feat: Redis cluster support.

[daanvinken]

what

• Add Redis Cluster mode support, auto-detected from config (no explicit mode flag)
• Introduce Config struct + NewWithConfig() constructor, deprecate old New()
• Change RedisDB.client to redis.Cmdable interface (common to Client and ClusterClient)
• Replace Keys() with Scan() in lock key migration (Scan fans out across cluster nodes, Keys doesn't)
• Add --redis-username flag for Redis 6+ ACL auth
• Add --redis-cluster-addresses flag accepting comma-separated node addresses
• /healthz now returns 503 when database is unreachable (added Ping() to Database interface)
• Apply lock check fails closed — rejects apply when lock backend is unreachable instead of silently proceeding
• Handle Kubernetes service link env var collisions (ATLANTIS_REDIS_PORT=tcp://... parsed as int)

why

• Picks up where feat: support redis cluster deployment #5529 left off, addressing reviewer feedback (auto-detection over explicit --redis-deployment flag)
• Keys() is incompatible with Redis Cluster — only queries the connected node. Scan() via go-redis ClusterClient fans out across all nodes.
• Cmdable interface is the narrowest type that covers both client types. Tradeoff: Close() requires a type switch since Cmdable doesn't expose it.
• Fail-open on apply lock check meant a Redis outage silently disabled the global apply lock — dangerous in production.
• K8s auto-creates ATLANTIS_REDIS_PORT=tcp://10.x.x.x:6379 for services in the same namespace; Viper picks it up via the ATLANTIS_ prefix and fails to parse as int.

tests

• Unit tests for NewWithConfig: single-node, cluster, empty-address edge cases
• Unit test for sanitizeKubernetesServiceLinks: URL detection and reset to defaults
• Updated TestApplyCommandRunner_IsLocked to verify fail-closed behavior
• Deployed against a 3-node Redis Cluster (Bitnami helm chart) in staging. Here I've tested some failovers, locking/unlocking on different PRs etc.

When Redis cluster down:

New logging:

{"level":"info","ts":"2026-03-09T15:26:43.365Z","caller":"server/server.go:505","msg":"Utilizing Redis DB in cluster mode, addresses: atlantis-redis-cluster-headless.atlantis.svc.cluster.local:6379","json":{}}

references

• Continues feat: support redis cluster deployment #5529
• Optionally, I've also worked on sentinel support will be submitted separately on top of this.

[Copilot]

Pull request overview

Adds Redis Cluster support and improves operational safety by enhancing health checks and making apply-lock enforcement fail closed when the lock backend is unreachable.

Changes:

• Introduces Redis Config + NewWithConfig() to auto-select single-node vs cluster client and adds username/cluster address configuration.
• Updates lock key migration to use SCAN (cluster-compatible) and adds Database.Ping() for /healthz 503 behavior.
• Adds Kubernetes service-link env var sanitization to avoid ATLANTIS_REDIS_PORT=tcp://... parsing issues and updates apply-lock error behavior/tests.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
server/user_config.go	Adds config fields for Redis username and cluster addresses.
server/server.go	Wires Redis config into server creation and makes /healthz depend on DB reachability.
server/events/apply_command_runner.go	Fails closed on apply-lock check errors and posts user-facing PR comment.
server/events/apply_command_runner_test.go	Updates expectations for fail-closed apply-lock behavior.
server/core/redis/redis.go	Adds Config, cluster support, SCAN migration, Ping(), and Cmdable client abstraction.
server/core/redis/redis_test.go	Adds unit tests for NewWithConfig behavior and edge cases.
server/core/db/db.go	Extends Database interface with Ping().
server/core/db/mocks/mock_database.go	Updates mock to implement Ping().
server/core/boltdb/boltdb.go	Implements Ping() for BoltDB.
cmd/server.go	Adds flags for Redis username/cluster addresses and sanitizes K8s service-link env collisions.
cmd/server_test.go	Adds tests for Kubernetes service-link sanitization.
runatlantis.io/docs/server-configuration.md	Documents new Redis username and cluster address options.

[daanvinken]

Note that this also works for BoltdB and performs a simple read.

[daanvinken]

@jamengual @javking07 do you mind having a look here?

[Copilot]

Pull request overview

Copilot reviewed 12 out of 12 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (1)

server/server.go:535

• The locking-db-type switch has no default case, so an unsupported value leaves database nil and the server continues initializing. With the new /healthz behavior this can now cause a nil dereference (and even without it, later DB usage will be unsafe). Add a default that returns an error for unknown db types (or otherwise enforce that database is always initialized).

	switch dbtype := userConfig.LockingDBType; dbtype {
	case "redis":
		var clusterAddrs []string
		if userConfig.RedisClusterAddresses != "" {
			rawClusterAddrs := strings.Split(userConfig.RedisClusterAddresses, ",")
			for _, addr := range rawClusterAddrs {
				trimmed := strings.TrimSpace(addr)
				if trimmed == "" {
					continue
				}
				clusterAddrs = append(clusterAddrs, trimmed)
			}
		}
		switch {
		case len(clusterAddrs) > 0:
			logger.Info(fmt.Sprintf("Utilizing Redis DB in cluster mode, addresses: %s", userConfig.RedisClusterAddresses))
		default:
			logger.Info(fmt.Sprintf("Utilizing Redis DB in single-node mode, host: %s, port: %d", userConfig.RedisHost, userConfig.RedisPort))
		}
		database, err = redis.NewWithConfig(redis.Config{
			Hostname:           userConfig.RedisHost,
			Port:               userConfig.RedisPort,
			Password:           userConfig.RedisPassword,
			Username:           userConfig.RedisUsername,
			TLSEnabled:         userConfig.RedisTLSEnabled,
			InsecureSkipVerify: userConfig.RedisInsecureSkipVerify,
			DB:                 userConfig.RedisDB,
			ClusterAddresses:   clusterAddrs,
		})
		if err != nil {
			return nil, err
		}
	case "boltdb":
		logger.Info("Utilizing BoltDB")
		database, err = boltdb.New(userConfig.DataDir)
		if err != nil {
			return nil, err
		}
	}

[matthewmrichter]

This would be really wonderful to be able to get horizontal autoscaling going.