By default, the Server SDK stores entitlements data in a local in-memory cache for fast and instant entitlement checks. If you restart the host process of the SDK or re-initialize the SDK, the local cache data will be lost. Usually, this behavior is acceptable because if the entitlements data is missing, the SDK will fetch it from the Stigg API over the network. Alternatively, you can provide an external store to persist the customer entitlements data which can survive restarts and be accessed by multiple processes. If your application is a fleet consisting of dozens of servers, or if you have a serverless infrastructure where each process is transient and can be de-provisioned after a limited period of time, a persistent cache can significantly reduce cache misses. Stigg offers a persistent cache setup for backend SDK users (currently supported in the Node SDK and Sidecar), providing several advantages over in-memory cache for high-scale services:
  • Redundancy - data is available on your Redis cluster even in case of an API outage
  • Low latency
  • Fewer requests from your services to the Stigg API
  • Reduce the memory footprint of the SDK in your services
  • Reducing cache coherence issues
  • Increased cache hits/misses ratio
To enable persistent caching in your environment, please contact the Stigg Support Team.

​
Overview

Stigg’s persistent cache relies on a running persistent-cache-service as part of your app deployment process. The service connects to a dedicated AWS SQS queue (owned and managed by Stigg), consumes messages that carry the state changes from the originating environment, and keeps your Redis cache instance with up-to-date data. The service can be scaled horizontally to keep up with the rate of arriving messages. In case of a cache miss, the SDK will fetch the data over the network directly from API, and update the cache to serve future requests.

​
Running the service

In order for the persistent cache to be updated properly in near real-time, you’ll need to install and run at least one instance of persistent-cache-service.

​
Prerequisites

  • An SQS provisioned by Stigg, to consume messages originating from your environment
  • Redis cluster to persist entitlements and usage for a future read operation

​
Usage

Run the service: 
docker run --publish=4000:4000 \
  --name stigg-persistent-cache \
  -e AWS_REGION="us-east-2" \
  -e QUEUE_URL="https://sqs.us-east-2.amazonaws.com/..." \
  -e REDIS_HOST="host.docker.internal" \
  -e SERVER_API_KEY="<SERVER_API_KEY>" \
  -e ENVIRONMENT_PREFIX="development" \
  public.ecr.aws/stigg/persistent-cache-service:latest
Available options:
Environment VariableTypeDefaultDescription
AWS_REGIONString*The AWS region of the SQS queue
QUEUE_URLString*The URL of the queue to consume from
SERVER_API_KEYString*Environment’s Server API key
ENVIRONMENT_PREFIXStringproductionThe Identifier of the environment needs to be the same as the one used by the SDK
REDIS_HOSTStringRedis host address
REDIS_PORTNumber6379Redis port
REDIS_DBNumber0Redis DB identifier
REDIS_USERString(Optional) Redis username
REDIS_PASSWORDString(Optional) Redis password
REDIS_TLSBoolean0(Optional) Redis use TLS encryption
BATCH_SIZENumber1Number of messages to receive in a single batch
KEYS_TTL_IN_SECSNumber604800 (7 days)The duration in milliseconds that data will be kept in the cache before evicted
EDGE_API_URLStringhttps://edge.api.stigg.ioEdge API address URL
SENTRY_DSNStringLoaded from remoteSentry DSN used for internal error reporting
DISABLE_SENTRYBoolean0Disables reporting internal errors to Sentry
PORTNumber4000HTTP server port
*Required fields

​
Health

The service exposes the following HTTP endpoints:

​
GET /livez

Returns 200 if the service is alive. Healthy response: {"status":"UP"}

​
GET /readyz

Returns 200 if the service is ready to receive messages. Healthy response: { "status": "UP", "consumers": 10, "redis": "CONNECTED" } Unhealthy response: { "status": "DOWN", "consumers": 0, "redis": "DISCONNECTED" }

​
GET /metrics

Returns service metrics in Prometheus format. This endpoint includes both system-level and cache specific metrics. These metrics are helpful for monitoring the health and performance of the service.
Metric NameTypeDescription
persistent_cache_write_duration_secondshistogramTime taken to write to Redis cache
persistent_cache_write_errors_totalcounterNumber of failed Redis cache write operations
persistent_cache_hit_ratiogaugePercentage of Redis cache hits vs total cache queries
persistent_cache_memory_usage_bytesgaugeAmount of memory used by the Redis cache in bytes
persistent_cache_hits_totalgaugeTotal number of cache hits in Redis
persistent_cache_misses_totalgaugeTotal number of cache misses in Redis
persistent_cache_message_handling_duration_secondshistogramTime taken to process a message from receipt to completion
persistent_cache_messages_processed_totalcounterTotal number of messages processed

​
Configure the SDK

To configure the SDK to use the Redis cache, provide the redis config during initialization: 
const stigg = Stigg.initialize({
    apiKey: '<SERVER_API_KEY>',
    redis: {
        host: 'localhost', 
        port: '6379',
        db: 0, // optionally, the db number to use
        environmentPrefix: 'production', // a string, for example 'production', 'staging' & etc
    }
});
Please make sure that the SDK and the persistent-cache-service are configured to use the same Redis instance, DB number, and the same environment prefix.
During initialization, the SDK will try to connect to the Redis instance. Once connected, all the entitlement checks will be evaluated against data from the persistent cache instead of fetching it over the network. The persistent cache is updated the same way the in-memory cache does - if customer data is missing, it will be fetched over the network and persisted in the cache.

​
Benchmarks

Based on internal benchmarks, a single instance of the persistent-cache-service runs efficiently with the following configuration:
  • CPU and memory: 1 vCPU and 512MB RAM.
  • CPU usage: Averages around 60%, with a range between 30% and 90% under constant load (i.e., when messages are consistently present in the queue).
  • Memory footprint: Approximately 250MB.
  • Throughput: Handles up to 100 messages per second, depending on message type and processing complexity.
  • Storage: Does not use any disk storage — the minimal container image size is sufficient.
  • Scalability: The service is stateless and can be horizontally scaled as needed to meet higher throughput demands.