raft/server.go

package raft

import (
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"net"
	"net/http"
	"sort"
	"strings"
	"sync"
	"time"

	"igit.com/xbase/raft/db"
)

// KVServer wraps Raft to provide a distributed key-value store
type KVServer struct {
	Raft       *Raft
	DB         *db.Engine
	CLI        *CLI
	AuthManager *AuthManager
	Watcher    *WebHookWatcher
	httpServer *http.Server
	stopCh     chan struct{}
	wg         sync.WaitGroup
	stopOnce   sync.Once
	// leavingNodes tracks nodes that are currently being removed
	// to prevent auto-rejoin/discovery logic from interfering
	leavingNodes sync.Map
	
	// Pending requests for synchronous operations (Read-Your-Writes)
	pendingRequests map[uint64]chan error
	pendingMu       sync.Mutex
}

// NewKVServer creates a new KV server
func NewKVServer(config *Config) (*KVServer, error) {
	// Initialize DB Engine
	// Use a subdirectory for DB to avoid conflict with Raft logs if they share DataDir
	dbPath := config.DataDir + "/kv_engine"
	engine, err := db.NewEngine(dbPath)
	if err != nil {
		return nil, fmt.Errorf("failed to create db engine: %w", err)
	}

	// Initialize LastAppliedIndex from DB to prevent re-applying entries
	config.LastAppliedIndex = engine.GetLastAppliedIndex()

	// Create stop channel early for use in callbacks
	stopCh := make(chan struct{})

	// Configure snapshot provider
	config.SnapshotProvider = func(minIncludeIndex uint64) ([]byte, error) {
		// Wait for DB to catch up to the requested index
		// This is critical for data integrity during compaction
		for engine.GetLastAppliedIndex() < minIncludeIndex {
			select {
			case <-stopCh:
				return nil, fmt.Errorf("server stopping")
			default:
				time.Sleep(10 * time.Millisecond)
			}
		}

		// Force sync to disk to ensure data durability before compaction
		// This prevents data loss if Raft logs are compacted but DB data is only in OS cache
		if err := engine.Sync(); err != nil {
			return nil, fmt.Errorf("failed to sync engine before snapshot: %w", err)
		}

		return engine.Snapshot()
	}

	// Configure get handler for remote reads
	config.GetHandler = func(key string) (string, bool) {
		return engine.Get(key)
	}

	applyCh := make(chan ApplyMsg, 1000) // Increase buffer for async processing
	transport := NewTCPTransport(config.ListenAddr, 10, config.Logger)

	// Initialize WebHookWatcher
	// 5 workers, 3 retries
	watcher := NewWebHookWatcher(5, 3, config.Logger)

	r, err := NewRaft(config, transport, applyCh)
	if err != nil {
		engine.Close()
		return nil, err
	}

	s := &KVServer{
		Raft:            r,
		DB:              engine,
		CLI:             nil,
		AuthManager:     nil, // initialized below
		Watcher:         watcher,
		stopCh:          stopCh,
		pendingRequests: make(map[uint64]chan error),
	}

	// Initialize AuthManager
	s.AuthManager = NewAuthManager(s)
	// Load Auth Data from DB (if any)
	if err := s.AuthManager.LoadFromDB(); err != nil {
		s.Raft.config.Logger.Warn("Failed to load auth data from DB: %v", err)
	}

	// Initialize CLI
	s.CLI = NewCLI(s)

	// Start applying entries
	go s.runApplyLoop(applyCh)

	// Start background maintenance loop
	s.wg.Add(1)
	go s.maintenanceLoop()

	return s, nil
}

func (s *KVServer) Start() error {
	// Start CLI if enabled
	if s.Raft.config.EnableCLI {
		go s.CLI.Start()
	}

	// Start HTTP Server if configured
	if s.Raft.config.HTTPAddr != "" {
		if err := s.startHTTPServer(s.Raft.config.HTTPAddr); err != nil {
			s.Raft.config.Logger.Warn("Failed to start HTTP server: %v", err)
		}
	}
	
	// Start TCP Server (Hardcoded offset +10 for demo or config?)
	// User didn't add TCPPort to config, so let's derive it or just fix it.
	// For demo, if ListenAddr is 127.0.0.1:9001, we try 9011.
	host, port, _ := net.SplitHostPort(s.Raft.config.ListenAddr)
	if port == "9001" {
		tcpAddr := fmt.Sprintf("%s:%s", host, "9011")
		if err := s.StartTCPServer(tcpAddr); err != nil {
			s.Raft.config.Logger.Warn("Failed to start TCP server: %v", err)
		}
	} else if port == "9002" {
		tcpAddr := fmt.Sprintf("%s:%s", host, "9012")
		s.StartTCPServer(tcpAddr)
	} else {
		// Fallback or ignore for other nodes in demo
	}

	return s.Raft.Start()
}

func (s *KVServer) Stop() error {
	var err error
	s.stopOnce.Do(func() {
		// Stop maintenance loop
		if s.stopCh != nil {
			close(s.stopCh)
			s.wg.Wait()
		}
		// Stop Watcher
		if s.Watcher != nil {
			s.Watcher.Stop()
		}
		// Stop HTTP Server
		if s.httpServer != nil {
			s.httpServer.Close()
		}
		// Stop Raft first
		if errRaft := s.Raft.Stop(); errRaft != nil {
			err = errRaft
		}
		// Close DB
		if s.DB != nil {
			if errDB := s.DB.Close(); errDB != nil {
				// Combine errors if both fail
				if err != nil {
					err = fmt.Errorf("raft stop error: %v, db close error: %v", err, errDB)
				} else {
					err = errDB
				}
			}
		}
	})
	return err
}

func (s *KVServer) runApplyLoop(applyCh chan ApplyMsg) {
	for msg := range applyCh {
		if msg.CommandValid {
			// Optimization: Skip if already applied
			// We check this here to avoid unmarshalling and locking DB for known duplicates
			if msg.CommandIndex <= s.DB.GetLastAppliedIndex() {
				// Even if duplicate, we might have someone waiting for it if they just proposed it
				// but this is unlikely for duplicates.
				// However, notify waiters just in case.
				s.notifyPending(msg.CommandIndex, nil)
				continue
			}

			var cmd KVCommand
			if err := json.Unmarshal(msg.Command, &cmd); err != nil {
				s.Raft.config.Logger.Error("Failed to unmarshal command: %v", err)
				s.notifyPending(msg.CommandIndex, err)
				continue
			}

			var err error
			switch cmd.Type {
			case KVSet:
				// Update Auth Cache for system keys
				// This includes AuthLockPrefix now
				if strings.HasPrefix(cmd.Key, SystemKeyPrefix) {
					s.AuthManager.UpdateCache(cmd.Key, cmd.Value, false)
				}
				
				err = s.DB.Set(cmd.Key, cmd.Value, msg.CommandIndex)
				if err == nil {
					s.Watcher.Notify(cmd.Key, cmd.Value, KVSet)
				}
			case KVDel:
				// Update Auth Cache for system keys
				if strings.HasPrefix(cmd.Key, SystemKeyPrefix) {
					s.AuthManager.UpdateCache(cmd.Key, "", true)
				}

				err = s.DB.Delete(cmd.Key, msg.CommandIndex)
				if err == nil {
					s.Watcher.Notify(cmd.Key, "", KVDel)
				}
			default:
				s.Raft.config.Logger.Error("Unknown command type: %d", cmd.Type)
			}

			if err != nil {
				s.Raft.config.Logger.Error("DB Apply failed: %v", err)
			}
			
			// Notify anyone waiting for this index
			s.notifyPending(msg.CommandIndex, err)

		} else if msg.SnapshotValid {
			if err := s.DB.Restore(msg.Snapshot); err != nil {
				s.Raft.config.Logger.Error("DB Restore failed: %v", err)
			}
		}
	}
}

// notifyPending notifies any waiting requests for the given index
func (s *KVServer) notifyPending(index uint64, err error) {
	s.pendingMu.Lock()
	defer s.pendingMu.Unlock()
	
	if ch, ok := s.pendingRequests[index]; ok {
		// Non-blocking send in case listener is gone (buffered channel recommended)
		select {
		case ch <- err:
		default:
		}
		close(ch)
		delete(s.pendingRequests, index)
	}
}

// SetAuthenticated sets a key-value pair with permission check
func (s *KVServer) SetAuthenticated(key, value, token string) error {
	if err := s.AuthManager.CheckPermission(token, key, ActionWrite, value); err != nil {
		return err
	}
	// Use SetSync for consistency in CLI/API
	return s.SetSync(key, value)
}

// DelAuthenticated deletes a key with permission check
func (s *KVServer) DelAuthenticated(key, token string) error {
	if err := s.AuthManager.CheckPermission(token, key, ActionWrite, ""); err != nil {
		return err
	}
	// Use SetSync (via Del which should be updated or create DelSync)
	// For simplicity, we implement DelSync logic here or update Del to be sync?
	// Let's implement DelSync
	return s.DelSync(key)
}

// Set sets a key-value pair (Async - eventually consistent)
func (s *KVServer) Set(key, value string) error {
	cmd := KVCommand{
		Type:  KVSet,
		Key:   key,
		Value: value,
	}
	data, err := json.Marshal(cmd)
	if err != nil {
		return err
	}

	_, _, err = s.Raft.ProposeWithForward(data)
	return err
}

// SetSync sets a key-value pair and waits for it to be applied (Read-Your-Writes)
func (s *KVServer) SetSync(key, value string) error {
	cmd := KVCommand{
		Type:  KVSet,
		Key:   key,
		Value: value,
	}
	data, err := json.Marshal(cmd)
	if err != nil {
		return err
	}

	index, _, err := s.Raft.ProposeWithForward(data)
	if err != nil {
		return err
	}

	// Wait for application
	ch := make(chan error, 1)
	s.pendingMu.Lock()
	s.pendingRequests[index] = ch
	s.pendingMu.Unlock()

	select {
	case applyErr := <-ch:
		return applyErr
	case <-time.After(5 * time.Second):
		s.pendingMu.Lock()
		delete(s.pendingRequests, index)
		s.pendingMu.Unlock()
		return fmt.Errorf("timeout waiting for apply")
	}
}

// Del deletes a key (Async)
func (s *KVServer) Del(key string) error {
	cmd := KVCommand{
		Type: KVDel,
		Key:  key,
	}
	data, err := json.Marshal(cmd)
	if err != nil {
		return err
	}

	_, _, err = s.Raft.ProposeWithForward(data)
	return err
}

// DelSync deletes a key and waits for it to be applied
func (s *KVServer) DelSync(key string) error {
	cmd := KVCommand{
		Type: KVDel,
		Key:  key,
	}
	data, err := json.Marshal(cmd)
	if err != nil {
		return err
	}

	index, _, err := s.Raft.ProposeWithForward(data)
	if err != nil {
		return err
	}

	// Wait for application
	ch := make(chan error, 1)
	s.pendingMu.Lock()
	s.pendingRequests[index] = ch
	s.pendingMu.Unlock()

	select {
	case applyErr := <-ch:
		return applyErr
	case <-time.After(5 * time.Second):
		s.pendingMu.Lock()
		delete(s.pendingRequests, index)
		s.pendingMu.Unlock()
		return fmt.Errorf("timeout waiting for apply")
	}
}

// GetAuthenticated gets a value with permission check (local read)
func (s *KVServer) GetAuthenticated(key, token string) (string, bool, error) {
	if err := s.AuthManager.CheckPermission(token, key, ActionRead, ""); err != nil {
		return "", false, err
	}
	val, ok := s.Get(key)
	return val, ok, nil
}

// Get gets a value (local read, can be stale)
// For linearizable reads, use GetLinear instead
func (s *KVServer) Get(key string) (string, bool) {
	return s.DB.Get(key)
}

// GetLinearAuthenticated gets a value with linearizable consistency and permission check
func (s *KVServer) GetLinearAuthenticated(key, token string) (string, bool, error) {
	if err := s.AuthManager.CheckPermission(token, key, ActionRead, ""); err != nil {
		return "", false, err
	}
	return s.GetLinear(key)
}

// Logout invalidates the current session
func (s *KVServer) Logout(token string) error {
	return s.AuthManager.Logout(token)
}

// GetSessionInfo returns the session details
func (s *KVServer) GetSessionInfo(token string) (*Session, error) {
	return s.AuthManager.GetSession(token)
}

// IsRoot checks if the token belongs to the root user
func (s *KVServer) IsRoot(token string) bool {
	sess, err := s.AuthManager.GetSession(token)
	if err != nil {
		return false
	}
	return sess.Username == "root"
}

// CreateUser creates a new user (Root only)
func (s *KVServer) CreateUser(username, password string, roles []string, token string) error {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return fmt.Errorf("permission denied: root access required")
	}
	// Use RegisterUserSync
	return s.AuthManager.RegisterUser(username, password, roles)
}

// DeleteUser deletes a user (Root only)
func (s *KVServer) DeleteUser(username string, token string) error {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return fmt.Errorf("permission denied: root access required")
	}
	// Check if user exists
	if _, err := s.AuthManager.GetUser(username); err != nil {
		return err
	}
	if username == "root" {
		return fmt.Errorf("cannot delete root user")
	}
	// Use DelSync
	return s.DelSync(AuthUserPrefix + username)
}

// UpdateUser updates generic user fields (Root only)
func (s *KVServer) UpdateUser(user User, token string) error {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return fmt.Errorf("permission denied: root access required")
	}
	// Check if user exists
	if _, err := s.AuthManager.GetUser(user.Username); err != nil {
		return err
	}
	return s.AuthManager.UpdateUser(user)
}

// ChangeUserPassword changes a user's password (Root or Self)
func (s *KVServer) ChangeUserPassword(username, newPassword string, token string) error {
	if s.AuthManager.IsEnabled() {
		session, err := s.AuthManager.GetSession(token)
		if err != nil {
			return err
		}
		if session.Username != "root" && session.Username != username {
			return fmt.Errorf("permission denied")
		}
	}
	// Use ChangePassword (which should use SetSync)
	return s.AuthManager.ChangePassword(username, newPassword)
}

// Role Management Helpers

// CreateRole creates a new role (Root only)
func (s *KVServer) CreateRole(name string, token string) error {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return fmt.Errorf("permission denied: root access required")
	}
	return s.AuthManager.CreateRole(name)
}

// DeleteRole deletes a role (Root only)
func (s *KVServer) DeleteRole(name string, token string) error {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return fmt.Errorf("permission denied: root access required")
	}
	return s.AuthManager.DeleteRole(name)
}

// UpdateRole updates a role (Root only)
func (s *KVServer) UpdateRole(role Role, token string) error {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return fmt.Errorf("permission denied: root access required")
	}
	return s.AuthManager.UpdateRole(role)
}

// ListUsers lists all users (Root only)
func (s *KVServer) ListUsers(token string) ([]*User, error) {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return nil, fmt.Errorf("permission denied: root access required")
	}
	return s.AuthManager.ListUsers(), nil
}

// ListRoles lists all roles (Root only)
func (s *KVServer) ListRoles(token string) ([]*Role, error) {
	if s.AuthManager.IsEnabled() && !s.IsRoot(token) {
		return nil, fmt.Errorf("permission denied: root access required")
	}
	return s.AuthManager.ListRoles(), nil
}

// SearchAuthenticated searches keys with permission checks
func (s *KVServer) SearchAuthenticated(pattern string, limit, offset int, token string) ([]db.QueryResult, error) {
	// Optimization: If user has full access (*), delegate to DB with limit/offset
	if s.AuthManager.HasFullAccess(token) {
		sql := fmt.Sprintf("key like \"%s\" LIMIT %d OFFSET %d", pattern, limit, offset)
		return s.DB.Query(sql)
	}

	// Slow path: Fetch all potential matches and filter
	// We construct a SQL query that retrieves CANDIDATES.
	// We cannot safely apply LIMIT/OFFSET in SQL because we might filter out some results.
	// So we must fetch ALL matches.
	// WARNING: This can be slow and memory intensive.
	
	// If pattern is wildcard, we might fetch everything.
	sql := fmt.Sprintf("key like \"%s\"", pattern)
	results, err := s.DB.Query(sql)
	if err != nil {
		return nil, err
	}

	filtered := make([]db.QueryResult, 0, len(results))
	
	// Apply Filtering
	for _, r := range results {
		if err := s.AuthManager.CheckPermission(token, r.Key, ActionRead, ""); err == nil {
			filtered = append(filtered, r)
		}
	}

	// Apply Pagination on Filtered Results
	if offset > len(filtered) {
		return []db.QueryResult{}, nil
	}
	
	start := offset
	end := offset + limit
	if end > len(filtered) {
		end = len(filtered)
	}
	
	return filtered[start:end], nil
}

// CountAuthenticated counts keys with permission checks
func (s *KVServer) CountAuthenticated(pattern string, token string) (int, error) {
	// Optimization: If user has full access
	if s.AuthManager.HasFullAccess(token) {
		sql := ""
		if pattern == "*" {
			sql = "*"
		} else {
			sql = fmt.Sprintf("key like \"%s\"", pattern)
		}
		return s.DB.Count(sql)
	}

	// Slow path: Iterate and check
	// We use DB.Query to get keys (Query returns values too, which is wasteful but DB engine doesn't expose keys-only Query yet)
	// Actually, we can access s.DB.Index.WalkPrefix if we want to be faster and avoid value read, 
	// but s.DB.Index is inside 'db' package. We can access it if it's exported.
	// 'db' package exports 'Engine' and 'FlatIndex'.
	// So s.DB.Index IS accessible.
	
	count := 0
	
	// Determine prefix from pattern
	prefix := ""
	if strings.HasSuffix(pattern, "*") {
		prefix = strings.TrimSuffix(pattern, "*")
	} else if pattern == "*" {
		prefix = ""
	} else {
		// Exact match check
		if err := s.AuthManager.CheckPermission(token, pattern, ActionRead, ""); err == nil {
			// Check if exists
			if _, ok := s.DB.Get(pattern); ok {
				return 1, nil
			}
			return 0, nil
		}
		return 0, nil // No perm or not found
	}

	// Walk
	// Note: WalkPrefix locks the DB Index (Read Lock).
	// Calling CheckPermission inside might involve some logic but it is memory-only and usually fast.
	// However, if CheckPermission takes time, we hold DB lock.
	s.DB.Index.WalkPrefix(prefix, func(key string, entry db.IndexEntry) bool {
		// Check pattern match first (WalkPrefix is just prefix, pattern might be more complex like "user.*.name")
		if !db.WildcardMatch(key, pattern) {
			return true
		}
		
		if err := s.AuthManager.CheckPermission(token, key, ActionRead, ""); err == nil {
			count++
		}
		return true
	})

	return count, nil
}

// GetLinear gets a value with linearizable consistency
// This ensures the read sees all writes committed before the read started
func (s *KVServer) GetLinear(key string) (string, bool, error) {
	// First, ensure we have up-to-date data via ReadIndex
	_, err := s.Raft.ReadIndex()
	if err != nil {
		// If we're not leader, try forwarding
		if errors.Is(err, ErrNotLeader) {
			return s.forwardGet(key)
		}
		return "", false, err
	}

	val, ok := s.DB.Get(key)
	return val, ok, nil
}

// forwardGet forwards a get request to the leader
func (s *KVServer) forwardGet(key string) (string, bool, error) {
	return s.Raft.ForwardGet(key)
}

// Join joins an existing cluster
func (s *KVServer) Join(nodeID, addr string) error {
	return s.Raft.AddNodeWithForward(nodeID, addr)
}

// Leave leaves the cluster
func (s *KVServer) Leave(nodeID string) error {
	// Mark node as leaving to prevent auto-rejoin
	s.leavingNodes.Store(nodeID, time.Now())
	
	// Auto-expire the leaving flag after a while
	go func() {
		time.Sleep(30 * time.Second)
		s.leavingNodes.Delete(nodeID)
	}()

	// Remove from RaftNode discovery key first to prevent auto-rejoin
	if err := s.removeNodeFromDiscovery(nodeID); err != nil {
		s.Raft.config.Logger.Warn("Failed to remove node from discovery key: %v", err)
		// Continue anyway, as the main goal is to leave the cluster
	}
	return s.Raft.RemoveNodeWithForward(nodeID)
}

// removeNodeFromDiscovery removes a node from the RaftNode key to prevent auto-rejoin
func (s *KVServer) removeNodeFromDiscovery(targetID string) error {
	val, ok := s.Get("RaftNode")
	if !ok || val == "" {
		return nil
	}

	parts := strings.Split(val, ";")
	var newParts []string
	changed := false

	for _, part := range parts {
		if part == "" {
			continue
		}
		kv := strings.SplitN(part, "=", 2)
		if len(kv) == 2 {
			if kv[0] == targetID {
				changed = true
				continue // Skip this node
			}
			newParts = append(newParts, part)
		}
	}

	if changed {
		newVal := strings.Join(newParts, ";")
		return s.Set("RaftNode", newVal)
	}
	return nil
}

// WaitForLeader waits until a leader is elected
func (s *KVServer) WaitForLeader(timeout time.Duration) error {
	deadline := time.Now().Add(timeout)
	for time.Now().Before(deadline) {
		leader := s.Raft.GetLeaderID()
		if leader != "" {
			return nil
		}
		time.Sleep(100 * time.Millisecond)
	}
	return fmt.Errorf("timeout waiting for leader")
}

// HealthCheck returns the health status of this server
func (s *KVServer) HealthCheck() HealthStatus {
	return s.Raft.HealthCheck()
}

// GetStats returns runtime statistics
func (s *KVServer) GetStats() Stats {
	return s.Raft.GetStats()
}

// GetMetrics returns runtime metrics
func (s *KVServer) GetMetrics() Metrics {
	return s.Raft.GetMetrics()
}

// TransferLeadership transfers leadership to the specified node
func (s *KVServer) TransferLeadership(targetID string) error {
	return s.Raft.TransferLeadership(targetID)
}

// GetClusterNodes returns current cluster membership
func (s *KVServer) GetClusterNodes() map[string]string {
	return s.Raft.GetClusterNodes()
}

// IsLeader returns true if this node is the leader
func (s *KVServer) IsLeader() bool {
	_, isLeader := s.Raft.GetState()
	return isLeader
}

// GetLeaderID returns the current leader ID
func (s *KVServer) GetLeaderID() string {
	return s.Raft.GetLeaderID()
}

// GetLogSize returns the raft log size
func (s *KVServer) GetLogSize() int64 {
	return s.Raft.log.GetLogSize()
}

// GetDBSize returns the db size
func (s *KVServer) GetDBSize() int64 {
	return s.DB.GetDBSize()
}

// WatchURL registers a webhook url for a key
func (s *KVServer) WatchURL(key, url string) {
	s.Watcher.Subscribe(key, url)
}

// UnwatchURL removes a webhook url for a key
func (s *KVServer) UnwatchURL(key, url string) {
	s.Watcher.Unsubscribe(key, url)
}

// WatchAll registers a watcher for all keys
func (s *KVServer) WatchAll(handler WatchHandler) {
	// s.FSM.WatchAll(handler)
	// TODO: Implement Watcher for DB
}

// Watch registers a watcher for a key
func (s *KVServer) Watch(key string, handler WatchHandler) {
	// s.FSM.Watch(key, handler)
	// TODO: Implement Watcher for DB
}

// Unwatch removes watchers for a key
func (s *KVServer) Unwatch(key string) {
	// s.FSM.Unwatch(key)
	// TODO: Implement Watcher for DB
}

func (s *KVServer) maintenanceLoop() {
	defer s.wg.Done()
	// Check every 1 second for faster reaction
	ticker := time.NewTicker(1 * time.Second)
	defer ticker.Stop()

	for {
		select {
		case <-s.stopCh:
			return
		case <-ticker.C:
			s.updateNodeInfo()
			s.checkConnections()
		}
	}
}

func (s *KVServer) updateNodeInfo() {
	// 1. Ensure "CreateNode/<NodeID>" is set to self address
	// We do this via Propose (Set) so it's replicated
	myID := s.Raft.config.NodeID
	myAddr := s.Raft.config.ListenAddr
	key := fmt.Sprintf("CreateNode/%s", myID)

	// Check if we need to update (avoid spamming logs/proposals)
	val, exists := s.Get(key)
	if !exists || val != myAddr {
		// Run in goroutine to avoid blocking
		go func() {
			if err := s.Set(key, myAddr); err != nil {
				s.Raft.config.Logger.Debug("Failed to update node info: %v", err)
			}
		}()
	}

	// 2. Only leader updates RaftNode aggregation
	if s.IsLeader() {
		// Read current RaftNode to preserve history
		currentVal, _ := s.Get("RaftNode")
		
		knownNodes := make(map[string]string)
		if currentVal != "" {
			parts := strings.Split(currentVal, ";")
			for _, part := range parts {
				if part == "" { continue }
				kv := strings.SplitN(part, "=", 2)
				if len(kv) == 2 {
					knownNodes[kv[0]] = kv[1]
				}
			}
		}

		// Merge current cluster nodes
		changed := false
		currentCluster := s.GetClusterNodes()
		for id, addr := range currentCluster {
			// Skip nodes that are marked as leaving
			if _, leaving := s.leavingNodes.Load(id); leaving {
				continue
			}

			if knownNodes[id] != addr {
				knownNodes[id] = addr
				changed = true
			}
		}

		// If changed, update RaftNode
		if changed {
			var peers []string
			for id, addr := range knownNodes {
				peers = append(peers, fmt.Sprintf("%s=%s", id, addr))
			}
			sort.Strings(peers)
			newVal := strings.Join(peers, ";")

			// Check again if we need to write to avoid loops if Get returned stale
			if newVal != currentVal {
				go func(k, v string) {
					if err := s.Set(k, v); err != nil {
						s.Raft.config.Logger.Warn("Failed to update RaftNode key: %v", err)
					}
				}("RaftNode", newVal)
			}
		}
	}
}

func (s *KVServer) checkConnections() {
	if !s.IsLeader() {
		return
	}

	// Read RaftNode key to find potential members that are missing
	val, ok := s.Get("RaftNode")
	if !ok || val == "" {
		return
	}

	// Parse saved nodes
	savedParts := strings.Split(val, ";")
	currentNodes := s.GetClusterNodes()

	// Invert currentNodes for address check
	currentAddrs := make(map[string]bool)
	for _, addr := range currentNodes {
		currentAddrs[addr] = true
	}

	for _, part := range savedParts {
		if part == "" {
			continue
		}
		// Expect id=addr
		kv := strings.SplitN(part, "=", 2)
		if len(kv) != 2 {
			continue
		}
		id, addr := kv[0], kv[1]

		// Skip invalid addresses
		if strings.HasPrefix(addr, ".") || !strings.Contains(addr, ":") {
			continue
		}

		if !currentAddrs[addr] {
			// Skip nodes that are marked as leaving
			if _, leaving := s.leavingNodes.Load(id); leaving {
				continue
			}

			// Found a node that was previously in the cluster but is now missing
			// Try to add it back
			// We use AddNodeWithForward which handles non-blocking internally somewhat,
			// but we should run this in goroutine to not block the loop
			go func(nodeID, nodeAddr string) {
				// Try to add node
				s.Raft.config.Logger.Info("Auto-rejoining node found in RaftNode: %s (%s)", nodeID, nodeAddr)
				if err := s.Join(nodeID, nodeAddr); err != nil {
					s.Raft.config.Logger.Debug("Failed to auto-rejoin node %s: %v", nodeID, err)
				}
			}(id, addr)
		}
	}
}

// startHTTPServer starts the HTTP API server
func (s *KVServer) startHTTPServer(addr string) error {
	mux := http.NewServeMux()

	// KV API
	mux.HandleFunc("/kv", func(w http.ResponseWriter, r *http.Request) {
		token := r.Header.Get("X-Raft-Token")
		
		switch r.Method {
		case http.MethodGet:
			key := r.URL.Query().Get("key")
			if key == "" {
				http.Error(w, "missing key", http.StatusBadRequest)
				return
			}
			// Use Authenticated method
			val, found, err := s.GetLinearAuthenticated(key, token)
			if err != nil {
				// Distinguish auth error vs raft error?
				if strings.Contains(err.Error(), "permission") || strings.Contains(err.Error(), "unauthorized") {
					http.Error(w, err.Error(), http.StatusForbidden)
				} else {
					http.Error(w, err.Error(), http.StatusInternalServerError)
				}
				return
			}
			if !found {
				http.Error(w, "not found", http.StatusNotFound)
				return
			}
			w.Write([]byte(val))

		case http.MethodPost:
			body, _ := io.ReadAll(r.Body)
			var req struct {
				Key   string `json:"key"`
				Value string `json:"value"`
			}
			if err := json.Unmarshal(body, &req); err != nil {
				http.Error(w, "invalid json", http.StatusBadRequest)
				return
			}
			if err := s.SetAuthenticated(req.Key, req.Value, token); err != nil {
				if strings.Contains(err.Error(), "permission") || strings.Contains(err.Error(), "unauthorized") {
					http.Error(w, err.Error(), http.StatusForbidden)
				} else {
					http.Error(w, err.Error(), http.StatusInternalServerError)
				}
				return
			}
			w.WriteHeader(http.StatusOK)

		case http.MethodDelete:
			key := r.URL.Query().Get("key")
			if key == "" {
				http.Error(w, "missing key", http.StatusBadRequest)
				return
			}
			if err := s.DelAuthenticated(key, token); err != nil {
				if strings.Contains(err.Error(), "permission") || strings.Contains(err.Error(), "unauthorized") {
					http.Error(w, err.Error(), http.StatusForbidden)
				} else {
					http.Error(w, err.Error(), http.StatusInternalServerError)
				}
				return
			}
			w.WriteHeader(http.StatusOK)
		
		default:
			http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
		}
	})

	// Auth API
	mux.HandleFunc("/auth/login", func(w http.ResponseWriter, r *http.Request) {
		if r.Method != http.MethodPost {
			http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
			return
		}
		var req struct {
			Username string `json:"username"`
			Password string `json:"password"`
			Code     string `json:"code"`
		}
		body, _ := io.ReadAll(r.Body)
		if err := json.Unmarshal(body, &req); err != nil {
			http.Error(w, "invalid json", http.StatusBadRequest)
			return
		}
		
		ip := r.RemoteAddr
		if host, _, err := net.SplitHostPort(r.RemoteAddr); err == nil {
			ip = host
		}
		
		token, err := s.AuthManager.Login(req.Username, req.Password, req.Code, ip)
		if err != nil {
			http.Error(w, err.Error(), http.StatusUnauthorized)
			return
		}
		
		resp := struct {
			Token string `json:"token"`
		}{Token: token}
		
		json.NewEncoder(w).Encode(resp)
	})

	// Watcher API
	mux.HandleFunc("/watch", func(w http.ResponseWriter, r *http.Request) {
		if r.Method != http.MethodPost {
			http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
			return
		}
		body, _ := io.ReadAll(r.Body)
		var req struct {
			Key string `json:"key"`
			URL string `json:"url"`
		}
		if err := json.Unmarshal(body, &req); err != nil {
			http.Error(w, "invalid json", http.StatusBadRequest)
			return
		}
		if req.Key == "" || req.URL == "" {
			http.Error(w, "missing key or url", http.StatusBadRequest)
			return
		}
		s.WatchURL(req.Key, req.URL)
		w.WriteHeader(http.StatusOK)
	})

	mux.HandleFunc("/unwatch", func(w http.ResponseWriter, r *http.Request) {
		if r.Method != http.MethodPost {
			http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
			return
		}
		body, _ := io.ReadAll(r.Body)
		var req struct {
			Key string `json:"key"`
			URL string `json:"url"`
		}
		if err := json.Unmarshal(body, &req); err != nil {
			http.Error(w, "invalid json", http.StatusBadRequest)
			return
		}
		s.UnwatchURL(req.Key, req.URL)
		w.WriteHeader(http.StatusOK)
	})

	s.httpServer = &http.Server{
		Addr:    addr,
		Handler: mux,
	}

	go func() {
		s.Raft.config.Logger.Info("HTTP API server listening on %s", addr)
		if err := s.httpServer.ListenAndServe(); err != nil && err != http.ErrServerClosed {
			s.Raft.config.Logger.Error("HTTP server failed: %v", err)
		}
	}()

	return nil
}