Option to read Redis URL from AWS Secret

charts/model-engine/values_sample.yaml

@@ -156,18 +156,37 @@ config:
       ml_account_id: "000000000000"
       # docker_repo_prefix [required] is the prefix for AWS ECR repositories
       docker_repo_prefix: "000000000000.dkr.ecr.us-east-1.amazonaws.com"
-      # redis_host [required] is the hostname of the redis cluster you wish to connect
+      # redis_host [required if redis_aws_secret_name not present] is the hostname of the redis cluster you wish to connect
       redis_host: llm-engine-prod-cache.use1.cache.amazonaws.com
+      # redis_aws_secret_name [optional] is the AWS secret that contains the connection info of the Redis cluster.
+      # The information provided should be as follows: 
+      # scheme: either redis:// or rediss://, will default to redis://
+      # auth_token (optional): an auth token for the Redis cluster
+      # host: the hostname of the Redis cluster
+      # port: the port of the Redis cluster
+      # query_params (optional): additional query parameters for the Redis cluster, will default to ""
+      # The url will be built as follows:
+      # {scheme}{host}:{port}/{db_index}{query_params} if auth_token is not provided,
+      # {scheme}:{auth_token}@{host}:{port}/{db_index}{query_params} if auth_token is provided
+      # db_index will be filled in by LLM Engine.
+      # This secret must be accessible by the default LLM Engine AWS role
+      # e.g. what is set by profile_ml_worker if provided
+      # redis_aws_secret_name: sample-prod/redis-credentials
       # s3_bucket [required] is the S3 bucket you wish to connect 
       s3_bucket: "llm-engine"
     launch:
       # endpoint_namespace [required] is K8s namespace the endpoints will be created in
       endpoint_namespace: llm-engine
       # cache_redis_aws_url is the full url for the redis cluster you wish to connect,
       # cache_redis_azure_host is the redis cluster host when using cloud_provider azure
-      # one of cache_redis_aws_url and cache_redis_azure_host must be provided
+      # cache_redis_aws_secret_name is an AWS secret that contains the Redis credentials.
+      # It has a field "cache-url" with the full URL of the Redis cluster (including db number).
+      # Other fields are ignored; e.g. you can use the secret for multiple purposes.
+      # This secret must be accessible by the default LLM Engine AWS role
+      # exactly one of cache_redis_aws_url, cache_redis_azure_host, or cache_redis_aws_secret_name must be provided
       cache_redis_aws_url: redis://llm-engine-prod-cache.use1.cache.amazonaws.com:6379/15
       cache_redis_azure_host: llm-engine-cache.redis.cache.windows.net:6380
+      cache_redis_aws_secret_name: sample-prod/redis-credentials
       # s3_file_llm_fine_tuning_job_repository [required] is the S3 URI for the S3 bucket/key that you wish to save fine-tuned assests
       s3_file_llm_fine_tuning_job_repository: "s3://llm-engine/llm-ft-job-repository"
       # dd_trace_enabled specifies whether to enable datadog tracing, datadog must be installed in the cluster

model-engine/model_engine_server/common/config.py

@@ -8,6 +8,7 @@
 
 import yaml
 from azure.identity import DefaultAzureCredential
+from model_engine_server.core.aws.secrets import get_key_file
 from model_engine_server.core.config import infra_config
 from model_engine_server.core.loggers import logger_name, make_logger
 
@@ -68,8 +69,12 @@ class HostedModelInferenceServiceConfig:
     user_inference_tensorflow_repository: str
     docker_image_layer_cache_repository: str
     sensitive_log_mode: bool
+    # Exactly one of the following three must be specified
     cache_redis_aws_url: Optional[str] = None  # also using this to store sync autoscaling metrics
     cache_redis_azure_host: Optional[str] = None
+    cache_redis_aws_secret_name: Optional[
+        str
+    ] = None  # Not an env var because the redis cache info is already here
 
     @classmethod
     def from_yaml(cls, yaml_path):
@@ -80,7 +85,18 @@ def from_yaml(cls, yaml_path):
     @property
     def cache_redis_url(self) -> str:
         if self.cache_redis_aws_url:
+            assert infra_config().cloud_provider == "aws", "cache_redis_aws_url is only for AWS"
+            if self.cache_redis_aws_secret_name:
+                logger.warning(
+                    "Both cache_redis_aws_url and cache_redis_aws_secret_name are set. Using cache_redis_aws_url"
+                )
             return self.cache_redis_aws_url
+        elif self.cache_redis_aws_secret_name:
+            assert (
+                infra_config().cloud_provider == "aws"
+            ), "cache_redis_aws_secret_name is only for AWS"
+            creds = get_key_file(self.cache_redis_aws_secret_name)  # Use default role
+            return creds["cache-url"]
 
         assert self.cache_redis_azure_host and infra_config().cloud_provider == "azure"
         username = os.getenv("AZURE_OBJECT_ID")

model-engine/model_engine_server/core/celery/app.py

@@ -10,6 +10,7 @@
 from celery.app.control import Inspect
 from celery.result import AsyncResult
 from model_engine_server.core.aws.roles import session
+from model_engine_server.core.aws.secrets import get_key_file
 from model_engine_server.core.config import infra_config
 from model_engine_server.core.loggers import (
     CustomJSONFormatter,
@@ -195,6 +196,17 @@ def get_redis_host_port():
 
 
 def get_redis_endpoint(db_index: int = 0) -> str:
+    if infra_config().redis_aws_secret_name is not None:
+        logger.info("Using infra_config().redis_aws_secret_name for Redis endpoint")
+        creds = get_key_file(infra_config().redis_aws_secret_name)  # Use default role
+        scheme = creds.get("scheme", "redis://")
+        host = creds["host"]
+        port = creds["port"]
+        query_params = creds.get("query_params", "")
+        auth_token = creds.get("auth_token", None)
+        if auth_token is not None:
+            return f"{scheme}:{auth_token}@{host}:{port}/{db_index}{query_params}"
+        return f"{scheme}{host}:{port}/{db_index}{query_params}"
     host, port = get_redis_host_port()
     auth_token = os.getenv("REDIS_AUTH_TOKEN")
     if auth_token:

model-engine/model_engine_server/core/config.py

@@ -38,8 +38,9 @@ class InfraConfig:
     default_region: str
     ml_account_id: str
     docker_repo_prefix: str
-    redis_host: str
     s3_bucket: str
+    redis_host: Optional[str] = None
+    redis_aws_secret_name: Optional[str] = None
     profile_ml_worker: str = "default"
     profile_ml_inference_worker: str = "default"
     identity_service_url: Optional[str] = None