sentinel.go

package redis

import (
	"context"
	"crypto/tls"
	"errors"
	"net"
	"strings"
	"sync"
	"time"

	"github.com/mujiantao/redis/v8/internal"
	"github.com/mujiantao/redis/v8/internal/pool"
	"github.com/mujiantao/redis/v8/internal/rand"
)

//------------------------------------------------------------------------------

// FailoverOptions are used to configure a failover client and should
// be passed to NewFailoverClient.
type FailoverOptions struct {
	// The master name.
	MasterName string
	// A seed list of host:port addresses of sentinel nodes.
	SentinelAddrs []string

	// If specified with SentinelPassword, enables ACL-based authentication (via
	// AUTH <user> <pass>).
	SentinelUsername string
	// Sentinel password from "requirepass <password>" (if enabled) in Sentinel
	// configuration, or, if SentinelUsername is also supplied, used for ACL-based
	// authentication.
	SentinelPassword string

	// Allows routing read-only commands to the closest master or slave node.
	// This option only works with NewFailoverClusterClient.
	RouteByLatency bool
	// Allows routing read-only commands to the random master or slave node.
	// This option only works with NewFailoverClusterClient.
	RouteRandomly bool

	// Route all commands to slave read-only nodes.
	SlaveOnly bool

	// Use slaves disconnected with master when cannot get connected slaves
	// Now, this option only works in RandomSlaveAddr function.
	UseDisconnectedSlaves bool

	// Following options are copied from Options struct.

	Dialer    func(ctx context.Context, network, addr string) (net.Conn, error)
	OnConnect func(ctx context.Context, cn *Conn) error

	Username string
	Password string
	DB       int

	MaxRetries      int
	MinRetryBackoff time.Duration
	MaxRetryBackoff time.Duration

	DialTimeout  time.Duration
	ReadTimeout  time.Duration
	WriteTimeout time.Duration

	// PoolFIFO uses FIFO mode for each node connection pool GET/PUT (default LIFO).
	PoolFIFO bool

	PoolSize           int
	MinIdleConns       int
	MaxConnAge         time.Duration
	PoolTimeout        time.Duration
	IdleTimeout        time.Duration
	IdleCheckFrequency time.Duration

	TLSConfig *tls.Config
}

func (opt *FailoverOptions) clientOptions() *Options {
	return &Options{
		Addr: "FailoverClient",

		Dialer:    opt.Dialer,
		OnConnect: opt.OnConnect,

		DB:       opt.DB,
		Username: opt.Username,
		Password: opt.Password,

		MaxRetries:      opt.MaxRetries,
		MinRetryBackoff: opt.MinRetryBackoff,
		MaxRetryBackoff: opt.MaxRetryBackoff,

		DialTimeout:  opt.DialTimeout,
		ReadTimeout:  opt.ReadTimeout,
		WriteTimeout: opt.WriteTimeout,

		PoolFIFO:           opt.PoolFIFO,
		PoolSize:           opt.PoolSize,
		PoolTimeout:        opt.PoolTimeout,
		IdleTimeout:        opt.IdleTimeout,
		IdleCheckFrequency: opt.IdleCheckFrequency,
		MinIdleConns:       opt.MinIdleConns,
		MaxConnAge:         opt.MaxConnAge,

		TLSConfig: opt.TLSConfig,
	}
}

func (opt *FailoverOptions) sentinelOptions(addr string) *Options {
	return &Options{
		Addr: addr,

		Dialer:    opt.Dialer,
		OnConnect: opt.OnConnect,

		DB:       0,
		Username: opt.SentinelUsername,
		Password: opt.SentinelPassword,

		MaxRetries:      opt.MaxRetries,
		MinRetryBackoff: opt.MinRetryBackoff,
		MaxRetryBackoff: opt.MaxRetryBackoff,

		DialTimeout:  opt.DialTimeout,
		ReadTimeout:  opt.ReadTimeout,
		WriteTimeout: opt.WriteTimeout,

		PoolFIFO:           opt.PoolFIFO,
		PoolSize:           opt.PoolSize,
		PoolTimeout:        opt.PoolTimeout,
		IdleTimeout:        opt.IdleTimeout,
		IdleCheckFrequency: opt.IdleCheckFrequency,
		MinIdleConns:       opt.MinIdleConns,
		MaxConnAge:         opt.MaxConnAge,

		TLSConfig: opt.TLSConfig,
	}
}

func (opt *FailoverOptions) clusterOptions() *ClusterOptions {
	return &ClusterOptions{
		Dialer:    opt.Dialer,
		OnConnect: opt.OnConnect,

		Username: opt.Username,
		Password: opt.Password,

		MaxRedirects: opt.MaxRetries,

		RouteByLatency: opt.RouteByLatency,
		RouteRandomly:  opt.RouteRandomly,

		MinRetryBackoff: opt.MinRetryBackoff,
		MaxRetryBackoff: opt.MaxRetryBackoff,

		DialTimeout:  opt.DialTimeout,
		ReadTimeout:  opt.ReadTimeout,
		WriteTimeout: opt.WriteTimeout,

		PoolFIFO:           opt.PoolFIFO,
		PoolSize:           opt.PoolSize,
		PoolTimeout:        opt.PoolTimeout,
		IdleTimeout:        opt.IdleTimeout,
		IdleCheckFrequency: opt.IdleCheckFrequency,
		MinIdleConns:       opt.MinIdleConns,
		MaxConnAge:         opt.MaxConnAge,

		TLSConfig: opt.TLSConfig,
	}
}

// NewFailoverClient returns a Redis client that uses Redis Sentinel
// for automatic failover. It's safe for concurrent use by multiple
// goroutines.
func NewFailoverClient(failoverOpt *FailoverOptions) *Client {
	if failoverOpt.RouteByLatency {
		panic("to route commands by latency, use NewFailoverClusterClient")
	}
	if failoverOpt.RouteRandomly {
		panic("to route commands randomly, use NewFailoverClusterClient")
	}

	sentinelAddrs := make([]string, len(failoverOpt.SentinelAddrs))
	copy(sentinelAddrs, failoverOpt.SentinelAddrs)

	rand.Shuffle(len(sentinelAddrs), func(i, j int) {
		sentinelAddrs[i], sentinelAddrs[j] = sentinelAddrs[j], sentinelAddrs[i]
	})

	failover := &sentinelFailover{
		opt:           failoverOpt,
		sentinelAddrs: sentinelAddrs,
	}

	opt := failoverOpt.clientOptions()
	opt.Dialer = masterSlaveDialer(failover)
	opt.init()

	connPool := newConnPool(opt)

	failover.mu.Lock()
	failover.onFailover = func(ctx context.Context, addr string) {
		_ = connPool.Filter(func(cn *pool.Conn) bool {
			return cn.RemoteAddr().String() != addr
		})
	}
	failover.mu.Unlock()

	c := Client{
		baseClient: newBaseClient(opt, connPool),
		ctx:        context.Background(),
	}
	c.cmdable = c.Process
	c.onClose = failover.Close

	return &c
}

func masterSlaveDialer(
	failover *sentinelFailover,
) func(ctx context.Context, network, addr string) (net.Conn, error) {
	return func(ctx context.Context, network, _ string) (net.Conn, error) {
		var addr string
		var err error

		if failover.opt.SlaveOnly {
			addr, err = failover.RandomSlaveAddr(ctx)
		} else {
			addr, err = failover.MasterAddr(ctx)
			if err == nil {
				failover.trySwitchMaster(ctx, addr)
			}
		}
		if err != nil {
			return nil, err
		}
		if failover.opt.Dialer != nil {
			return failover.opt.Dialer(ctx, network, addr)
		}

		netDialer := &net.Dialer{
			Timeout:   failover.opt.DialTimeout,
			KeepAlive: 5 * time.Minute,
		}
		if failover.opt.TLSConfig == nil {
			return netDialer.DialContext(ctx, network, addr)
		}
		return tls.DialWithDialer(netDialer, network, addr, failover.opt.TLSConfig)
	}
}

//------------------------------------------------------------------------------

// SentinelClient is a client for a Redis Sentinel.
type SentinelClient struct {
	*baseClient
	hooks
	ctx context.Context
}

func NewSentinelClient(opt *Options) *SentinelClient {
	opt.init()
	c := &SentinelClient{
		baseClient: &baseClient{
			opt:      opt,
			connPool: newConnPool(opt),
		},
		ctx: context.Background(),
	}
	return c
}

func (c *SentinelClient) Context() context.Context {
	return c.ctx
}

func (c *SentinelClient) WithContext(ctx context.Context) *SentinelClient {
	if ctx == nil {
		panic("nil context")
	}
	clone := *c
	clone.ctx = ctx
	return &clone
}

func (c *SentinelClient) Process(ctx context.Context, cmd Cmder) error {
	return c.hooks.process(ctx, cmd, c.baseClient.process)
}

func (c *SentinelClient) pubSub() *PubSub {
	pubsub := &PubSub{
		opt: c.opt,

		newConn: func(ctx context.Context, channels []string) (*pool.Conn, error) {
			return c.newConn(ctx)
		},
		closeConn: c.connPool.CloseConn,
	}
	pubsub.init()
	return pubsub
}

// Ping is used to test if a connection is still alive, or to
// measure latency.
func (c *SentinelClient) Ping(ctx context.Context) *StringCmd {
	cmd := NewStringCmd(ctx, "ping")
	_ = c.Process(ctx, cmd)
	return cmd
}

// Subscribe subscribes the client to the specified channels.
// Channels can be omitted to create empty subscription.
func (c *SentinelClient) Subscribe(ctx context.Context, channels ...string) *PubSub {
	pubsub := c.pubSub()
	if len(channels) > 0 {
		_ = pubsub.Subscribe(ctx, channels...)
	}
	return pubsub
}

// PSubscribe subscribes the client to the given patterns.
// Patterns can be omitted to create empty subscription.
func (c *SentinelClient) PSubscribe(ctx context.Context, channels ...string) *PubSub {
	pubsub := c.pubSub()
	if len(channels) > 0 {
		_ = pubsub.PSubscribe(ctx, channels...)
	}
	return pubsub
}

func (c *SentinelClient) GetMasterAddrByName(ctx context.Context, name string) *StringSliceCmd {
	cmd := NewStringSliceCmd(ctx, "sentinel", "get-master-addr-by-name", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

func (c *SentinelClient) Sentinels(ctx context.Context, name string) *SliceCmd {
	cmd := NewSliceCmd(ctx, "sentinel", "sentinels", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

// Failover forces a failover as if the master was not reachable, and without
// asking for agreement to other Sentinels.
func (c *SentinelClient) Failover(ctx context.Context, name string) *StatusCmd {
	cmd := NewStatusCmd(ctx, "sentinel", "failover", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

// Reset resets all the masters with matching name. The pattern argument is a
// glob-style pattern. The reset process clears any previous state in a master
// (including a failover in progress), and removes every slave and sentinel
// already discovered and associated with the master.
func (c *SentinelClient) Reset(ctx context.Context, pattern string) *IntCmd {
	cmd := NewIntCmd(ctx, "sentinel", "reset", pattern)
	_ = c.Process(ctx, cmd)
	return cmd
}

// FlushConfig forces Sentinel to rewrite its configuration on disk, including
// the current Sentinel state.
func (c *SentinelClient) FlushConfig(ctx context.Context) *StatusCmd {
	cmd := NewStatusCmd(ctx, "sentinel", "flushconfig")
	_ = c.Process(ctx, cmd)
	return cmd
}

// Master shows the state and info of the specified master.
func (c *SentinelClient) Master(ctx context.Context, name string) *StringStringMapCmd {
	cmd := NewStringStringMapCmd(ctx, "sentinel", "master", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

// Masters shows a list of monitored masters and their state.
func (c *SentinelClient) Masters(ctx context.Context) *SliceCmd {
	cmd := NewSliceCmd(ctx, "sentinel", "masters")
	_ = c.Process(ctx, cmd)
	return cmd
}

// Slaves shows a list of slaves for the specified master and their state.
func (c *SentinelClient) Slaves(ctx context.Context, name string) *SliceCmd {
	cmd := NewSliceCmd(ctx, "sentinel", "slaves", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

// CkQuorum checks if the current Sentinel configuration is able to reach the
// quorum needed to failover a master, and the majority needed to authorize the
// failover. This command should be used in monitoring systems to check if a
// Sentinel deployment is ok.
func (c *SentinelClient) CkQuorum(ctx context.Context, name string) *StringCmd {
	cmd := NewStringCmd(ctx, "sentinel", "ckquorum", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

// Monitor tells the Sentinel to start monitoring a new master with the specified
// name, ip, port, and quorum.
func (c *SentinelClient) Monitor(ctx context.Context, name, ip, port, quorum string) *StringCmd {
	cmd := NewStringCmd(ctx, "sentinel", "monitor", name, ip, port, quorum)
	_ = c.Process(ctx, cmd)
	return cmd
}

// Set is used in order to change configuration parameters of a specific master.
func (c *SentinelClient) Set(ctx context.Context, name, option, value string) *StringCmd {
	cmd := NewStringCmd(ctx, "sentinel", "set", name, option, value)
	_ = c.Process(ctx, cmd)
	return cmd
}

// Remove is used in order to remove the specified master: the master will no
// longer be monitored, and will totally be removed from the internal state of
// the Sentinel.
func (c *SentinelClient) Remove(ctx context.Context, name string) *StringCmd {
	cmd := NewStringCmd(ctx, "sentinel", "remove", name)
	_ = c.Process(ctx, cmd)
	return cmd
}

//------------------------------------------------------------------------------

type sentinelFailover struct {
	opt *FailoverOptions

	sentinelAddrs []string

	onFailover func(ctx context.Context, addr string)
	onUpdate   func(ctx context.Context)

	mu          sync.RWMutex
	_masterAddr string
	sentinel    *SentinelClient
	pubsub      *PubSub
}

func (c *sentinelFailover) Close() error {
	c.mu.Lock()
	defer c.mu.Unlock()
	if c.sentinel != nil {
		return c.closeSentinel()
	}
	return nil
}

func (c *sentinelFailover) closeSentinel() error {
	firstErr := c.pubsub.Close()
	c.pubsub = nil

	err := c.sentinel.Close()
	if err != nil && firstErr == nil {
		firstErr = err
	}
	c.sentinel = nil

	return firstErr
}

func (c *sentinelFailover) RandomSlaveAddr(ctx context.Context) (string, error) {
	if c.opt == nil {
		return "", errors.New("opt is nil")
	}

	addresses, err := c.slaveAddrs(ctx, false)
	if err != nil {
		return "", err
	}

	if len(addresses) == 0 && c.opt.UseDisconnectedSlaves {
		addresses, err = c.slaveAddrs(ctx, true)
		if err != nil {
			return "", err
		}
	}

	if len(addresses) == 0 {
		return c.MasterAddr(ctx)
	}
	return addresses[rand.Intn(len(addresses))], nil
}

func (c *sentinelFailover) MasterAddr(ctx context.Context) (string, error) {
	c.mu.RLock()
	sentinel := c.sentinel
	c.mu.RUnlock()

	if sentinel != nil {
		addr := c.getMasterAddr(ctx, sentinel)
		if addr != "" {
			return addr, nil
		}
	}

	c.mu.Lock()
	defer c.mu.Unlock()

	if c.sentinel != nil {
		addr := c.getMasterAddr(ctx, c.sentinel)
		if addr != "" {
			return addr, nil
		}
		_ = c.closeSentinel()
	}

	for i, sentinelAddr := range c.sentinelAddrs {
		sentinel := NewSentinelClient(c.opt.sentinelOptions(sentinelAddr))

		masterAddr, err := sentinel.GetMasterAddrByName(ctx, c.opt.MasterName).Result()
		if err != nil {
			internal.Logger.Printf(ctx, "sentinel: GetMasterAddrByName master=%q failed: %s",
				c.opt.MasterName, err)
			_ = sentinel.Close()
			continue
		}

		// Push working sentinel to the top.
		c.sentinelAddrs[0], c.sentinelAddrs[i] = c.sentinelAddrs[i], c.sentinelAddrs[0]
		c.setSentinel(ctx, sentinel)

		addr := net.JoinHostPort(masterAddr[0], masterAddr[1])
		return addr, nil
	}

	return "", errors.New("redis: all sentinels specified in configuration are unreachable")
}

func (c *sentinelFailover) slaveAddrs(ctx context.Context, useDisconnected bool) ([]string, error) {
	c.mu.RLock()
	sentinel := c.sentinel
	c.mu.RUnlock()

	if sentinel != nil {
		addrs := c.getSlaveAddrs(ctx, sentinel)
		if len(addrs) > 0 {
			return addrs, nil
		}
	}

	c.mu.Lock()
	defer c.mu.Unlock()

	if c.sentinel != nil {
		addrs := c.getSlaveAddrs(ctx, c.sentinel)
		if len(addrs) > 0 {
			return addrs, nil
		}
		_ = c.closeSentinel()
	}

	var sentinelReachable bool

	for i, sentinelAddr := range c.sentinelAddrs {
		sentinel := NewSentinelClient(c.opt.sentinelOptions(sentinelAddr))

		slaves, err := sentinel.Slaves(ctx, c.opt.MasterName).Result()
		if err != nil {
			internal.Logger.Printf(ctx, "sentinel: Slaves master=%q failed: %s",
				c.opt.MasterName, err)
			_ = sentinel.Close()
			continue
		}
		sentinelReachable = true
		addrs := parseSlaveAddrs(slaves, useDisconnected)
		if len(addrs) == 0 {
			continue
		}
		// Push working sentinel to the top.
		c.sentinelAddrs[0], c.sentinelAddrs[i] = c.sentinelAddrs[i], c.sentinelAddrs[0]
		c.setSentinel(ctx, sentinel)

		return addrs, nil
	}

	if sentinelReachable {
		return []string{}, nil
	}
	return []string{}, errors.New("redis: all sentinels specified in configuration are unreachable")
}

func (c *sentinelFailover) getMasterAddr(ctx context.Context, sentinel *SentinelClient) string {
	addr, err := sentinel.GetMasterAddrByName(ctx, c.opt.MasterName).Result()
	if err != nil {
		internal.Logger.Printf(ctx, "sentinel: GetMasterAddrByName name=%q failed: %s",
			c.opt.MasterName, err)
		return ""
	}
	return net.JoinHostPort(addr[0], addr[1])
}

func (c *sentinelFailover) getSlaveAddrs(ctx context.Context, sentinel *SentinelClient) []string {
	addrs, err := sentinel.Slaves(ctx, c.opt.MasterName).Result()
	if err != nil {
		internal.Logger.Printf(ctx, "sentinel: Slaves name=%q failed: %s",
			c.opt.MasterName, err)
		return []string{}
	}
	return parseSlaveAddrs(addrs, false)
}

func parseSlaveAddrs(addrs []interface{}, keepDisconnected bool) []string {
	nodes := make([]string, 0, len(addrs))
	for _, node := range addrs {
		ip := ""
		port := ""
		flags := []string{}
		lastkey := ""
		isDown := false

		for _, key := range node.([]interface{}) {
			switch lastkey {
			case "ip":
				ip = key.(string)
			case "port":
				port = key.(string)
			case "flags":
				flags = strings.Split(key.(string), ",")
			}
			lastkey = key.(string)
		}

		for _, flag := range flags {
			switch flag {
			case "s_down", "o_down":
				isDown = true
			case "disconnected":
				if !keepDisconnected {
					isDown = true
				}
			}
		}

		if !isDown {
			nodes = append(nodes, net.JoinHostPort(ip, port))
		}
	}

	return nodes
}

func (c *sentinelFailover) trySwitchMaster(ctx context.Context, addr string) {
	c.mu.RLock()
	currentAddr := c._masterAddr //nolint:ifshort
	c.mu.RUnlock()

	if addr == currentAddr {
		return
	}

	c.mu.Lock()
	defer c.mu.Unlock()

	if addr == c._masterAddr {
		return
	}
	c._masterAddr = addr

	internal.Logger.Printf(ctx, "sentinel: new master=%q addr=%q",
		c.opt.MasterName, addr)
	if c.onFailover != nil {
		c.onFailover(ctx, addr)
	}
}

func (c *sentinelFailover) setSentinel(ctx context.Context, sentinel *SentinelClient) {
	if c.sentinel != nil {
		panic("not reached")
	}
	c.sentinel = sentinel
	c.discoverSentinels(ctx)

	c.pubsub = sentinel.Subscribe(ctx, "+switch-master", "+slave-reconf-done")
	go c.listen(c.pubsub)
}

func (c *sentinelFailover) discoverSentinels(ctx context.Context) {
	sentinels, err := c.sentinel.Sentinels(ctx, c.opt.MasterName).Result()
	if err != nil {
		internal.Logger.Printf(ctx, "sentinel: Sentinels master=%q failed: %s", c.opt.MasterName, err)
		return
	}
	for _, sentinel := range sentinels {
		vals := sentinel.([]interface{})
		var ip, port string
		for i := 0; i < len(vals); i += 2 {
			key := vals[i].(string)
			switch key {
			case "ip":
				ip = vals[i+1].(string)
			case "port":
				port = vals[i+1].(string)
			}
		}
		if ip != "" && port != "" {
			sentinelAddr := net.JoinHostPort(ip, port)
			if !contains(c.sentinelAddrs, sentinelAddr) {
				internal.Logger.Printf(ctx, "sentinel: discovered new sentinel=%q for master=%q",
					sentinelAddr, c.opt.MasterName)
				c.sentinelAddrs = append(c.sentinelAddrs, sentinelAddr)
			}
		}
	}
}

func (c *sentinelFailover) listen(pubsub *PubSub) {
	ctx := context.TODO()

	if c.onUpdate != nil {
		c.onUpdate(ctx)
	}

	ch := pubsub.Channel()
	for msg := range ch {
		if msg.Channel == "+switch-master" {
			parts := strings.Split(msg.Payload, " ")
			if parts[0] != c.opt.MasterName {
				internal.Logger.Printf(pubsub.getContext(), "sentinel: ignore addr for master=%q", parts[0])
				continue
			}
			addr := net.JoinHostPort(parts[3], parts[4])
			c.trySwitchMaster(pubsub.getContext(), addr)
		}

		if c.onUpdate != nil {
			c.onUpdate(ctx)
		}
	}
}

func contains(slice []string, str string) bool {
	for _, s := range slice {
		if s == str {
			return true
		}
	}
	return false
}

//------------------------------------------------------------------------------

// NewFailoverClusterClient returns a client that supports routing read-only commands
// to a slave node.
func NewFailoverClusterClient(failoverOpt *FailoverOptions) *ClusterClient {
	sentinelAddrs := make([]string, len(failoverOpt.SentinelAddrs))
	copy(sentinelAddrs, failoverOpt.SentinelAddrs)

	failover := &sentinelFailover{
		opt:           failoverOpt,
		sentinelAddrs: sentinelAddrs,
	}

	opt := failoverOpt.clusterOptions()
	opt.ClusterSlots = func(ctx context.Context) ([]ClusterSlot, error) {
		masterAddr, err := failover.MasterAddr(ctx)
		if err != nil {
			return nil, err
		}

		nodes := []ClusterNode{{
			Addr: masterAddr,
		}}

		slaveAddrs, err := failover.slaveAddrs(ctx, false)
		if err != nil {
			return nil, err
		}

		for _, slaveAddr := range slaveAddrs {
			nodes = append(nodes, ClusterNode{
				Addr: slaveAddr,
			})
		}

		slots := []ClusterSlot{
			{
				Start: 0,
				End:   16383,
				Nodes: nodes,
			},
		}
		return slots, nil
	}

	c := NewClusterClient(opt)

	failover.mu.Lock()
	failover.onUpdate = func(ctx context.Context) {
		c.ReloadState(ctx)
	}
	failover.mu.Unlock()

	return c
}