Enabling in-transit encryption on a self-designed Redis OSS Cluster using Python

The following guide will demonstrate how to enable in-transit encryption on a Redis OSS 7.0 cluster that was originally created with in-transit encryption disabled. TCP and TLS clients will continue communicating with the cluster during this process without downtime.

Boto3 will get the credentials it needs (aws_access_key_id, aws_secret_access_key, and aws_session_token) from the environment variables. Those credentials will be pasted in advance in the same bash terminal where we will run python3 to process the Python code shown in this guide. The code in the example below was process from an EC2 instance that was launched in the same VPC that will be used to create the ElastiCache Redis OSS Cluster in it.

Define the string constants that will launch the ElastiCache Valkey or Redis OSS Cluster

First, let’s define some simple Python string constants that will hold the names of the AWS entities required to create the ElastiCache cluster such as security-group, Cache Subnet group, and a default parameter group. All of these AWS entities must be created in advance in your AWS account in the Region you are willing to use.

#Constants definitions 
SECURITY_GROUP = "sg-0492aa0a29c558427"
CLUSTER_DESCRIPTION = "This cluster has been launched as part of the online TLS migration user guide"
EC_SUBNET_GROUP = "client-testing"
DEFAULT_PARAMETER_GROUP_REDIS_7_CLUSTER_MODE_ENABLED = "default.redis7.cluster.on"

Define the classes for the cluster configuration

Now, let’s define some simple Python classes that will represent a configuration of a cluster, which will hold metadata about the cluster such as the Valkey or Redis OSS version, the instance type, and whether in-transit encryption (TLS) is enabled or disabled.

#Class definitions

class Config:
    def __init__(
        self,
        instance_type: str = "cache.t4g.small",
        version: str = "7.0",
        multi_az: bool = True,
        TLS: bool = True,
        name: str = None,
    ):
        self.instance_type = instance_type
        self.version = version
        self.multi_az = multi_az
        self.TLS = TLS
        self.name = name or f"tls-test"

    def create_base_launch_request(self):
        return {
            "ReplicationGroupId": self.name,
            "TransitEncryptionEnabled": self.TLS,
            "MultiAZEnabled": self.multi_az,
            "CacheNodeType": self.instance_type,
            "Engine": "redis",
            "EngineVersion": self.version,
            "CacheSubnetGroupName": EC_SUBNET_GROUP ,
            "CacheParameterGroupName": DEFAULT_PARAMETER_GROUP_REDIS_7_CLUSTER_MODE_ENABLED ,
            "ReplicationGroupDescription": CLUSTER_DESCRIPTION,
            "SecurityGroupIds": [SECURITY_GROUP],
        }
        
class ConfigCME(Config):
    def __init__(
        self,
        instance_type: str = "cache.t4g.small",
        version: str = "7.0",
        multi_az: bool = True,
        TLS: bool = True,
        name: str = None,
        num_shards: int = 2,
        num_replicas_per_shard: int = 1,
    ):
        super().__init__(instance_type, version, multi_az, TLS, name)
        self.num_shards = num_shards
        self.num_replicas_per_shard = num_replicas_per_shard

    def create_launch_request(self) -> dict:
        launch_request = self.create_base_launch_request()
        launch_request["NumNodeGroups"] = self.num_shards
        launch_request["ReplicasPerNodeGroup"] = self.num_replicas_per_shard
        return launch_request

Define a class that will represent the cluster itself

Now, let’s define some simple Python classes that will represent the ElastiCache Valkey or Redis OSS Cluster itself. This class will have a client field which will hold a boto3 client for ElastiCache management operations such as creating the cluster and querying the ElastiCache API.

import botocore.config
import boto3

# Create boto3 client
def init_client(region: str = "us-east-1"):
    config = botocore.config.Config(retries={"max_attempts": 10, "mode": "standard"})
    init_request = dict()
    init_request["config"] = config
    init_request["service_name"] = "elasticache"
    init_request["region_name"] = region
    return boto3.client(**init_request) 
 
 
class ElastiCacheClusterBase:
    def __init__(self, name: str):
        self.name = name
        self.elasticache_client = init_client()

    def get_first_replication_group(self):
        return self.elasticache_client.describe_replication_groups(
        ReplicationGroupId=self.name
        )["ReplicationGroups"][0]
 
    def get_status(self) -> str:
        return self.get_first_replication_group()["Status"]
 
    def get_transit_encryption_enabled(self) -> bool:
        return self.get_first_replication_group()["TransitEncryptionEnabled"]
 
    def is_available(self) -> bool:
        return self.get_status() == "available"
        
    def is_modifying(self) -> bool:
        return self.get_status() == "modifying"
        
    def wait_for_available(self):
        while True:
            if self.is_available():
                break
            else:
                time.sleep(5)

    def wait_for_modifying(self):
        while True:
            if self.is_modifying():
                break
            else:
                time.sleep(5)
                
    def delete_cluster(self) -> bool:
        self.elasticache_client.delete_replication_group(
            ReplicationGroupId=self.name, RetainPrimaryCluster=False
        )
        
    def modify_transit_encryption_mode(self, new_transit_encryption_mode: str):
        # generate api call to migrate the cluster to TLS preffered or to TLS required
            self.elasticache_client.modify_replication_group(
                ReplicationGroupId=self.name,
                TransitEncryptionMode=new_transit_encryption_mode,
                TransitEncryptionEnabled=True,
                ApplyImmediately=True,
            )  
        self.wait_for_modifying()
              
 class ElastiCacheClusterCME(ElastiCacheClusterBase):
    def __init__(self, name: str):
        super().__init__(name)

    @classmethod
    def launch(cls, config: ConfigCME = None) -> ElastiCacheClusterCME:
        config = config or ConfigCME()
        print(config)
        new_cluster = ElastiCacheClusterCME(config.name)
        launch_request = config.create_launch_request()
        new_cluster.elasticache_client.create_replication_group(**launch_request)
        new_cluster.wait_for_available()
        return new_cluster

    def get_configuration_endpoint(self) -> str:
        return self.get_first_replication_group()["ConfigurationEndpoint"]["Address"]
     
#Since the code can throw exceptions, we define this class to make the code more readable and 
#so we won't forget to delete the cluster    
class ElastiCacheCMEManager:
    def __init__(self, config: ConfigCME = None):
        self.config = config or ConfigCME()

    def __enter__(self) -> ElastiCacheClusterCME:
        self.cluster = ElastiCacheClusterCME.launch(self.config)
        return self.cluster 
          
    def __exit__(self, exc_type, exc_val, exc_tb):
        self.cluster.delete_cluster()

(Optional) Create a wrapper class to demo client connection to Valkey or Redis OSS cluster

Now, let’s create a wrapper class for the redis-py-cluster client. This wrapper class will support pre-filling the cluster with some keys and then doing random repeated get commands.

import redis
improt random
from time import perf_counter_ns, time


class DowntimeTestClient:
    def __init__(self, client):
        self.client = client

        # num of keys prefilled
        self.prefilled = 0
        # percent of get above prefilled
        self.percent_get_above_prefilled = 10 # nil result expected when get hit above prefilled
        # total downtime in nano seconds 
        self.downtime_ns = 0
        # num of success and fail operations
        self.success_ops = 0
        self.fail_ops = 0
        self.connection_errors = 0
        self.timeout_errors = 0
        

    def replace_client(self, client):
        self.client = client

    def prefill_data(self, timelimit_sec=60):
        end_time = time() + timelimit_sec
        while time() < end_time:
            self.client.set(self.prefilled, self.prefilled)
            self.prefilled += 1

    # unsuccesful operations throw exceptions
    def _exec(self, func):
        try:
            start_ns = perf_counter_ns()
            func()
            self.success_ops += 1
            elapsed_ms = (perf_counter_ns() - start_ns) // 10 ** 6
            # upon succesful execution of func
            # reset random_key to None so that the next command
            # will use a new random key
            self.random_key = None

        except Exception as e:
            elapsed_ns = perf_counter_ns() - start_ns
            self.downtime_ns += elapsed_ns
            # in case of failure- increment the relevant counters so that we will keep track 
            # of how many connection issues we had while trying to communicate with
            # the cluster.
            self.fail_ops += 1
            if e.__class__ is redis.exceptions.ConnectionError:
                self.connection_errors += 1
            if e.__class__ is redis.exceptions.TimeoutError:
                self.timeout_errors += 1

    def _repeat_exec(self, func, seconds):
        end_time = time() + seconds
        while time() < end_time:
            self._exec(func)

    def _new_random_key_if_needed(self, percent_above_prefilled):
        if self.random_key is None:
            max = int((self.prefilled * (100 + percent_above_prefilled)) / 100)
            return random.randint(0, max)
        return self.random_key

    def _random_get(self):
        key = self._new_random_key_if_needed(self.percent_get_above_prefilled)
        result = self.client.get(key)
        # we know the key was set for sure only in the case key < self.prefilled
        if key < self.prefilled:
            assert result.decode("UTF-8") == str(key)


    def repeat_get(self, seconds=60):
        self._repeat_exec(self._random_get, seconds)

    def get_downtime_ms(self) -> int:
        return self.downtime_ns // 10 ** 6


    def do_get_until(self, cond_check):
        while not cond_check():
            self.repeat_get()
        # do one more get cycle once condition is met
        self.repeat_get()

Create the main function that demos the process of changing in-transit encryption configuration

Now, let’s define the main function, which will do the following:

import redis

def init_cluster_client(
    cluster: ElastiCacheClusterCME, prefill_data: bool, TLS: bool = True) -> DowntimeTestClient:
    # we must use for the host name the cluster configuration endpoint. 
    redis_client = redis.RedisCluster(
        host=cluster.get_configuration_endpoint(), ssl=TLS, socket_timeout=0.25, socket_connect_timeout=0.1
    )
    test_client = DowntimeTestClient(redis_client)
    if prefill_data:
        test_client.prefill_data()
    return test_client

if __name__ == '__main__':
    config = ConfigCME(TLS=False, instance_type="cache.m5.large")

    with ElastiCacheCMEManager(config) as cluster:
        # create a client that will connect to the cluster with clear tcp connection
        test_client_tcp = init_cluster_client(cluster, prefill_data=True, TLS=False)
        
       # migrate the cluster to TLS Preferred
        cluster.modify_transit_encryption_mode(new_transit_encryption_mode="preferred")
        
        # do repeated get commands until the cluster finishes the migration to TLS Preferred
        test_client_tcp.do_get_until(cluster.is_available)
        
       # verify that in transit encryption is enabled so that clients will be able to connect to the cluster with TLS
        assert cluster.get_transit_encryption_enabled() == True
        
       # create a client that will connect to the cluster with TLS connection. 
        # we must first make sure that the cluster indeed supports TLS
        test_client_tls = init_cluster_client(cluster, prefill_data=True, TLS=True)
        
        # by doing get commands with the tcp client for 60 more seconds
       # we can verify that the existing tcp connection to the cluster still works 
        test_client_tcp.repeat_get(seconds=60)
        
        # do get commands with the new TLS client for 60 more seconds
        test_client_tcp.repeat_get(seconds=60)
        
       # migrate the cluster to TLS required
        cluster.modify_transit_encryption_mode(new_transit_encryption_mode="required")
        
       # from this point the tcp clients will be disconnected and we must not use them anymore.
       # do get commands with the TLS client until the cluster finishes migartion to TLS required mode.
        test_client_tls.do_get_until(cluster.is_available)