Alerting: Add option to use Redis in cluster mode for Alerting HA (#88696)

* Add config option to use Redis in cluster mode

* Use UniversalOptions

conf/defaults.ini

@@ -1180,7 +1180,12 @@ admin_config_poll_interval = 60s
 # The interval string is a possibly signed sequence of decimal numbers, followed by a unit suffix (ms, s, m, h, d), e.g. 30s or 1m.
 alertmanager_config_poll_interval = 60s
 
-# The redis server address that should be connected to.
+# Set to true when using redis in cluster mode.
+ha_redis_cluster_mode_enabled = false
+
+# The redis server address(es) that should be connected to.
+# Can either be a single address, or if using redis in cluster mode,
+# the cluster configuration address or a comma-separated list of addresses.
 ha_redis_address =
 
 # The username that should be used to authenticate with the redis server.

conf/sample.ini

@@ -1106,7 +1106,12 @@
 # The interval string is a possibly signed sequence of decimal numbers, followed by a unit suffix (ms, s, m, h, d), e.g. 30s or 1m.
 ;alertmanager_config_poll_interval = 60s
 
-# The redis server address that should be connected to.
+# Set to true when using redis in cluster mode.
+;ha_redis_cluster_mode_enabled = false
+
+# The redis server address(es) that should be connected to.
+# Can either be a single address, or if using redis in cluster mode,
+# the cluster configuration address or a comma-separated list of addresses.
 ;ha_redis_address =
 
 # The username that should be used to authenticate with the redis server.

pkg/services/ngalert/notifier/redis_peer.go

@@ -2,9 +2,11 @@ package notifier
 
 import (
 	"context"
+	"crypto/tls"
 	"slices"
 	"sort"
 	"strconv"
+	"strings"
 	"sync"
 	"time"
 
@@ -28,6 +30,7 @@ type redisConfig struct {
 	name        string
 	prefix      string
 	maxConns    int
+	clusterMode bool
 
 	tlsEnabled bool
 	tls        dstls.ClientConfig
@@ -55,7 +58,7 @@ const (
 
 type redisPeer struct {
 	name      string
-	redis     *redis.Client
+	redis     redis.UniversalClient
 	prefix    string
 	logger    log.Logger
 	states    map[string]alertingCluster.State
@@ -95,25 +98,34 @@ func newRedisPeer(cfg redisConfig, logger log.Logger, reg prometheus.Registerer,
 		poolSize = cfg.maxConns
 	}
 
-	opts := &redis.Options{
-		Addr:     cfg.addr,
+	addrs := strings.Split(cfg.addr, ",")
+
+	var tlsClientConfig *tls.Config
+	var err error
+	if cfg.tlsEnabled {
+		tlsClientConfig, err = cfg.tls.GetTLSConfig()
+		if err != nil {
+			logger.Error("Failed to get TLS config", "err", err)
+			return nil, err
+		}
+	}
+
+	opts := &redis.UniversalOptions{
+		Addrs:     addrs,
 		Username:  cfg.username,
 		Password:  cfg.password,
 		DB:        cfg.db,
 		PoolSize:  poolSize,
+		TLSConfig: tlsClientConfig,
 	}
 
-	if cfg.tlsEnabled {
-		tlsClientConfig, err := cfg.tls.GetTLSConfig()
-		if err != nil {
-			logger.Error("Failed to get TLS config", "err", err)
-			return nil, err
+	var rdb redis.UniversalClient
+	if cfg.clusterMode {
+		rdb = redis.NewClusterClient(opts.Cluster())
 	} else {
-			opts.TLSConfig = tlsClientConfig
-		}
+		rdb = redis.NewClient(opts.Simple())
 	}
 
-	rdb := redis.NewClient(opts)
 	cmd := rdb.Ping(context.Background())
 	if cmd.Err() != nil {
 		logger.Error("Failed to ping redis - redis-based alertmanager clustering may not be available", "err", cmd.Err())

pkg/services/ngalert/notifier/redis_peer_test.go

@@ -16,6 +16,35 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+func TestNewRedisPeerClusterMode(t *testing.T) {
+	// Write client and server certificates/keys to tempDir, both issued by the same CA
+	certPaths := createX509TestDir(t)
+
+	// Set up tls.Config and start miniredis with server-side TLS
+	x509Cert, err := tls.LoadX509KeyPair(certPaths.serverCert, certPaths.serverKey)
+	require.NoError(t, err)
+
+	mr, err := miniredis.RunTLS(&tls.Config{
+		Certificates: []tls.Certificate{x509Cert},
+		ClientAuth:   tls.NoClientCert,
+	})
+	require.NoError(t, err)
+	defer mr.Close()
+
+	redisPeer, err := newRedisPeer(redisConfig{
+		clusterMode: true,
+		addr:        mr.Addr(),
+		tlsEnabled:  true,
+		tls: dstls.ClientConfig{
+			CAPath:     certPaths.ca,
+			ServerName: "localhost",
+		}}, log.NewNopLogger(), prometheus.NewRegistry(), time.Second*60)
+	require.NoError(t, err)
+
+	ping := redisPeer.redis.Ping(context.Background())
+	require.NoError(t, ping.Err())
+}
+
 func TestNewRedisPeerWithTLS(t *testing.T) {
 	// Write client and server certificates/keys to tempDir, both issued by the same CA
 	certPaths := createX509TestDir(t)

pkg/setting/setting_unified_alerting.go

@@ -73,6 +73,7 @@ type UnifiedAlertingSettings struct {
 	HAGossipInterval               time.Duration
 	HAPushPullInterval             time.Duration
 	HALabel                        string
+	HARedisClusterModeEnabled      bool
 	HARedisAddr                    string
 	HARedisPeerName                string
 	HARedisPrefix                  string
@@ -222,6 +223,7 @@ func (cfg *Cfg) ReadUnifiedAlertingSettings(iniFile *ini.File) error {
 	uaCfg.HAListenAddr = ua.Key("ha_listen_address").MustString(alertmanagerDefaultClusterAddr)
 	uaCfg.HAAdvertiseAddr = ua.Key("ha_advertise_address").MustString("")
 	uaCfg.HALabel = ua.Key("ha_label").MustString("")
+	uaCfg.HARedisClusterModeEnabled = ua.Key("ha_redis_cluster_mode_enabled").MustBool(false)
 	uaCfg.HARedisAddr = ua.Key("ha_redis_address").MustString("")
 	uaCfg.HARedisPeerName = ua.Key("ha_redis_peer_name").MustString("")
 	uaCfg.HARedisPrefix = ua.Key("ha_redis_prefix").MustString("")