 {
			break
		}
		if time.Now().After(deadline) {
			t.Fatalf("mobile claim count = %d while telecom is blocked", mobileClaims)
		}
		time.Sleep(time.Millisecond)
	}
	close(releaseTelecom)
	consumer.channelGroup.Wait()
}

func TestClaimedBatchRenewsLeaseUntilAcknowledged(t *testing.T) {
	fake := &fakeRedis{}
	handler := &testHandler{query: func(_ context.Context, task queue.Task) Result {
		time.Sleep(45 * time.Millisecond)
		return Result{Task: task, Success: true}
	}}
	consumer := newTestConsumer(t, fake, handler, Config{LeaseTTL: 30 * time.Millisecond, TaskTimeout: 20 * time.Millisecond, MobileConcurrency: 1, TelecomConcurrency: 1})
	claimed := []queue.ClaimedTask{claimedTask(queue.Task{TaskID: "1", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1})}
	if err := consumer.handleClaims(context.Background(), claimed); err != nil {
		t.Fatal(err)
	}
	fake.mu.Lock()
	defer fake.mu.Unlock()
	if len(fake.calls) < 2 {
		t.Fatalf("Redis calls = %d, want lease renewal and ACK", len(fake.calls))
	}
}

func TestLostBatchLeasePreventsApplyAndAck(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New(int64(0))}}
	applyCalls := 0
	handler := &testHandler{
		query: func(ctx context.Context, task queue.Task) Result {
			<-ctx.Done()
			return Result{Task: task, ErrorClass: "canceled"}
		},
		apply: func(context.Context, []Result) error {
			applyCalls++
			return nil
		},
	}
	consumer := newTestConsumer(t, fake, handler, Config{LeaseTTL: 30 * time.Millisecond, TaskTimeout: 20 * time.Millisecond, MobileConcurrency: 1, TelecomConcurrency: 1})
	claimed := []queue.ClaimedTask{claimedTask(queue.Task{TaskID: "1", RunID: "run", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1})}
	err := consumer.handleClaims(context.Background(), claimed)
	if !errors.Is(err, ErrLeaseLost) {
		t.Fatalf("handleClaims() error = %v, want ErrLeaseLost", err)
	}
	if applyCalls != 0 {
		t.Fatalf("apply calls = %d, want 0 after lease loss", applyCalls)
	}
	fake.mu.Lock()
	defer fake.mu.Unlock()
	if len(fake.calls) != 1 {
		t.Fatalf("Redis calls = %d, want only failed lease renewal", len(fake.calls))
	}
}

func newTrackingHandler(release chan struct{}, current, maximum *atomic.Int32) *testHandler {
	return &testHandler{query: func(_ context.Context, task queue.Task) Result {
		active := current.Add(1)
		for {
			observed := maximum.Load()
			if active <= observed || maximum.CompareAndSwap(observed, active) {
				break
			}
		}
		<-release
		current.Add(-1)
		return Result{Task: task, Success: true}
	}}
}

func TestMobileChannelsHaveIndependentConcurrencyPools(t *testing.T) {
	var current atomic.Int32
	var maximum atomic.Int32
	release := make(chan struct{})
	consumer := newTestConsumer(t, &fakeRedis{}, newTrackingHandler(release, &current, &maximum), Config{MobileConcurrency: 2, TelecomConcurrency: 1})
	var group sync.WaitGroup
	for index := 0; index < 3; index++ {
		group.Add(1)
		go func(index int) {
			defer group.Done()
			task := queue.Task{TaskID: string(rune('1' + index)), ChanKey: "mobile-" + string(rune('a'+index)), TempKey: mobileTempKey, CardID: int64(index + 1)}
			_ = consumer.queryClaim(context.Background(), claimedTask(task), carrierMobile, mobileQPS)
		}(index)
	}
	time.Sleep(100 * time.Millisecond)
	if maximum.Load() != 3 {
		t.Fatalf("maximum in-flight across channels = %d, want 3 (one pool per channel)", maximum.Load())
	}
	close(release)
	group.Wait()
}

func TestChannelConcurrencyPoolCapsWithinChannel(t *testing.T) {
	var current atomic.Int32
	var maximum atomic.Int32
	release := make(chan struct{})
	consumer := newTestConsumer(t, &fakeRedis{}, newTrackingHandler(release, &current, &maximum), Config{MobileConcurrency: 2, TelecomConcurrency: 1})
	var group sync.WaitGroup
	for index := 0; index < 3; index++ {
		group.Add(1)
		go func(index int) {
			defer group.Done()
			task := queue.Task{TaskID: string(rune('1' + index)), ChanKey: "mobile-a", TempKey: mobileTempKey, CardID: int64(index + 1)}
			_ = consumer.queryClaim(context.Background(), claimedTask(task), carrierMobile, mobileQPS)
		}(index)
	}
	time.Sleep(100 * time.Millisecond)
	if maximum.Load() != 2 {
		t.Fatalf("maximum in-flight within one channel = %d, want 2", maximum.Load())
	}
	close(release)
	group.Wait()
}

func TestMobileAndTelecomUseIndependentChannelPools(t *testing.T) {
	started := make(chan carrier, 2)
	release := make(chan struct{})
	handler := &testHandler{query: func(_ context.Context, task queue.Task) Result {
		channelCarrier, _ := policyForTemplate(task.TempKey, Config{})
		started <- channelCarrier
		<-release
		return Result{Task: task, Success: true}
	}}
	consumer := newTestConsumer(t, &fakeRedis{}, handler, Config{MobileConcurrency: 1, TelecomConcurrency: 1})
	go func() {
		task := queue.Task{TaskID: "mobile", ChanKey: "mobile-a", TempKey: mobileTempKey, Operator: 1}
		_ = consumer.queryClaim(context.Background(), claimedTask(task), carrierMobile, mobileQPS)
	}()
	go func() {
		task := queue.Task{TaskID: "telecom", ChanKey: "telecom-a", TempKey: telecomTempKey, Operator: 2}
		_ = consumer.queryClaim(context.Background(), claimedTask(task), carrierTelecom, telecomQPS)
	}()

	seen := map[carrier]bool{}
	for len(seen) < 2 {
		select {
		case value := <-started:
			seen[value] = true
		case <-time.After(time.Second):
			t.Fatalf("carrier pools did not progress independently: %+v", seen)
		}
	}
	close(release)
}

func TestUnknownTemplateUsesGenericStrategy(t *testing.T) {
	fake := &fakeRedis{responses: []*gvar.Var{gvar.New(int64(2))}}
	queryCalls := 0
	failures := 0
	applyCalls := 0
	handler := &testHandler{
		query: func(_ context.Context, task queue.Task) Result {
			queryCalls++
			return Result{Task: task, Success: true, UsageMB: 1}
		},
		apply: func(_ context.Context, results []Result) error {
			applyCalls++
			for _, result := range results {
				if !result.Success || result.BlockChannel {
					t.Fatalf("unexpected result = %+v", result)
				}
			}
			return nil
		},
	}
	consumer := newTestConsumer(t, fake, handler, Config{OnFailure: func(context.Context, Result) { failures++ }})
	claimed := []queue.ClaimedTask{
		claimedTask(queue.Task{TaskID: "1", RunID: "run", ChanKey: "unknown-a", TempKey: "unknown-template", Operator: 1, CardID: 1}),
		claimedTask(queue.Task{TaskID: "2", RunID: "run", ChanKey: "unknown-a", TempKey: "unknown-template", Operator: 2, CardID: 2}),
	}
	if err := consumer.handleClaims(context.Background(), claimed); err != nil {
		t.Fatal(err)
	}
	if queryCalls != 2 || failures != 0 || applyCalls != 1 {
		t.Fatalf("query calls = %d, failures = %d, apply calls = %d", queryCalls, failures, applyCalls)
	}
	if len(fake.calls) != 1 || fake.calls[0].command != "EVAL" {
		t.Fatalf("Redis calls = %+v, want one ack script", fake.calls)
	}
	if len(fake.responses) != 0 {
		t.Fatalf("unexpected leftover scripted responses: %+v", fake.responses)
	}
}

func TestChannelPacerLimitsEachChannelIndependently(t *testing.T) {
	pacer := newChannelPacer()
	now := time.Unix(100, 0)
	pacedQPS := 30.0
	if delay := pacer.reserve("mobile-a", pacedQPS, now); delay != 0 {
		t.Fatalf("first reservation delay = %s", delay)
	}
	if delay := pacer.reserve("mobile-a", pacedQPS, now); delay < 33*time.Millisecond || delay > 34*time.Millisecond {
		t.Fatalf("second reservation delay = %s", delay)
	}
	if delay := pacer.reserve("mobile-b", pacedQPS, now); delay != 0 {
		t.Fatalf("independent channel delay = %s", delay)
	}
}

func TestChannelPacerSkipsWhenQPSIsDisabled(t *testing.T) {
	pacer := newChannelPacer()
	now := time.Unix(100, 0)
	if delay := pacer.reserve("mobile-a", 0, now); delay != 0 {
		t.Fatalf("disabled qps delay = %s", delay)
	}
	if len(pacer.next) != 0 {
		t.Fatalf("disabled qps should not reserve channel slots, got %+v", pacer.next)
	}
}

func TestNewDefaultsTaskTimeout(t *testing.T) {
	consumer := newTestConsumer(t, &fakeRedis{}, &testHandler{}, Config{})
	if consumer.config.TaskTimeout != 30*time.Second {
		t.Fatalf("TaskTimeout = %s, want 30s default", consumer.config.TaskTimeout)
	}
	if consumer.config.GenericQPS != defaultGenericQPS {
		t.Fatalf("GenericQPS = %v, want %v default", consumer.config.GenericQPS, defaultGenericQPS)
	}
	if _, err := New(consumer.store, &testHandler{}, Config{LeaseTTL: 10 * time.Second}); err == nil {
		t.Fatal("New() with LeaseTTL below default TaskTimeout should fail")
	}
}

func TestPolicyForTemplateUsesTemplateValues(t *testing.T) {
	tests := []struct {
		name    string
		tempKey string
		carrier carrier
		qps     float64
	}{
		{name: "mobile", tempKey: mobileTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "mobile mock", tempKey: mobileMockKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Tengyu", tempKey: sharedTengyuTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Xuzhou mobile", tempKey: sharedXuzhouMobileTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Lelian", tempKey: sharedLelianTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Qianxun", tempKey: sharedQianxunTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Henan", tempKey: sharedHenanTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Daihe", tempKey: sharedDaiheTempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Beixin", tempKey: sharedBeixinTempKey, carrier: carrierMobile, qps: mobileQPS},
		// lanshang_8000/8003 are OneLink shared-token templates whose names
		// omit the sim_onelink_ prefix. Before this case was added they fell
		// through to carrierGeneric and were throttled to 1.5 QPS, which made a
		// 6k-card status/traffic/IMEI run take over an hour per channel.
		{name: "shared Lanshang 8000", tempKey: sharedLanshang8000TempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Lanshang 8003", tempKey: sharedLanshang8003TempKey, carrier: carrierMobile, qps: mobileQPS},
		{name: "shared Lanshang case and padding", tempKey: " LANSHANG_8000 ", carrier: carrierMobile, qps: mobileQPS},
		{name: "telecom", tempKey: telecomTempKey, carrier: carrierTelecom, qps: telecomQPS},
		{name: "telecom v3", tempKey: " SIM_DX_V3 ", carrier: carrierTelecom, qps: telecomQPS},
		{name: "unknown telecom version", tempKey: "sim_dx_v2_future", carrier: carrierGeneric, qps: defaultGenericQPS},
		{name: "unknown template", tempKey: "unknown-template", carrier: carrierGeneric, qps: defaultGenericQPS},
		{name: "missing template", tempKey: "", carrier: carrierGeneric, qps: defaultGenericQPS},
	}
	for _, test := range tests {
		t.Run(test.name, func(t *testing.T) {
			gotCarrier, gotQPS := policyForTemplate(test.tempKey, Config{})
			if gotCarrier != test.carrier || gotQPS != test.qps {
				t.Fatalf("policyForTemplate(%q) = %q,%v", test.tempKey, gotCarrier, gotQPS)
			}
		})
	}
}

func TestPolicyForTemplateConfigurationOverridesDefaults(t *testing.T) {
	config := Config{
		GenericQPS: 2.5,
		TemplateQPS: map[string]float64{
			" SIM_YD_V5 ":      12,
			"unknown-template": 3,
			"invalid":          0,
		},
	}
	config.setDefaults()

	if gotCarrier, gotQPS := policyForTemplate("sim_yd_v5", config); gotCarrier != carrierMobile || gotQPS != 12 {
		t.Fatalf("mobile override = %q,%v, want mobile,12", gotCarrier, gotQPS)
	}
	if gotCarrier, gotQPS := policyForTemplate("unknown-template", config); gotCarrier != carrierGeneric || gotQPS != 3 {
		t.Fatalf("generic override = %q,%v, want generic,3", gotCarrier, gotQPS)
	}
	if gotCarrier, gotQPS := policyForTemplate("other-template", config); gotCarrier != carrierGeneric || gotQPS != 2.5 {
		t.Fatalf("generic default = %q,%v, want generic,2.5", gotCarrier, gotQPS)
	}
}

// TestFailurePoolAlertThresholdIsObservabilityOnly 覆盖失败池告警：
//   - 未配置阀值（0）时不发 HLEN 查询
//   - 失败数低于阀值时不告警
//   - 超过阀值时发一次 HLEN 并打 WARN，但绝不截断 Run
//
// 告警本身是纯观测行为，所以断言的是「有没有多打 Redis 命令」，
// 而不是「Run 有没有被改坏」——后者必须永远为「没被改坏」。
func TestFailurePoolAlertThresholdIsObservabilityOnly(t *testing.T) {
	newMeta := func(failed int64) queue.RunMeta {
		return queue.RunMeta{RunID: "run-1", Status: "completed_with_failures", FailedCount: failed, CurrentRound: 1}
	}

	cases := []struct {
		name          string
		threshold     int64
		failed        int64
		wantHLenCalls int
	}{
		{"未配置阀值不查询", 0, 1_000_000, 0},
		{"低于阀值不告警", 50000, 49999, 0},
		{"等于阀值不告警", 50000, 50000, 0},
		{"超过阀值告警一次", 50000, 50001, 1},
		{"远超阀值也只告警一次", 50000, 500_000, 1},
	}
	for _, tc := range cases {
		fake := &fakeRedis{}
		consumer := newTestConsumer(t, fake, &testHandler{}, Config{
			FailurePoolAlertThreshold: tc.threshold,
		})
		consumer.reportFailurePoolSize(context.Background(), "run-1", newMeta(tc.failed))

		hlen := 0
		for _, call := range fake.calls {
			if call.command == "HLEN" {
				hlen++
			}
		}
		if hlen != tc.wantHLenCalls {
			t.Fatalf("%s: HLEN 调用 %d 次, want %d", tc.name, hlen, tc.wantHLenCalls)
		}
	}
}

// TestFailurePoolAlertReadsCorrectKey 告警必须查该 Run 自己的失败池键。
func TestFailurePoolAlertReadsCorrectKey(t *testing.T) {
	fake := &fakeRedis{}
	consumer := newTestConsumer(t, fake, &testHandler{}, Config{FailurePoolAlertThreshold: 10})
	consumer.reportFailurePoolSize(context.Background(), "run-abc", queue.RunMeta{
		RunID: "run-abc", Status: "completed_with_failures", FailedCount: 11, CurrentRound: 1,
	})

	var key string
	for _, call := range fake.calls {
		if call.command == "HLEN" {
			if len(call.args) > 0 {
				key, _ = call.args[0].(string)
			}
		}
	}
	want, _ := consumer.store.FailedKey("run-abc")
	if key != want {
		t.Fatalf("HLEN key = %q, want %q", key, want)
	}
}