package store

import "testing"

func TestNormalizeModelMatchKeyRemovesWhitespace(t *testing.T) {
	got := normalizeModelMatchKey(" doubao-5.0 图像\t编辑\n")
	if got != "doubao-5.0图像编辑" {
		t.Fatalf("expected whitespace-insensitive model key, got %q", got)
	}
}

func TestTaskBillingModelIdentityKeepsRequestedModelPrimary(t *testing.T) {
	identity := taskBillingModelIdentity(GatewayTask{
		Model:          "doubao-5.0 图像编辑",
		RequestedModel: "doubao-5.0 图像编辑",
		ResolvedModel:  "doubao-image-real",
		Metrics: map[string]any{
			"modelAlias":    "doubao-5.0图像编辑",
			"modelName":     "doubao-image-real",
			"providerModel": "doubao-provider-image",
		},
	})

	if identity.Model != "doubao-5.0 图像编辑" || identity.RequestedModel != "doubao-5.0 图像编辑" {
		t.Fatalf("expected persisted model to keep requested model, got %+v", identity)
	}
	if identity.ResolvedModel != "doubao-5.0图像编辑" {
		t.Fatalf("expected resolved model to keep system alias, got %+v", identity)
	}
	if identity.ModelName != "doubao-image-real" || identity.ProviderModelName != "doubao-provider-image" {
		t.Fatalf("expected model name/provider model to stay available, got %+v", identity)
	}
}

func TestTaskBillingModelIdentityFallsBackToBillingLines(t *testing.T) {
	identity := taskBillingModelIdentity(GatewayTask{
		Model: "front end alias",
		Billings: []any{
			map[string]any{
				"model":      "system-model-name",
				"modelAlias": "System Model Alias",
			},
		},
	})

	if identity.Model != "front end alias" || identity.ModelName != "system-model-name" || identity.ResolvedModel != "System Model Alias" {
		t.Fatalf("expected billing lines to provide system model identity, got %+v", identity)
	}
}

func TestRuntimeCandidateLoadUsesMaxLimitedMetric(t *testing.T) {
	candidate := RuntimeModelCandidate{}
	applyRuntimeCandidateLoad(&candidate, runtimeCandidateLoadInput{
		Policy: map[string]any{"rules": []any{
			map[string]any{"metric": "rpm", "limit": 100},
			map[string]any{"metric": "tpm_total", "limit": 1000},
			map[string]any{"metric": "concurrent", "limit": 10},
		}},
		RPMUsed:          40,
		RPMReserved:      10,
		TPMUsed:          900,
		ConcurrentActive: 3,
		QueuedWaiting:    2,
	})

	if !candidate.LoadLimited {
		t.Fatal("expected load to be limited when rate limit rules exist")
	}
	if candidate.LoadMetrics.RPMRatio != 0.5 {
		t.Fatalf("expected rpm ratio 0.5, got %v", candidate.LoadMetrics.RPMRatio)
	}
	if candidate.LoadMetrics.TPMRatio != 0.9 {
		t.Fatalf("expected tpm ratio 0.9, got %v", candidate.LoadMetrics.TPMRatio)
	}
	if candidate.LoadMetrics.ConcurrentRatio != 0.5 {
		t.Fatalf("expected concurrent ratio 0.5, got %v", candidate.LoadMetrics.ConcurrentRatio)
	}
	if candidate.LoadRatio != 0.9 {
		t.Fatalf("expected max load ratio 0.9, got %v", candidate.LoadRatio)
	}
}

func TestRuntimeCandidateSortingAvoidsFullCandidatesButKeepsFallback(t *testing.T) {
	candidates := []RuntimeModelCandidate{
		{
			PlatformID:       "high-priority-full",
			PlatformPriority: 1,
			LoadLimited:      true,
			LoadRatio:        1.2,
		},
		{
			PlatformID:       "lower-priority-available",
			PlatformPriority: 50,
			LoadLimited:      true,
			LoadRatio:        0.2,
		},
	}

	sortRuntimeModelCandidates(candidates)

	if candidates[0].PlatformID != "lower-priority-available" {
		t.Fatalf("expected non-full candidate to be tried first, got %+v", candidates)
	}
	if candidates[1].PlatformID != "high-priority-full" || !candidates[1].LoadAvoided {
		t.Fatalf("expected full high-priority candidate to remain as avoided fallback, got %+v", candidates)
	}
}

func TestRuntimeCandidateSortingPrefersCacheAffinityWithinAvailableCandidates(t *testing.T) {
	candidates := []RuntimeModelCandidate{
		{
			PlatformID:       "base-priority",
			PlatformPriority: 10,
		},
		{
			PlatformID:       "cache-affinity",
			PlatformPriority: 20,
		},
	}
	applyRuntimeCandidateCacheAffinity(&candidates[0], ListModelCandidatesOptions{
		CacheAffinityKey:    "explicit:test",
		CacheAffinityPolicy: map[string]any{"enabled": true, "minSamples": 3, "maxPriorityBoost": 20, "modelTypes": []any{"text_generate"}},
	}, runtimeCandidateCacheAffinityInput{
		RequestCount: 1,
		EMAHitRatio:  1,
	})
	applyRuntimeCandidateCacheAffinity(&candidates[1], ListModelCandidatesOptions{
		CacheAffinityKey:    "explicit:test",
		CacheAffinityPolicy: map[string]any{"enabled": true, "minSamples": 3, "maxPriorityBoost": 20, "modelTypes": []any{"text_generate"}},
	}, runtimeCandidateCacheAffinityInput{
		RequestCount: 3,
		EMAHitRatio:  0.9,
	})

	sortRuntimeModelCandidates(candidates)

	if candidates[0].PlatformID != "cache-affinity" || !candidates[0].CacheAffinity.Applied {
		t.Fatalf("expected cache affinity candidate first, got %+v", candidates)
	}
	if candidates[1].PlatformID != "base-priority" || candidates[1].CacheAffinity.Applied {
		t.Fatalf("expected low sample candidate to keep base priority, got %+v", candidates)
	}
}

func TestRuntimeCandidateSortingPromotesObservedCacheHitAbovePriority(t *testing.T) {
	candidates := []RuntimeModelCandidate{
		{
			PlatformID:       "volces-priority",
			PlatformPriority: 1,
		},
		{
			PlatformID:       "deepseek-cache-hit",
			PlatformPriority: 100,
		},
	}
	policy := map[string]any{"enabled": true, "minSamples": 3, "maxPriorityBoost": 20, "modelTypes": []any{"text_generate"}}
	applyRuntimeCandidateCacheAffinity(&candidates[0], ListModelCandidatesOptions{
		CacheAffinityKey:    "conversation:test",
		CacheAffinityPolicy: policy,
	}, runtimeCandidateCacheAffinityInput{})
	applyRuntimeCandidateCacheAffinity(&candidates[1], ListModelCandidatesOptions{
		CacheAffinityKey:    "conversation:test",
		CacheAffinityPolicy: policy,
	}, runtimeCandidateCacheAffinityInput{
		RequestCount:      1,
		InputTokens:       7945,
		CachedInputTokens: 7296,
		EMAHitRatio:       0.918,
		LastHitRatio:      0.918,
	})

	sortRuntimeModelCandidates(candidates)

	if candidates[0].PlatformID != "deepseek-cache-hit" || !candidates[0].CacheAffinity.Applied {
		t.Fatalf("expected observed cache hit to outrank platform priority, got %+v", candidates)
	}
	if candidates[0].CacheAffinity.Score <= 0 || candidates[0].CacheAffinity.Boost <= 0 {
		t.Fatalf("expected observed cache hit to carry score and boost, got %+v", candidates[0].CacheAffinity)
	}
	if candidates[1].PlatformID != "volces-priority" || candidates[1].CacheAffinity.Applied {
		t.Fatalf("expected priority-only candidate second, got %+v", candidates)
	}
}

func TestRuntimeCandidateSortingOrdersMultipleCacheAffinityCandidatesByScore(t *testing.T) {
	candidates := []RuntimeModelCandidate{
		{
			PlatformID:       "lower-cache-score",
			PlatformPriority: 1,
		},
		{
			PlatformID:       "higher-cache-score",
			PlatformPriority: 50,
		},
	}
	policy := map[string]any{"enabled": true, "minSamples": 1, "maxPriorityBoost": 20, "modelTypes": []any{"text_generate"}}
	applyRuntimeCandidateCacheAffinity(&candidates[0], ListModelCandidatesOptions{
		CacheAffinityKey:    "conversation:test",
		CacheAffinityPolicy: policy,
	}, runtimeCandidateCacheAffinityInput{
		RequestCount:      2,
		InputTokens:       8000,
		CachedInputTokens: 1000,
		EMAHitRatio:       0.5,
		LastHitRatio:      0.5,
	})
	applyRuntimeCandidateCacheAffinity(&candidates[1], ListModelCandidatesOptions{
		CacheAffinityKey:    "conversation:test",
		CacheAffinityPolicy: policy,
	}, runtimeCandidateCacheAffinityInput{
		RequestCount:      1,
		InputTokens:       8000,
		CachedInputTokens: 7000,
		EMAHitRatio:       0.75,
		LastHitRatio:      0.75,
	})

	sortRuntimeModelCandidates(candidates)

	if candidates[0].PlatformID != "higher-cache-score" {
		t.Fatalf("expected higher cache score candidate first, got %+v", candidates)
	}
	if candidates[0].CacheAffinity.Score <= candidates[1].CacheAffinity.Score {
		t.Fatalf("expected score to drive cache-affinity order, got %+v", candidates)
	}
}

func TestRuntimeCandidateSortingKeepsFullCacheAffinityCandidateAvoided(t *testing.T) {
	candidates := []RuntimeModelCandidate{
		{
			PlatformID:       "cache-affinity-full",
			PlatformPriority: 50,
			LoadLimited:      true,
			LoadRatio:        1.2,
		},
		{
			PlatformID:       "available",
			PlatformPriority: 10,
		},
	}
	applyRuntimeCandidateCacheAffinity(&candidates[0], ListModelCandidatesOptions{
		CacheAffinityKey:    "explicit:test",
		CacheAffinityPolicy: map[string]any{"enabled": true, "minSamples": 1, "maxPriorityBoost": 100, "modelTypes": []any{"text_generate"}},
	}, runtimeCandidateCacheAffinityInput{
		RequestCount: 10,
		EMAHitRatio:  1,
	})

	sortRuntimeModelCandidates(candidates)

	if candidates[0].PlatformID != "available" {
		t.Fatalf("expected available candidate before full cache-affinity candidate, got %+v", candidates)
	}
	if candidates[1].PlatformID != "cache-affinity-full" || !candidates[1].LoadAvoided {
		t.Fatalf("expected full cache-affinity candidate to remain avoided fallback, got %+v", candidates)
	}
}

func TestDefaultRunnerPriorityDemotePolicyUsesAutoMode(t *testing.T) {
	policy := defaultRunnerPriorityDemotePolicy()

	if _, ok := policy["demoteStep"]; ok {
		t.Fatal("priority demotion should be automatic and must not expose a demoteStep policy")
	}
	if policy["enabled"] != true {
		t.Fatalf("expected default priority demotion to stay enabled, got %+v", policy["enabled"])
	}
	if policy["deprecated"] != true || policy["replacedBy"] != "failoverPolicy.actionRules" {
		t.Fatalf("expected legacy priority demotion policy to be marked deprecated, got %+v", policy)
	}
}

func TestDefaultRunnerFailoverPolicyCarriesComprehensiveDefaultsInActionRules(t *testing.T) {
	policy := defaultRunnerFailoverPolicy()
	if policy["legacyFieldsDeprecated"] != true || policy["replacedBy"] != "actionRules" {
		t.Fatalf("expected legacy failover summary fields to be marked deprecated, got %+v", policy)
	}
	rules, ok := policy["actionRules"].([]any)
	if !ok || len(rules) < 5 {
		t.Fatalf("expected default action rules to be present, got %+v", policy["actionRules"])
	}

	cooldownRule := defaultActionRuleByAction(t, rules, "cooldown_and_next")
	for _, code := range []string{"rate_limit", "too_many_requests", "resource_exhausted", "throttled", "quota_exceeded", "rpm_limit", "tpm_limit"} {
		if !actionRuleHasString(cooldownRule, "codes", code) {
			t.Fatalf("expected cooldown rule to include rate-limit code %q, got %+v", code, cooldownRule["codes"])
		}
	}
	for _, keyword := range []string{"rate limit", "too many requests", "resource exhausted", "request limit", "requests per minute", "tokens per minute"} {
		if !actionRuleHasString(cooldownRule, "keywords", keyword) {
			t.Fatalf("expected cooldown rule to include rate-limit keyword %q, got %+v", keyword, cooldownRule["keywords"])
		}
	}
	if !actionRuleHasInt(cooldownRule, "statusCodes", 429) {
		t.Fatalf("expected cooldown rule to include 429, got %+v", cooldownRule["statusCodes"])
	}

	disableRule := defaultActionRuleByAction(t, rules, "disable_and_next")
	for _, code := range []string{"auth_failed", "invalid_api_key", "missing_credentials", "account_disabled", "billing_not_active", "permission_denied"} {
		if !actionRuleHasString(disableRule, "codes", code) {
			t.Fatalf("expected disable rule to include auth/platform code %q, got %+v", code, disableRule["codes"])
		}
	}
	for _, keyword := range []string{"invalid api key", "unauthorized", "credential", "permission denied", "account disabled", "billing_not_active", "余额不足"} {
		if !actionRuleHasString(disableRule, "keywords", keyword) {
			t.Fatalf("expected disable rule to include auth/platform keyword %q, got %+v", keyword, disableRule["keywords"])
		}
	}
	for _, status := range []int{401, 403} {
		if !actionRuleHasInt(disableRule, "statusCodes", status) {
			t.Fatalf("expected disable rule to include status %d, got %+v", status, disableRule["statusCodes"])
		}
	}

	demoteRule := defaultActionRuleByAction(t, rules, "demote_and_next")
	for _, category := range []string{"timeout", "provider_5xx", "provider_overloaded"} {
		if !actionRuleHasString(demoteRule, "categories", category) {
			t.Fatalf("expected demote rule to include transient category %q, got %+v", category, demoteRule["categories"])
		}
	}
	for _, code := range []string{"timeout", "server_error", "overloaded", "provider_failed", "invalid_response", "script_timeout", "upload_failed"} {
		if !actionRuleHasString(demoteRule, "codes", code) {
			t.Fatalf("expected demote rule to include transient code %q, got %+v", code, demoteRule["codes"])
		}
	}
	for _, status := range []int{408, 500, 502, 503, 504, 520, 522, 524, 529, 530} {
		if !actionRuleHasInt(demoteRule, "statusCodes", status) {
			t.Fatalf("expected demote rule to include transient status %d, got %+v", status, demoteRule["statusCodes"])
		}
	}
	for _, keyword := range []string{"deadline exceeded", "temporarily unavailable", "service unavailable", "bad gateway", "upstream timeout", "try again later"} {
		if !actionRuleHasString(demoteRule, "keywords", keyword) {
			t.Fatalf("expected demote rule to include transient keyword %q, got %+v", keyword, demoteRule["keywords"])
		}
	}

	nextRule := defaultActionRuleByAction(t, rules, "next")
	for _, category := range []string{"network", "stream_error"} {
		if !actionRuleHasString(nextRule, "categories", category) {
			t.Fatalf("expected next rule to include network category %q, got %+v", category, nextRule["categories"])
		}
	}
	for _, code := range []string{"network", "stream_read_error", "upload_network", "upload_config_failed", "upload_source_fetch_failed"} {
		if !actionRuleHasString(nextRule, "codes", code) {
			t.Fatalf("expected next rule to include network code %q, got %+v", code, nextRule["codes"])
		}
	}
	for _, keyword := range []string{"socket hang up", "connection reset", "econnreset", "enotfound", "unexpected eof", "stream error", "fetch failed"} {
		if !actionRuleHasString(nextRule, "keywords", keyword) {
			t.Fatalf("expected next rule to include network keyword %q, got %+v", keyword, nextRule["keywords"])
		}
	}

	stopRule := defaultActionRuleByAction(t, rules, "stop")
	for _, code := range []string{"bad_request", "invalid_request", "invalid_parameter", "missing_required", "unsupported_kind", "unsupported_model", "request_asset_input_format_unsupported", "upload_source_too_large", "invalid_multipart_image", "script_error", "insufficient_balance", "permission_denied"} {
		if !actionRuleHasString(stopRule, "codes", code) {
			t.Fatalf("expected stop rule to include legacy hard-stop code %q, got %+v", code, stopRule["codes"])
		}
	}
	for _, status := range []int{400, 404, 405, 409, 413, 415, 422} {
		if !actionRuleHasInt(stopRule, "statusCodes", status) {
			t.Fatalf("expected stop rule to include request-error status %d, got %+v", status, stopRule["statusCodes"])
		}
	}
	for _, keyword := range []string{"invalid parameter", "missing required", "not supported", "image is required", "too large", "not found"} {
		if !actionRuleHasString(stopRule, "keywords", keyword) {
			t.Fatalf("expected stop rule to include hard-stop keyword %q, got %+v", keyword, stopRule["keywords"])
		}
	}
}

func TestLegacyDefaultRunnerActionRulesAreDetectedForUpgrade(t *testing.T) {
	legacy := []any{
		map[string]any{
			"enabled":    true,
			"categories": []any{"network", "timeout", "stream_error", "provider_5xx", "provider_overloaded"},
			"action":     "next",
		},
		map[string]any{
			"enabled":         true,
			"categories":      []any{"rate_limit"},
			"codes":           []any{"rate_limit"},
			"statusCodes":     []any{429},
			"keywords":        []any{"rate_limit", "429", "too many requests"},
			"action":          "cooldown_and_next",
			"target":          "model",
			"cooldownSeconds": 300,
		},
		map[string]any{
			"enabled":     true,
			"keywords":    []any{"rate limit", "too many requests", "temporarily_unavailable"},
			"action":      "demote_and_next",
			"target":      "platform",
			"demoteSteps": 1,
		},
		map[string]any{
			"enabled":     true,
			"categories":  []any{"auth_error"},
			"codes":       []any{"auth_failed", "invalid_api_key", "missing_credentials"},
			"statusCodes": []any{401, 403},
			"keywords":    []any{"invalid api key", "api key is invalid", "unauthorized", "forbidden", "permission denied"},
			"action":      "disable_and_next",
			"target":      "platform",
		},
		map[string]any{
			"enabled":    true,
			"categories": []any{"request_error", "unsupported_model", "user_permission", "insufficient_balance"},
			"codes":      []any{"bad_request", "invalid_request", "invalid_parameter", "missing_required", "unsupported_kind", "unsupported_model", "insufficient_balance", "permission_denied"},
			"keywords":   []any{"invalid_parameter", "missing required", "bad request", "insufficient balance"},
			"action":     "stop",
		},
	}
	if !runnerActionRulesLookLikeLegacyDefaults(legacy) {
		t.Fatal("expected old compact default action rules to be detected")
	}
	if runnerActionRulesLookLikeLegacyDefaults(defaultRunnerFailoverPolicy()["actionRules"]) {
		t.Fatal("enhanced defaults should not be treated as legacy")
	}
}

func defaultActionRuleByAction(t *testing.T, rules []any, action string) map[string]any {
	t.Helper()
	for _, raw := range rules {
		rule, ok := raw.(map[string]any)
		if ok && rule["action"] == action {
			return rule
		}
	}
	t.Fatalf("expected action rule %q in %+v", action, rules)
	return nil
}

func actionRuleHasString(rule map[string]any, key string, want string) bool {
	for _, raw := range ruleSlice(rule, key) {
		if raw == want {
			return true
		}
	}
	return false
}

func actionRuleHasInt(rule map[string]any, key string, want int) bool {
	for _, raw := range ruleSlice(rule, key) {
		if raw == want {
			return true
		}
	}
	return false
}

func ruleSlice(rule map[string]any, key string) []any {
	values, _ := rule[key].([]any)
	return values
}