synapse/internal/components/harbor/harbor.go

// Package harbor 灵泊 - 异步队列与暂存层
//
// "泊"为船舶停靠。数据包如远行之舟，
// 遇风浪（网络抖动）时暂泊于此，
// 待风平浪静（目标就绪）再启航。
// 不丢包、不超时，给予 Agent 异步协作的从容。
package harbor

import (
	"context"
	"encoding/json"
	"fmt"
	"log"
	"sync"
	"time"

	"synapse/pkg/types"
)

// =============================================================================
// Message Queue - 消息队列
// =============================================================================

// QueueType 队列类型
type QueueType string

const (
	QueueFIFO     QueueType = "fifo"     // 先进先出
	QueuePriority QueueType = "priority" // 优先级队列
	QueueLIFO     QueueType = "lifo"     // 后进先出
)

// QueuedMessage 队列中的消息
type QueuedMessage struct {
	ID          string
	Message     *types.UnifiedSynapseMessage
	Priority    int
	EnqueueTime time.Time
	RetryCount  int
	MaxRetries  int
	NextRetry   time.Time
	Status      QueuedStatus
	Target      string // 目标 endpoint
}

// QueuedStatus 队列消息状态
type QueuedStatus string

const (
	QueuedPending    QueuedStatus = "pending"
	QueuedProcessing QueuedStatus = "processing"
	QueuedDelivered  QueuedStatus = "delivered"
	QueuedFailed     QueuedStatus = "failed"
	QueuedDead       QueuedStatus = "dead"
)

// MessageQueue 消息队列
type MessageQueue struct {
	mu       sync.RWMutex
	messages []*QueuedMessage
	config   *QueueConfig
}

// QueueConfig 队列配置
type QueueConfig struct {
	MaxSize           int           // 最大队列大小
	DefaultMaxRetries int           // 默认最大重试次数
	DefaultRetryDelay time.Duration // 默认重试延迟
}

// DefaultQueueConfig 返回默认队列配置
func DefaultQueueConfig() *QueueConfig {
	return &QueueConfig{
		MaxSize:           10000,
		DefaultMaxRetries: 5,
		DefaultRetryDelay: 30 * time.Second,
	}
}

// NewMessageQueue 创建消息队列
func NewMessageQueue(config *QueueConfig) *MessageQueue {
	if config == nil {
		config = DefaultQueueConfig()
	}
	return &MessageQueue{
		messages: make([]*QueuedMessage, 0),
		config:   config,
	}
}

// Enqueue 入队
func (mq *MessageQueue) Enqueue(msg *types.UnifiedSynapseMessage, target string, priority int) (*QueuedMessage, error) {
	mq.mu.Lock()
	defer mq.mu.Unlock()

	if len(mq.messages) >= mq.config.MaxSize {
		return nil, fmt.Errorf("queue is full")
	}

	qm := &QueuedMessage{
		ID:          fmt.Sprintf("q-%d", time.Now().UnixNano()),
		Message:     msg,
		Priority:    priority,
		EnqueueTime: time.Now(),
		RetryCount:  0,
		MaxRetries:  mq.config.DefaultMaxRetries,
		NextRetry:   time.Now(),
		Status:      QueuedPending,
		Target:      target,
	}

	// 按优先级插入
	inserted := false
	for i, m := range mq.messages {
		if priority > m.Priority {
			mq.messages = append(mq.messages[:i], append([]*QueuedMessage{qm}, mq.messages[i:]...)...)
			inserted = true
			break
		}
	}
	if !inserted {
		mq.messages = append(mq.messages, qm)
	}

	log.Printf("[灵泊] 消息入队: %s (目标: %s, 优先级: %d)", qm.ID, target, priority)
	return qm, nil
}

// Dequeue 出队
func (mq *MessageQueue) Dequeue() (*QueuedMessage, error) {
	mq.mu.Lock()
	defer mq.mu.Unlock()

	now := time.Now()
	for i, qm := range mq.messages {
		if qm.Status == QueuedPending && now.After(qm.NextRetry) {
			qm.Status = QueuedProcessing
			// 移到待处理
			mq.messages = append(mq.messages[:i], mq.messages[i+1:]...)
			return qm, nil
		}
	}

	return nil, fmt.Errorf("no messages available")
}

// MarkDelivered 标记为已送达
func (mq *MessageQueue) MarkDelivered(msgID string) error {
	mq.mu.Lock()
	defer mq.mu.Unlock()

	for _, qm := range mq.messages {
		if qm.ID == msgID {
			qm.Status = QueuedDelivered
			log.Printf("[灵泊] 消息送达: %s", msgID)
			return nil
		}
	}
	return fmt.Errorf("message %s not found", msgID)
}

// MarkFailed 标记为失败（会重试）
func (mq *MessageQueue) MarkFailed(msgID string, err error) error {
	mq.mu.Lock()
	defer mq.mu.Unlock()

	for _, qm := range mq.messages {
		if qm.ID == msgID {
			qm.RetryCount++
			if qm.RetryCount >= qm.MaxRetries {
				qm.Status = QueuedDead
				log.Printf("[灵泊] 消息失败(已死): %s (重试: %d次)", msgID, qm.RetryCount)
			} else {
				qm.Status = QueuedPending
				// 指数退避
				delay := mq.config.DefaultRetryDelay * time.Duration(1<<uint(qm.RetryCount-1))
				qm.NextRetry = time.Now().Add(delay)
				log.Printf("[灵泊] 消息失败(将重试): %s (重试: %d/%d, 下次: %v)",
					msgID, qm.RetryCount, qm.MaxRetries, delay)
			}
			return nil
		}
	}
	return fmt.Errorf("message %s not found", msgID)
}

// Size 队列大小
func (mq *MessageQueue) Size() int {
	mq.mu.RLock()
	defer mq.mu.RUnlock()
	return len(mq.messages)
}

// =============================================================================
// Message Buffer - 消息缓冲
// =============================================================================

// MessageBuffer 消息缓冲（暂存离线消息）
type MessageBuffer struct {
	mu      sync.RWMutex
	buffers map[string][]*types.UnifiedSynapseMessage // agent_id -> messages
	config  *BufferConfig
}

// BufferConfig 缓冲配置
type BufferConfig struct {
	MaxPerAgent     int           // 每个 Agent 最大缓冲数
	MaxAge          time.Duration // 消息最大存活时间
	CleanupInterval time.Duration // 清理间隔
}

// DefaultBufferConfig 返回默认缓冲配置
func DefaultBufferConfig() *BufferConfig {
	return &BufferConfig{
		MaxPerAgent:     1000,
		MaxAge:          24 * time.Hour,
		CleanupInterval: 1 * time.Hour,
	}
}

// NewMessageBuffer 创建消息缓冲
func NewMessageBuffer(config *BufferConfig) *MessageBuffer {
	if config == nil {
		config = DefaultBufferConfig()
	}
	mb := &MessageBuffer{
		buffers: make(map[string][]*types.UnifiedSynapseMessage),
		config:  config,
	}

	// 启动清理协程
	go mb.cleanupLoop()

	return mb
}

// BufferMessage 缓冲消息
func (mb *MessageBuffer) BufferMessage(agentID string, msg *types.UnifiedSynapseMessage) error {
	mb.mu.Lock()
	defer mb.mu.Unlock()

	buf, ok := mb.buffers[agentID]
	if !ok {
		buf = make([]*types.UnifiedSynapseMessage, 0)
	}

	if len(buf) >= mb.config.MaxPerAgent {
		// 丢弃最旧的
		buf = buf[1:]
	}

	mb.buffers[agentID] = append(buf, msg)

	log.Printf("[灵泊] 缓冲消息: %s -> %s (缓冲区大小: %d)",
		msg.Sender.AgentID, agentID, len(mb.buffers[agentID]))

	return nil
}

// FlushBuffer 清空缓冲并返回所有消息
func (mb *MessageBuffer) FlushBuffer(agentID string) []*types.UnifiedSynapseMessage {
	mb.mu.Lock()
	defer mb.mu.Unlock()

	buf, ok := mb.buffers[agentID]
	if !ok {
		return []*types.UnifiedSynapseMessage{}
	}

	delete(mb.buffers, agentID)

	log.Printf("[灵泊] 清空缓冲: %s (消息数: %d)", agentID, len(buf))
	return buf
}

// GetBufferSize 获取缓冲大小
func (mb *MessageBuffer) GetBufferSize(agentID string) int {
	mb.mu.RLock()
	defer mb.mu.RUnlock()
	return len(mb.buffers[agentID])
}

// cleanupLoop 清理过期消息
func (mb *MessageBuffer) cleanupLoop() {
	ticker := time.NewTicker(mb.config.CleanupInterval)
	defer ticker.Stop()

	for range ticker.C {
		mb.cleanup()
	}
}

func (mb *MessageBuffer) cleanup() {
	mb.mu.Lock()
	defer mb.mu.Unlock()

	now := time.Now()
	cleaned := 0

	for agentID, buf := range mb.buffers {
		var newBuf []*types.UnifiedSynapseMessage
		for _, msg := range buf {
			msgTime, _ := time.Parse(time.RFC3339, string(msg.Timestamp))
			if now.Sub(msgTime) < mb.config.MaxAge {
				newBuf = append(newBuf, msg)
			} else {
				cleaned++
			}
		}
		mb.buffers[agentID] = newBuf
	}

	if cleaned > 0 {
		log.Printf("[灵泊] 清理过期消息: %d 条", cleaned)
	}
}

// =============================================================================
// Retry Manager - 重试管理器
// =============================================================================

// RetryManager 重试管理器
type RetryManager struct {
	mu      sync.RWMutex
	retries map[string]*RetryEntry
	config  *RetryConfig
}

// RetryEntry 重试条目
type RetryEntry struct {
	ID          string
	Message     *types.UnifiedSynapseMessage
	Target      string
	RetryCount  int
	MaxRetries  int
	NextRetry   time.Time
	LastError   string
	BackoffBase time.Duration
}

// RetryConfig 重试配置
type RetryConfig struct {
	MaxRetries   int
	BaseDelay    time.Duration
	MaxDelay     time.Duration
	JitterFactor float64
}

// DefaultRetryConfig 返回默认重试配置
func DefaultRetryConfig() *RetryConfig {
	return &RetryConfig{
		MaxRetries:   5,
		BaseDelay:    1 * time.Second,
		MaxDelay:     5 * time.Minute,
		JitterFactor: 0.1,
	}
}

// NewRetryManager 创建重试管理器
func NewRetryManager(config *RetryConfig) *RetryManager {
	if config == nil {
		config = DefaultRetryConfig()
	}
	return &RetryManager{
		retries: make(map[string]*RetryEntry),
		config:  config,
	}
}

// AddRetry 添加重试条目
func (rm *RetryManager) AddRetry(msg *types.UnifiedSynapseMessage, target string) {
	rm.mu.Lock()
	defer rm.mu.Unlock()

	id := fmt.Sprintf("retry-%d", time.Now().UnixNano())
	entry := &RetryEntry{
		ID:          id,
		Message:     msg,
		Target:      target,
		RetryCount:  0,
		MaxRetries:  rm.config.MaxRetries,
		NextRetry:   time.Now().Add(rm.config.BaseDelay),
		BackoffBase: rm.config.BaseDelay,
	}

	rm.retries[id] = entry

	log.Printf("[灵泊] 添加重试: %s -> %s (最多 %d 次)", id, target, rm.config.MaxRetries)
}

// GetPendingRetries 获取待重试的条目
func (rm *RetryManager) GetPendingRetries() []*RetryEntry {
	rm.mu.RLock()
	defer rm.mu.RUnlock()

	now := time.Now()
	var pending []*RetryEntry
	for _, entry := range rm.retries {
		if now.After(entry.NextRetry) && entry.RetryCount < entry.MaxRetries {
			pending = append(pending, entry)
		}
	}
	return pending
}

// MarkSuccess 标记成功
func (rm *RetryManager) MarkSuccess(retryID string) {
	rm.mu.Lock()
	defer rm.mu.Unlock()
	delete(rm.retries, retryID)
	log.Printf("[灵泊] 重试成功: %s", retryID)
}

// MarkFailure 标记失败
func (rm *RetryManager) MarkFailure(retryID string, err error) bool {
	rm.mu.Lock()
	defer rm.mu.Unlock()

	entry, ok := rm.retries[retryID]
	if !ok {
		return false
	}

	entry.RetryCount++
	entry.LastError = err.Error()

	if entry.RetryCount >= entry.MaxRetries {
		delete(rm.retries, retryID)
		log.Printf("[灵泊] 重试失败(已达上限): %s (重试 %d 次)", retryID, entry.RetryCount)
		return false
	}

	// 指数退避 + 抖动
	delay := entry.BackoffBase * time.Duration(1<<uint(entry.RetryCount-1))
	if delay > rm.config.MaxDelay {
		delay = rm.config.MaxDelay
	}
	entry.NextRetry = time.Now().Add(delay)

	log.Printf("[灵泊] 重试失败: %s (%d/%d, 下次: %v)",
		retryID, entry.RetryCount, entry.MaxRetries, delay)

	return true
}

// Stats 获取统计信息
func (rm *RetryManager) Stats() (pending, failed int) {
	rm.mu.RLock()
	defer rm.mu.RUnlock()
	for _, entry := range rm.retries {
		if entry.RetryCount == 0 {
			pending++
		}
	}
	return pending, 0
}

// =============================================================================
// Harbor - 灵泊整体入口
// =============================================================================

// Harbor 灵泊 - 异步队列与暂存
type Harbor struct {
	Queue  *MessageQueue
	Buffer *MessageBuffer
	Retry  *RetryManager
}

// NewHarbor 创建灵泊
func NewHarbor() *Harbor {
	return &Harbor{
		Queue:  NewMessageQueue(nil),
		Buffer: NewMessageBuffer(nil),
		Retry:  NewRetryManager(nil),
	}
}

// SendWithRetry 发送消息并重试
func (h *Harbor) SendWithRetry(ctx context.Context, msg *types.UnifiedSynapseMessage, target string, sendFunc func(ctx context.Context, msg *types.UnifiedSynapseMessage, target string) error) error {
	// 先入队
	qm, err := h.Queue.Enqueue(msg, target, 5)
	if err != nil {
		return err
	}

	// 尝试发送
	err = sendFunc(ctx, msg, target)
	if err == nil {
		h.Queue.MarkDelivered(qm.ID)
		return nil
	}

	// 失败，加入重试
	h.Retry.AddRetry(msg, target)
	h.Queue.MarkFailed(qm.ID, err)

	return err
}

// BufferForOffline 为离线 Agent 缓冲消息
func (h *Harbor) BufferForOffline(agentID string, msg *types.UnifiedSynapseMessage) {
	h.Buffer.BufferMessage(agentID, msg)
}

// FlushAndSend 清空缓冲并发送
func (h *Harbor) FlushAndSend(agentID string, sendFunc func(msg *types.UnifiedSynapseMessage) error) int {
	msgs := h.Buffer.FlushBuffer(agentID)
	sent := 0
	for _, msg := range msgs {
		if err := sendFunc(msg); err == nil {
			sent++
		}
	}
	return sent
}

// ToJSON 导出为 JSON
func (qm *QueuedMessage) ToJSON() string {
	data, _ := json.MarshalIndent(qm, "", "  ")
	return string(data)
}