Valkey
DIAL Core keeps volatile data in Valkey for fast access: cached resources, writes waiting to be synced to blob storage, per-request API keys, per-trace usage records, and coordination between Core instances. These static settings connect Core to a single Valkey server, a Valkey cluster, or a managed cache with IAM authentication. This page is for platform engineers who provision Valkey for Core.
Valkey or Redis
Core talks to its cache over the Redis protocol, so the settings keep the redis prefix, addresses use the redis:// scheme, and the same settings connect Core to Redis. Since version 7.0.0 of the dial Helm chart, the DIAL Helm chart deploys Valkey instead of Redis.
Check the license before you choose Redis. Redis 7.2 and earlier are under BSD-3-Clause, Redis 7.4 under RSALv2 or SSPLv1, and Redis 8.0 and later under RSALv2, SSPLv1, or AGPLv3; see Redis licenses. Valkey is under BSD-3-Clause.
Connection
Set exactly one of the two connection settings.
| Setting | Default | Description |
|---|---|---|
redis.singleServerConfig.address | — | The address of a single server, for example redis://valkey:6379. Use rediss:// for TLS. |
redis.clusterServersConfig.nodeAddresses | — | A JSON array of cluster node addresses, for example ["redis://host1:6379","redis://host2:6379"]. Use it only for a cache that runs in cluster mode. |
The DIAL Helm chart runs a single Valkey node without replicas. For production, use a replicated Valkey deployment or a managed cache.
Authentication
By default, Valkey's default user has no password and can run every command. Turn on ACL authentication and give Core a user that can run only the commands it needs, for example in valkey.conf:
user dial-core on ><YOUR_PASSWORD> ~* allchannels +@read +@write +ping +info +@hash +@list +@pubsub +@scripting +TIME
| Setting | Default | Description |
|---|---|---|
redis.singleServerConfig.username | — | The ACL user. Without it, Valkey signs Core in as default. For a cluster, use redis.clusterServersConfig.username. |
redis.singleServerConfig.password | — | The user's password. For a cluster, use redis.clusterServersConfig.password. Pass it as a secret. |
With core.valkey.auth.enabled: true, the DIAL Helm chart sets both for you. It signs Core in as default, with the password from the chart's own Secret or from core.valkey.auth.usersExistingSecret. The chart's default permissions for that user also allow +psync and +replconf.
Managed cache with IAM authentication
redis.provider.* turns on provider-specific authentication: Core gets a short-lived token from the cloud provider and signs in with it instead of a password. Core finds its cloud credentials the standard way for each provider: the AWS default credentials provider chain, Google application default credentials, or Azure DefaultAzureCredential.
| Setting | Applies to | Description |
|---|---|---|
redis.provider.name | All | The provider: aws-elasti-cache (ElastiCache for Valkey or Redis OSS, IAM authentication), gcp-memory-store (Memorystore for Valkey, IAM authentication), or azure-redis-cache (Azure Cache for Redis, Microsoft Entra authentication). |
redis.provider.userId | aws-elasti-cache | The IAM-enabled user ID. |
redis.provider.region | aws-elasti-cache | The region of the cache. |
redis.provider.clusterName | aws-elasti-cache | The cluster name. |
redis.provider.serverless | aws-elasti-cache | Whether the cache is serverless. |
redis.provider.accountName | gcp-memory-store | The service account resource name, in the form projects/-/serviceAccounts/{ACCOUNT_EMAIL_OR_UNIQUEID}. The - wildcard is required; a project ID in its place is invalid. |
Every setting listed for the chosen provider is required; without one, Core does not start. ElastiCache supports IAM authentication for Valkey 7.2 and later and for Redis OSS 7.0 and later, and only with in-transit encryption, so use rediss:// addresses.
{
"redis": {
"clusterServersConfig": {
"nodeAddresses": ["rediss://dial-cache.example.cache.amazonaws.com:6379"]
},
"provider": {
"name": "aws-elasti-cache",
"userId": "dial-core",
"region": "eu-west-1",
"clusterName": "dial-cache",
"serverless": false
}
}
}
Memory and eviction
Valkey can run as a cache with a volatile-* eviction policy, for example:
maxmemory 4G
maxmemory-policy volatile-lfu
Some data lives only in Valkey and has a TTL, so these policies can evict it under memory pressure:
- Per-request API keys (
perRequestApiKey.ttl). - Per-trace usage records (
deploymentStats.ttlMs). An evicted record loses its trace's roll-up:aggCostin the analytics log,statistics.usage_per_model, and the aggregated costs the limits API reports. Limit enforcement is not affected.
Size maxmemory so that eviction stays rare.
Valkey can use more memory than maxmemory, so set the container memory limit higher to avoid out-of-memory kills. The DIAL Helm chart pairs maxmemory 2G (2×10⁹ bytes) with a 2Gi (2³¹ bytes) limit; change both together.
Further reading
- Resources — what Core caches in Valkey and when it syncs to blob storage
- Storage — blob storage settings and provider requirements
- DIAL Core chart configuration — the bundled Valkey and its chart values
- High availability — run a managed, replicated cache for production