本文為英文版的機器翻譯版本,如內容有任何歧義或不一致之處,概以英文版為準。
Python 和 ElastiCache
在本教學課程中,您可以使用 for Python AWS SDK(Boto3) 編寫簡單的程式來執行下列 ElastiCache (RedisOSS) 操作:
建立 ElastiCache (Redis OSS) 叢集 (啟用叢集模式和停用叢集模式)
檢查使用者或使用者群組是否存在,否則請建立它們。(此功能適用於 Valkey 7.2 及更新版本,以及 Redis OSS 6.0 及更新版本。)
連線至 ElastiCache
執行各種作業,例如設定和取得字串、讀取和寫入串流,以及從發佈/訂閱頻道發佈和訂閱。
完成本教學課程時,您可以參考 for Python AWS SDK(Boto) 文件。以下章節是 特有的 ElastiCache:ElastiCache 低階用戶端
教學課程事前準備
設定 AWS 存取金鑰以使用 AWS SDKs。如需詳細資訊,請參閱設定 ElastiCache。
安裝 Python 3.0 或更新版本。如需詳細資訊,請參閱 https://https://www.python.org/downloads
。如需說明,請參閱 Boto 3 文件中的快速入門 。
教學課程:建立 ElastiCache 叢集和使用者
下列範例使用 boto3 SDK進行 ElastiCache (Redis OSS) 管理操作 (叢集或使用者建立) 和重新解散/redis-py-cluster 進行資料處理。
建立停用叢集模式的叢集
複製下列程式,並將其貼到名為 CreateClusterModeDisabledCluster.py 的檔案中。
import boto3 import logging logging.basicConfig(level=logging.INFO) client = boto3.client('elasticache') def create_cluster_mode_disabled(CacheNodeType='cache.t3.small',EngineVersion='6.0',NumCacheClusters=2,ReplicationGroupDescription='Sample cache cluster',ReplicationGroupId=None): """Creates an ElastiCache Cluster with cluster mode disabled Returns a dictionary with the API response :param CacheNodeType: Node type used on the cluster. If not specified, cache.t3.small will be used Refer to https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/CacheNodes.SupportedTypes.html for supported node types :param EngineVersion: Engine version to be used. If not specified, latest will be used. :param NumCacheClusters: Number of nodes in the cluster. Minimum 1 (just a primary node) and maximun 6 (1 primary and 5 replicas). If not specified, cluster will be created with 1 primary and 1 replica. :param ReplicationGroupDescription: Description for the cluster. :param ReplicationGroupId: Name for the cluster :return: dictionary with the API results """ if not ReplicationGroupId: return 'ReplicationGroupId parameter is required' response = client.create_replication_group( AutomaticFailoverEnabled=True, CacheNodeType=CacheNodeType, Engine='valkey', EngineVersion=EngineVersion, NumCacheClusters=NumCacheClusters, ReplicationGroupDescription=ReplicationGroupDescription, ReplicationGroupId=ReplicationGroupId, SnapshotRetentionLimit=30, ) return response if __name__ == '__main__': # Creates an ElastiCache Cluster mode disabled cluster, based on cache.m6g.large nodes, Valkey 7.2, one primary and two replicas elasticacheResponse = create_cluster_mode_disabled( #CacheNodeType='cache.m6g.large', EngineVersion='7.2', NumCacheClusters=3, ReplicationGroupDescription='Valkey cluster mode disabled with replicas', ReplicationGroupId='valkey202104053' ) logging.info(elasticacheResponse)
若要執行程式,請輸入下列命令:
python CreateClusterModeDisabledCluster.py
如需詳細資訊,請參閱在 中管理叢集 ElastiCache。
使用 TLS和 建立停用叢集模式的叢集 RBAC
為了確保安全,您可以在建立停用叢集模式的叢集時,使用 Transport Layer Security (TLS) 和角色型存取控制 (RBAC)。與 Valkey 或 Redis OSS 不同AUTH,在驗證所有已驗證的用戶端的權杖時,其具有完整複寫群組存取權,RBAC可讓您透過使用者群組控制叢集存取權。這些使用者群組的設計目的是整理複寫群組的存取權。如需詳細資訊,請參閱角色型存取控制 (RBAC)。
複製下列程式,並將其貼到名為 ClusterModeDisabledWithRBAC.py 的檔案中。
import boto3 import logging logging.basicConfig(level=logging.INFO) client = boto3.client('elasticache') def create_cluster_mode_disabled_rbac(CacheNodeType='cache.t3.small',EngineVersion='6.0',NumCacheClusters=2,ReplicationGroupDescription='Sample cache cluster',ReplicationGroupId=None, UserGroupIds=None, SecurityGroupIds=None,CacheSubnetGroupName=None): """Creates an ElastiCache Cluster with cluster mode disabled and RBAC Returns a dictionary with the API response :param CacheNodeType: Node type used on the cluster. If not specified, cache.t3.small will be used Refer to https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/CacheNodes.SupportedTypes.html for supported node types :param EngineVersion: Engine version to be used. If not specified, latest will be used. :param NumCacheClusters: Number of nodes in the cluster. Minimum 1 (just a primary node) and maximun 6 (1 primary and 5 replicas). If not specified, cluster will be created with 1 primary and 1 replica. :param ReplicationGroupDescription: Description for the cluster. :param ReplicationGroupId: Mandatory name for the cluster. :param UserGroupIds: The ID of the user group to be assigned to the cluster. :param SecurityGroupIds: List of security groups to be assigned. If not defined, default will be used :param CacheSubnetGroupName: subnet group where the cluster will be placed. If not defined, default will be used. :return: dictionary with the API results """ if not ReplicationGroupId: return {'Error': 'ReplicationGroupId parameter is required'} elif not isinstance(UserGroupIds,(list)): return {'Error': 'UserGroupIds parameter is required and must be a list'} params={'AutomaticFailoverEnabled': True, 'CacheNodeType': CacheNodeType, 'Engine': 'valkey', 'EngineVersion': EngineVersion, 'NumCacheClusters': NumCacheClusters, 'ReplicationGroupDescription': ReplicationGroupDescription, 'ReplicationGroupId': ReplicationGroupId, 'SnapshotRetentionLimit': 30, 'TransitEncryptionEnabled': True, 'UserGroupIds':UserGroupIds } # defaults will be used if CacheSubnetGroupName or SecurityGroups are not explicit. if isinstance(SecurityGroupIds,(list)): params.update({'SecurityGroupIds':SecurityGroupIds}) if CacheSubnetGroupName: params.update({'CacheSubnetGroupName':CacheSubnetGroupName}) response = client.create_replication_group(**params) return response if __name__ == '__main__': # Creates an ElastiCache Cluster mode disabled cluster, based on cache.m6g.large nodes, Valkey 7.2, one primary and two replicas. # Assigns the existent user group "mygroup" for RBAC authentication response=create_cluster_mode_disabled_rbac( CacheNodeType='cache.m6g.large', EngineVersion='7.2', NumCacheClusters=3, ReplicationGroupDescription='Valkey cluster mode disabled with replicas', ReplicationGroupId='valkey202104', UserGroupIds=[ 'mygroup' ], SecurityGroupIds=[ 'sg-7cc73803' ], CacheSubnetGroupName='default' ) logging.info(response)
若要執行程式,請輸入下列命令:
python ClusterModeDisabledWithRBAC.py
如需詳細資訊,請參閱在 中管理叢集 ElastiCache。
建立啟用叢集模式的叢集
複製下列程式,並將其貼到名為 ClusterModeEnabled.py 的檔案中。
import boto3 import logging logging.basicConfig(level=logging.INFO) client = boto3.client('elasticache') def create_cluster_mode_enabled(CacheNodeType='cache.t3.small',EngineVersion='6.0',NumNodeGroups=1,ReplicasPerNodeGroup=1, ReplicationGroupDescription='Sample cache with cluster mode enabled',ReplicationGroupId=None): """Creates an ElastiCache Cluster with cluster mode enabled Returns a dictionary with the API response :param CacheNodeType: Node type used on the cluster. If not specified, cache.t3.small will be used Refer to https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/CacheNodes.SupportedTypes.html for supported node types :param EngineVersion: Engine version to be used. If not specified, latest will be used. :param NumNodeGroups: Number of shards in the cluster. Minimum 1 and maximun 90. If not specified, cluster will be created with 1 shard. :param ReplicasPerNodeGroup: Number of replicas per shard. If not specified 1 replica per shard will be created. :param ReplicationGroupDescription: Description for the cluster. :param ReplicationGroupId: Name for the cluster :return: dictionary with the API results """ if not ReplicationGroupId: return 'ReplicationGroupId parameter is required' response = client.create_replication_group( AutomaticFailoverEnabled=True, CacheNodeType=CacheNodeType, Engine='valkey', EngineVersion=EngineVersion, ReplicationGroupDescription=ReplicationGroupDescription, ReplicationGroupId=ReplicationGroupId, # Creates a cluster mode enabled cluster with 1 shard(NumNodeGroups), 1 primary node (implicit) and 2 replicas (replicasPerNodeGroup) NumNodeGroups=NumNodeGroups, ReplicasPerNodeGroup=ReplicasPerNodeGroup, CacheParameterGroupName='default.valkey7.2.cluster.on' ) return response # Creates a cluster mode enabled response = create_cluster_mode_enabled( CacheNodeType='cache.m6g.large', EngineVersion='6.0', ReplicationGroupDescription='Valkey cluster mode enabled with replicas', ReplicationGroupId='valkey20210', # Creates a cluster mode enabled cluster with 1 shard(NumNodeGroups), 1 primary (implicit) and 2 replicas (replicasPerNodeGroup) NumNodeGroups=2, ReplicasPerNodeGroup=1, ) logging.info(response)
若要執行程式,請輸入下列命令:
python ClusterModeEnabled.py
如需詳細資訊,請參閱在 中管理叢集 ElastiCache。
使用 TLS和 建立啟用叢集模式的叢集 RBAC
為了確保安全,您可以在建立啟用叢集模式的叢集時,使用 Transport Layer Security (TLS) 和角色型存取控制 (RBAC)。與 Valkey 或 Redis OSS 不同AUTH,在驗證所有已驗證的用戶端的權杖時,其具有完整複寫群組存取權,RBAC可讓您透過使用者群組控制叢集存取權。這些使用者群組的設計目的是整理複寫群組的存取權。如需詳細資訊,請參閱角色型存取控制 (RBAC)。
複製下列程式,並將其貼到名為 ClusterModeEnabledWithRBAC.py 的檔案中。
import boto3 import logging logging.basicConfig(level=logging.INFO) client = boto3.client('elasticache') def create_cluster_mode_enabled(CacheNodeType='cache.t3.small',EngineVersion='6.0',NumNodeGroups=1,ReplicasPerNodeGroup=1, ReplicationGroupDescription='Sample cache with cluster mode enabled',ReplicationGroupId=None,UserGroupIds=None, SecurityGroupIds=None,CacheSubnetGroupName=None,CacheParameterGroupName='default.valkey7.2.cluster.on'): """Creates an ElastiCache Cluster with cluster mode enabled and RBAC Returns a dictionary with the API response :param CacheNodeType: Node type used on the cluster. If not specified, cache.t3.small will be used Refer to https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/CacheNodes.SupportedTypes.html for supported node types :param EngineVersion: Engine version to be used. If not specified, latest will be used. :param NumNodeGroups: Number of shards in the cluster. Minimum 1 and maximun 90. If not specified, cluster will be created with 1 shard. :param ReplicasPerNodeGroup: Number of replicas per shard. If not specified 1 replica per shard will be created. :param ReplicationGroupDescription: Description for the cluster. :param ReplicationGroupId: Name for the cluster. :param CacheParameterGroupName: Parameter group to be used. Must be compatible with the engine version and cluster mode enabled. :return: dictionary with the API results """ if not ReplicationGroupId: return 'ReplicationGroupId parameter is required' elif not isinstance(UserGroupIds,(list)): return {'Error': 'UserGroupIds parameter is required and must be a list'} params={'AutomaticFailoverEnabled': True, 'CacheNodeType': CacheNodeType, 'Engine': 'valkey', 'EngineVersion': EngineVersion, 'ReplicationGroupDescription': ReplicationGroupDescription, 'ReplicationGroupId': ReplicationGroupId, 'SnapshotRetentionLimit': 30, 'TransitEncryptionEnabled': True, 'UserGroupIds':UserGroupIds, 'NumNodeGroups': NumNodeGroups, 'ReplicasPerNodeGroup': ReplicasPerNodeGroup, 'CacheParameterGroupName': CacheParameterGroupName } # defaults will be used if CacheSubnetGroupName or SecurityGroups are not explicit. if isinstance(SecurityGroupIds,(list)): params.update({'SecurityGroupIds':SecurityGroupIds}) if CacheSubnetGroupName: params.update({'CacheSubnetGroupName':CacheSubnetGroupName}) response = client.create_replication_group(**params) return response if __name__ == '__main__': # Creates a cluster mode enabled cluster response = create_cluster_mode_enabled( CacheNodeType='cache.m6g.large', EngineVersion='7.2', ReplicationGroupDescription='Valkey cluster mode enabled with replicas', ReplicationGroupId='valkey2021', # Creates a cluster mode enabled cluster with 1 shard(NumNodeGroups), 1 primary (implicit) and 2 replicas (replicasPerNodeGroup) NumNodeGroups=2, ReplicasPerNodeGroup=1, UserGroupIds=[ 'mygroup' ], SecurityGroupIds=[ 'sg-7cc73803' ], CacheSubnetGroupName='default' ) logging.info(response)
若要執行程式,請輸入下列命令:
python ClusterModeEnabledWithRBAC.py
如需詳細資訊,請參閱在 中管理叢集 ElastiCache。
檢查使用者/使用者群組是否存在,若不存在則加以建立
透過 RBAC,您可以使用存取字串建立使用者並為其指派特定許可。您可以將使用者指派給與特定角色 (管理員、人力資源) 一致的使用者群組,然後部署到一或多個 ElastiCache (RedisOSS) 複寫群組。透過這樣做,您可以使用相同的 Valkey 或 Redis OSS複寫群組在用戶端之間建立安全界限,並防止用戶端存取彼此的資料。如需詳細資訊,請參閱角色型存取控制 (RBAC)。
複製下列程式,並將其貼到名為 UserAndUserGroups.py 的檔案中。更新提供憑證的機制。此範例中的憑證顯示為可取代,並且有指派的未宣告項目。避免對憑證進行硬式編碼。
此範例使用具有使用者許可的存取字串。如需存取字串的詳細資訊,請參閱 使用存取字串指定許可。
import boto3 import logging logging.basicConfig(level=logging.INFO) client = boto3.client('elasticache') def check_user_exists(UserId): """Checks if UserId exists Returns True if UserId exists, otherwise False :param UserId: ElastiCache User ID :return: True|False """ try: response = client.describe_users( UserId=UserId, ) if response['Users'][0]['UserId'].lower() == UserId.lower(): return True except Exception as e: if e.response['Error']['Code'] == 'UserNotFound': logging.info(e.response['Error']) return False else: raise def check_group_exists(UserGroupId): """Checks if UserGroupID exists Returns True if Group ID exists, otherwise False :param UserGroupId: ElastiCache User ID :return: True|False """ try: response = client.describe_user_groups( UserGroupId=UserGroupId ) if response['UserGroups'][0]['UserGroupId'].lower() == UserGroupId.lower(): return True except Exception as e: if e.response['Error']['Code'] == 'UserGroupNotFound': logging.info(e.response['Error']) return False else: raise def create_user(UserId=None,UserName=None,Password=None,AccessString=None): """Creates a new user Returns the ARN for the newly created user or the error message :param UserId: ElastiCache user ID. User IDs must be unique :param UserName: ElastiCache user name. ElastiCache allows multiple users with the same name as long as the associated user ID is unique. :param Password: Password for user. Must have at least 16 chars. :param AccessString: Access string with the permissions for the user. :return: user ARN """ try: response = client.create_user( UserId=UserId, UserName=UserName, Engine='Redis', Passwords=[Password], AccessString=AccessString, NoPasswordRequired=False ) return response['ARN'] except Exception as e: logging.info(e.response['Error']) return e.response['Error'] def create_group(UserGroupId=None, UserIds=None): """Creates a new group. A default user is required (mandatory) and should be specified in the UserIds list Return: Group ARN :param UserIds: List with user IDs to be associated with the new group. A default user is required :param UserGroupId: The ID (name) for the group :return: Group ARN """ try: response = client.create_user_group( UserGroupId=UserGroupId, Engine='Redis', UserIds=UserIds ) return response['ARN'] except Exception as e: logging.info(e.response['Error']) if __name__ == '__main__': groupName='mygroup2' userName = 'myuser2' userId=groupName+'-'+userName # Creates a new user if the user ID does not exist. for tmpUserId,tmpUserName in [ (userId,userName), (groupName+'-default','default')]: if not check_user_exists(tmpUserId): response=create_user(UserId=tmpUserId, UserName=
EXAMPLE
,Password=EXAMPLE
,AccessString='on ~* +@all') logging.info(response) # assigns the new user ID to the user group if not check_group_exists(groupName): UserIds = [ userId , groupName+'-default'] response=create_group(UserGroupId=groupName,UserIds=UserIds) logging.info(response)
若要執行程式,請輸入下列命令:
python UserAndUserGroups.py
教學課程:連線至 ElastiCache
下列範例使用 Valkey 或 Redis OSS用戶端連線至 ElastiCache。
連線至停用叢集模式的叢集
複製下列程式,並將其貼到名為 ConnectClusterModeDisabled.py 的檔案中。更新提供憑證的機制。此範例中的憑證顯示為可取代,並且有指派的未宣告項目。避免對憑證進行硬式編碼。
from redis import Redis import logging logging.basicConfig(level=logging.INFO) redis = Redis(host='primary.xxx.yyyyyy.zzz1.cache.amazonaws.com', port=6379, decode_responses=True, ssl=True, username=
example
, password=EXAMPLE
) if redis.ping(): logging.info("Connected to Redis")
若要執行程式,請輸入下列命令:
python ConnectClusterModeDisabled.py
連線至啟用叢集模式的叢集
複製下列程式,並將其貼到名為 ConnectClusterModeEnabled.py 的檔案中。
from rediscluster import RedisCluster import logging logging.basicConfig(level=logging.INFO) redis = RedisCluster(startup_nodes=[{"host": "xxx.yyy.clustercfg.zzz1.cache.amazonaws.com","port": "6379"}], decode_responses=True,skip_full_coverage_check=True) if redis.ping(): logging.info("Connected to Redis")
若要執行程式,請輸入下列命令:
python ConnectClusterModeEnabled.py
使用範例
下列範例使用 boto3 SDK ElastiCache 讓 使用 ElastiCache (Redis OSS)。
設定和取得字串
複製下列程式,並將其貼到名為 SetAndGetStrings.py 的檔案中。
import time import logging logging.basicConfig(level=logging.INFO,format='%(asctime)s: %(message)s') keyName='mykey' currTime=time.ctime(time.time()) # Set the key 'mykey' with the current date and time as value. # The Key will expire and removed from cache in 60 seconds. redis.set(keyName, currTime, ex=60) # Sleep just for better illustration of TTL (expiration) value time.sleep(5) # Retrieve the key value and current TTL keyValue=redis.get(keyName) keyTTL=redis.ttl(keyName) logging.info("Key {} was set at {} and has {} seconds until expired".format(keyName, keyValue, keyTTL))
若要執行程式,請輸入下列命令:
python SetAndGetStrings.py
設定和取得具有多個項目的雜湊
複製下列程式,並將其貼到名為 SetAndGetHash.py 的檔案中。
import logging import time logging.basicConfig(level=logging.INFO,format='%(asctime)s: %(message)s') keyName='mykey' keyValues={'datetime': time.ctime(time.time()), 'epochtime': time.time()} # Set the hash 'mykey' with the current date and time in human readable format (datetime field) and epoch number (epochtime field). redis.hset(keyName, mapping=keyValues) # Set the key to expire and removed from cache in 60 seconds. redis.expire(keyName, 60) # Sleep just for better illustration of TTL (expiration) value time.sleep(5) # Retrieves all the fields and current TTL keyValues=redis.hgetall(keyName) keyTTL=redis.ttl(keyName) logging.info("Key {} was set at {} and has {} seconds until expired".format(keyName, keyValues, keyTTL))
若要執行程式,請輸入下列命令:
python SetAndGetHash.py
從發佈/訂閱頻道發佈 (寫入) 和訂閱 (讀取)
複製下列程式,並將其貼到名為 PubAndSub.py 的檔案中。
import logging import time def handlerFunction(message): """Prints message got from PubSub channel to the log output Return None :param message: message to log """ logging.info(message) logging.basicConfig(level=logging.INFO) redis = Redis(host="redis202104053.tihewd.ng.0001.use1.cache.amazonaws.com", port=6379, decode_responses=True) # Creates the subscriber connection on "mychannel" subscriber = redis.pubsub() subscriber.subscribe(**{'mychannel': handlerFunction}) # Creates a new thread to watch for messages while the main process continues with its routines thread = subscriber.run_in_thread(sleep_time=0.01) # Creates publisher connection on "mychannel" redis.publish('mychannel', 'My message') # Publishes several messages. Subscriber thread will read and print on log. while True: redis.publish('mychannel',time.ctime(time.time())) time.sleep(1)
若要執行程式,請輸入下列命令:
python PubAndSub.py
從串流寫入和讀取
複製下列程式,並將其貼到名為 ReadWriteStream.py 的檔案中。
from redis import Redis import redis.exceptions as exceptions import logging import time import threading logging.basicConfig(level=logging.INFO) def writeMessage(streamName): """Starts a loop writting the current time and thread name to 'streamName' :param streamName: Stream (key) name to write messages. """ fieldsDict={'writerId':threading.currentThread().getName(),'myvalue':None} while True: fieldsDict['myvalue'] = time.ctime(time.time()) redis.xadd(streamName,fieldsDict) time.sleep(1) def readMessage(groupName=None,streamName=None): """Starts a loop reading from 'streamName' Multiple threads will read from the same stream consumer group. Consumer group is used to coordinate data distribution. Once a thread acknowleges the message, it won't be provided again. If message wasn't acknowledged, it can be served to another thread. :param groupName: stream group were multiple threads will read. :param streamName: Stream (key) name where messages will be read. """ readerID=threading.currentThread().getName() while True: try: # Check if the stream has any message if redis.xlen(streamName)>0: # Check if if the messages are new (not acknowledged) or not (already processed) streamData=redis.xreadgroup(groupName,readerID,{streamName:'>'},count=1) if len(streamData) > 0: msgId,message = streamData[0][1][0] logging.info("{}: Got {} from ID {}".format(readerID,message,msgId)) #Do some processing here. If the message has been processed sucessfuly, acknowledge it and (optional) delete the message. redis.xack(streamName,groupName,msgId) logging.info("Stream message ID {} read and processed successfuly by {}".format(msgId,readerID)) redis.xdel(streamName,msgId) else: pass except: raise time.sleep(0.5) # Creates the stream 'mystream' and consumer group 'myworkergroup' where multiple threads will write/read. try: redis.xgroup_create('mystream','myworkergroup',mkstream=True) except exceptions.ResponseError as e: logging.info("Consumer group already exists. Will continue despite the error: {}".format(e)) except: raise # Starts 5 writer threads. for writer_no in range(5): writerThread = threading.Thread(target=writeMessage, name='writer-'+str(writer_no), args=('mystream',),daemon=True) writerThread.start() # Starts 10 reader threads for reader_no in range(10): readerThread = threading.Thread(target=readMessage, name='reader-'+str(reader_no), args=('myworkergroup','mystream',),daemon=True) readerThread.daemon = True readerThread.start() # Keep the code running for 30 seconds time.sleep(30)
若要執行程式,請輸入下列命令:
python ReadWriteStream.py