Amazon ElastiCache FAQs

ElastiCache Serverless is a serverless option that lets you get started with a cache in under a minute without infrastructure provisioning or capacity planning. ElastiCache Serverless removes the need for time-consuming capacity planning by continuously monitoring a cache’s compute, memory, and network utilization, so it can instantly scale to meet demand without downtime or performance degradation. ElastiCache Serverless automatically replicates data across multiple Availability Zones (AZs) and provides customers with a 99.99% availability service level agreement (SLA) for each cache. With ElastiCache Serverless, you only pay for the data you store and the compute resources your application uses. To get started, create an ElastiCache Serverless cache in just a few steps by specifying a cache name using the Console, ElastiCache Development Kit (SDK), or AWS Command Line Interface (AWS CLI).

You can move an existing ElastiCache workload by changing the Valkey, Memcached, or Redis OSS endpoint to your new ElastiCache Serverless cache endpoint in your application. You can migrate existing ElastiCache data to ElastiCache Serverless by specifying the Amazon Simple Storage Service (Amazon S3) location of a backup file. Visit our ElastiCache Serverless documentation to learn more about migrating your workloads.

ElastiCache Serverless supports Valkey 7.2, Memcached version 1.6.21, and Redis OSS version 7.0 and above.

ElastiCache Serverless continuously monitors your cache’s memory, compute, and network utilization to instantly scale. ElastiCache Serverless scales without downtime or performance degradation to the application by allowing the cache to scale up and initiating scale-out in parallel, to meet application requirements just in time. Visit our ElastiCache Serverless documentation to learn more about scaling.

ElastiCache Serverless automatically stores data redundantly across multiple AZs and provides a 99.99% availability Service Level Agreement (SLA) for all workloads.

With ElastiCache Serverless, you only pay for the data you store and the compute your application uses. Visit the ElastiCache pricing page to learn more.

Reserved Nodes or Reserved Instances (RIs) provide you with a significant discount over on-demand usage when you commit to a one-year or three-year term. With Reserved Nodes, you can make a one-time, up-front payment to create a one- or three-year reservation to run your cache in a specific Region and receive a significant discount off the ongoing hourly usage charge. There are three Reserved Node types – All Upfront, No Upfront, and Partial Upfront – that allow you to balance the amount you pay upfront with your effective hourly price.

Reserved Nodes provide a discount that applies to ElastiCache on-demand usage. ElastiCache Serverless is not compatible with Reserved Nodes.

You can purchase up to 300 Reserved Nodes. If you wish to run more than 300 nodes, please complete the ElastiCache node request form.

Purchase a node reservation with the same node class within the same Region as the node you are currently running and would like to reserve. If the reservation purchase is successful, ElastiCache will automatically apply your new hourly usage charge to your existing node.

Pricing changes associated with a Reserved Node are activated once your request is received and while the payment authorization is processed. You can follow the status of your reservation on the AWS Account Activity page or by using the DescribeReservedCacheNodes API. If the one-time payment cannot be successfully authorized by the next billing period, the discounted price will not take effect.

When your reservation term expires, your Reserved Node will revert to the appropriate On-Demand hourly usage rate for your node class and Region.

The ElastiCache APIs to create, modify, and delete nodes do not distinguish between On-Demand and Reserved Nodes, so that you can seamlessly use both. When computing your bill, our system will automatically apply your reservations, such that all eligible nodes are charged at the lower hourly Reserved Cache Node rate.

No. You cannot cancel your node reservation, and the one-time payment (if applicable) is not refundable. You will continue to pay for every hour during your Reserved Node term regardless of your usage.

Each Reserved Node is associated with a specific Region, which is fixed for the lifetime of the reservation and cannot be changed. Each reservation can, however, be used in any of the available AZs within the associated Region.

When you purchase a Reserved Node under the All Upfront payment option, you pay for the entire term of the Reserved Node in one upfront payment. You can choose to pay nothing upfront by choosing the No Upfront option. The entire value of the No Upfront Reserved Node is spread across every hour in the term and you will be billed for every hour in the term, regardless of usage. The Partial Upfront payment option is a hybrid of the All Upfront and No Upfront options. You make a small upfront payment, and you are billed a low hourly rate for every hour in the term regardless of usage.

Yes, ElastiCache reserved nodes offer size flexibility within an instance family (or node family) and AWS region. This means that your existing discounted reserved node rate will be applied automatically to usage of all sizes in the same node family.

ElastiCache allows you to configure encryption of data at rest using AWS Key Management Service (AWS KMS), encryption of data in transit using Transport Layer Security (TLS), authentication using AWS Identity and Access Management (IAM), and network access control with Amazon Elastic Compute Cloud (Amazon EC2) security groups.

When not using Amazon Virtual Private Cloud (Amazon VPC), ElastiCache allows you to control access to your caches through network security groups. A security group acts like a firewall, controlling network access to your cache. By default, network access is turned off to your caches. If you want your applications to access your cache, you must explicitly enable access from hosts in specific Amazon EC2 security groups.

You can also control access to your ElastiCache resources using IAM authentication. For more information, see the authenticating with IAM documentation.

ElastiCache supports compliance programs such as SOC 1, SOC 2, SOC 3, ISO, MTCS, C5, PCI DSS, HIPAA, and FedRAMP. See AWS Services in Scope by Compliance Program for a current list of supported compliance programs.

Yes, the AWS PCI compliance program includes ElastiCache as a PCI-compliant service. To learn more, see the following resources:

• ElastiCache Compliance
• AWS PCI Compliance page

To see the current list of compliance programs that ElastiCache is in scope for, see AWS Services in Scope by Compliance Program.

Yes, ElastiCache is a HIPAA-eligible service and is covered under the AWS Business Associate Addendum (BAA). This means you can use ElastiCache to help you process, maintain, and store protected health information (PHI) and power healthcare applications.

If you have an executed Business Associate Agreement (BAA) with AWS, you can use ElastiCache to build applications that store and process HI under HIPAA. If you do not have a BAA, or have other questions about using AWS for your applications, contact us for more information. 

No, there is no additional cost for using compliance features.

The AWS FedRAMP compliance program includes ElastiCache as a FedRAMP-authorized service. U.S. government customers and their partners can now use the latest version of ElastiCache to process and store their FedRAMP systems, data, and mission-critical, high-impact workloads in the AWS GovCloud (US-East) and AWS GovCloud (US-West) Regions, and at moderate-level impact in the US East (Ohio), US East (N. Virginia), US West (N. California), and US West (Oregon) Regions.

To learn more, see the following resources:

• ElastiCache Compliance
• AWS FedRAMP Compliance

To see the current list of compliance programs that ElastiCache is in scope for, see AWS Services in Scope by Compliance Program.

There are several performance benefits.

ElastiCache provides enhanced I/O threads that deliver significant improvements to throughput and latency at scale through multiplexing, presentation layer offloading, and more. Enhanced I/O threads improve performance by using more cores for processing I/O and dynamically adjusting to the workload. ElastiCache improves the throughput of TLS-enabled clusters by offloading encryption to the same enhanced I/O threads. This enables ElastiCache for Valkey to deliver up to 100% more throughput and 50% lower P99 latency than ElastiCache version 7.0 for Redis OSS. You can achieve over 1 million requests per second per node, or 500 million request per second per cluster, on r7g.4xlarge nodes or larger.

In addition, with ElastiCache version 8.0 for Valkey provides you improved memory efficiency for node-based clusters with Cluster Mode, using 32 bytes less memory per key compared to ElastiCache version 7.2 for Valkey and version 7.1 for Redis OSS. The Serverless configuration has enhanced performance, scaling to 5 million requests per second per cache in minutes, up to 5x faster than Valkey 7.2, with microsecond read latency.

ElastiCache provides two different sets of metrics to measure the CPU utilization of your cache depending on your cache deployment choice. When using ElastiCache Serverless, you can monitor the CPU utilization with the ElastiCache Processing Units (ECPU) metric. The number of ECPUs consumed by your requests depends on the vCPU time taken and the amount of data transferred. Each read and write, like the Valkey GET and SET commands or the Memcached get and set commands, requires 1 ECPU for each kilobyte (KB) of data transferred. Some commands that operate on in-memory data structures can consume more vCPU time than a GET or SET command. ElastiCache calculates the number of ECPUs consumed based on the vCPU time taken by the command compared to a baseline of the vCPU time taken by a SET or GET command. If your command takes additional vCPU time and transfers more data than the baseline of 1 ECPU, then ElastiCache calculates the ECPUs required based on the higher of the two dimensions.

When designing your own cluster, you can monitor EngineCPUUtilization and CPUUtilization. The CPUUtilization metric measures the CPU utilization for the instance (node), and the EngineCPUUtilization metric measures the utilization at the engine process level. You need the EngineCPUUtilization metric in addition to the CPUUtilization metric, as the main engine process is single threaded and uses just one CPU of the multiple CPU cores available on an instance. Therefore, the CPUUtilization metric does not provide precise visibility into the CPU utilization rates at the process level. We recommend that you use both the CPUUtilization and EngineCPUUtilization metrics together to get a detailed understanding of CPU utilization for your Valkey clusters.

Both sets of metrics are available in all AWS Regions, and you can access these metrics using Amazon CloudWatch or in the console. Additionally, we recommend that you visit the documentation to learn about useful metrics for performance monitoring.

Read replicas serve two purposes:

• Failure handling
• Read scaling

When you run a cache with a read replica, the primary serves both writes and reads. The replica serves exclusively read traffic and is also available as a warm standby in the event that the primary becomes impaired.

With ElastiCache Serverless, read replicas are automatically maintained by the service. When designing your own cache, there are a variety of scenarios where deploying one or more read replicas for a given primary node might make sense. Common reasons for deploying a read replica include:

• Scaling beyond the compute or I/O capacity of a single primary node for read-heavy workloads: This excess read traffic can be directed to one or more read replicas.
• Serving read traffic while the primary is unavailable: If your primary node cannot take I/O requests (for example, due to I/O suspension for backups or scheduled maintenance), you can direct read traffic to your read replicas. For this use case, keep in mind that the data on the read replica might be stale, since the primary instance is unavailable. The read replica can also be used warmed up to restart a failed primary.

Data protection scenarios: In the unlikely event of primary node failure or that the AZ in which your primary node resides becomes unavailable, you can promote a read replica in a different AZ to become the new primary.

You can connect to a read replica just as you would connect to a primary cache node. If you have multiple read replicas, it is up to your application to determine how read traffic will be distributed amongst them. Here are more details:

• Valkey or Redis (cluster mode disabled) OSS clusters, use the individual node endpoints for read operations. (In the API/CLI these are referred to as read endpoints.)
• Valkey or Redis OSS (cluster mode enabled) clusters, use the cluster's configuration endpoint for all operations. You can still read from individual node endpoints. (In the API and CLI these are referred to as read endpoints.)

ElastiCache allows you to create up to five (5) read replicas for a given primary cache node.

In the event of a failover, any associated and available read replicas should automatically resume replication once failover has completed (acquiring updates from the newly promoted read replica).

Updates to a primary cache node will automatically be replicated to any associated read replicas. However, with Valkey or Redis OSS asynchronous replication technology, a read replica can fall behind its primary cache node for a variety of reasons. Typical reasons include:

• Write I/O volume to the primary cache node exceeds the rate at which changes can be applied to the read replica.
• Network partitions or latency between the primary cache node and a read replica.

Read replicas are subject to the strengths and weaknesses of Valkey or Redis OSS replication. If you are using read replicas, you should be aware of the potential for lag between a read replica and its primary cache node or “inconsistency.” ElastiCache emits a metric to help you understand the inconsistency.

A read replica is billed as a standard cache node and at the same rates. Just like a standard cache node, the rate per cache node hour for a read replica is determined by the cache node class of the read replica: visit the ElastiCache pricing page for up-to-date pricing. You are not charged for the data transfer incurred in replicating data between your primary cache node and read replica. Billing for a read replica begins as soon as the read replica has been successfully created (when the status is listed as active). The read replica will continue being billed at standard ElastiCache cache node hour rates until you issue a command to delete it.

Initiated failover is supported by ElastiCache so that you can resume cache operations as quickly as possible. When failing over, ElastiCache flips the DNS record for your cache node to point at the read replica, which is in turn promoted to become the new primary. We encourage you to follow best practices and implement cache node connection retry at the application layer. Typically, start to finish, steps one to five below complete within six minutes.

These are the automatic failover events, listed in order of occurrence:

1. Replication group message: Test Failover API called for node group
2. Cache cluster message: Failover from primary node to replica node completed
3. Replication group message: Failover from primary node to replica node completed
4. Cache cluster message: Recovering cache nodes
5. Cache cluster message: Finished recovery for cache nodes

No, your read replica can only be provisioned in the same or different AZ of the same Region as your cache node primary. You can, however, use the Global Datastore to work with fully managed, fast, reliable, and security-focused replication across AWS Regions. Using this feature, you can create cross-Region read replica clusters for ElastiCache to enable low-latency reads and disaster recovery across AWS Regions.

Yes. You can add or remove a read replica across one or more shards in a cluster environment. The cluster continues to stay online and serve incoming I/O during this operation.

Initiated failover is supported by ElastiCache so that you can resume cache operations as quickly as possible. When failing over, ElastiCache flips the DNS record for your cache node to point at the read replica, which is in turn promoted to become the new primary. We encourage you to follow best practices and implement cache node connection retry at the application layer. Typically, a new healthy node joins the cluster within six minutes. The node synchronization time varies based on the cluster's workload and size, which impacts the completion time of steps one to five below. These are the automatic failover events, listed in order of occurrence:

• Replication group message: Test Failover API called for node group <node-group-id>
• Cache cluster message: Failover from primary node <primary-node-id> to replica node <node-id> completed
• Replication group message: Failover from primary node <primary-node-id> to replica node <node-id> completed
• Cache cluster message: Recovering cache nodes <node-id>
• Cache cluster message: Finished recovery for cache nodes <node-id>

Multi-AZ is a feature that allows you to run in a more highly available configuration when designing your own ElastiCache cache. All ElastiCache Serverless caches are automatically run in a Multi-AZ configuration. An ElastiCache replication group consists of a primary and up to five read replicas. If Multi-AZ is enabled, then at least one replica is required per primary. During certain types of planned maintenance, or in the unlikely event of an ElastiCache node failure or AZ failure, ElastiCache will automatically detect the failure of a primary, select a read replica, and promote it to become the new primary. ElastiCache also propagates the DNS changes of the promoted read replica, so if your application is writing to the primary node endpoint, no endpoint change will be needed.

The main benefits of running your ElastiCache in Multi-AZ mode are enhanced availability and a smaller need for administration. When running ElastiCache in a Multi-AZ configuration, your caches are eligible for the 99.99% availability SLA. If an ElastiCache primary node failure occurs, the impact on your ability to read and write to the primary is limited to the time it takes for automatic failover to complete. When Multi-AZ is enabled, ElastiCache node failover is automatic and requires no administration.

You can use Multi-AZ if you are using ElastiCache and have a replication group consisting of a primary node and one or more read replicas. If the primary node fails, ElastiCache will automatically detect the failure, select one from the available read replicas, and promote it to become the new primary. ElastiCache will propagate the DNS changes of the promoted replica so that your application can keep writing to the primary endpoint. ElastiCache will also spin up a new node to replace the promoted read replica in the same AZ of the failed primary. In case the primary failed due to temporary AZ disruption, the new replica will be launched once that AZ has recovered.

Yes. Note that placing both the primary and the replicas in the same AZ will not make your ElastiCache replication group resilient to an AZ disruption.

ElastiCache will fail over to a read replica in the event of any of the following:

• Loss of availability in primary’s AZ
• Loss of network connectivity to primary
• Compute unit failure on primary

If there is more than one read replica, the read replica with the smaller asynchronous replication lag to the primary will be promoted.

Yes, ElastiCache will create an event to inform you that automatic failover occurred. You can use the DescribeEvents API to return information about events related to your ElastiCache node, or select the Events section in the ElastiCache Management Console.

AZs are engineered to provide low latency network connectivity to other AZs in the same Region. You should consider architecting your application and other AWS resources with redundancy across multiple AZs so your application will be resilient in the event of an AZ disruption.

For more information about Multi-AZ, see ElastiCache documentation.

Backup and Restore is a feature that allows you to create snapshots of your ElastiCache caches. ElastiCache stores the snapshots, allowing users to subsequently use them to restore caches. This is currently supported with ElastiCache for Valkey, ElastiCache for Redis OSS and Serverless.

Creating snapshots can be useful in case of data loss caused by node failure, as well as the unlikely event of a hardware failure. Another common reason to use backups is for archiving purposes. Snapshots are stored in Amazon S3.

Yes, you can export your ElastiCache snapshots to an authorized S3 bucket in the same Region as your cache.

Yes. You must first copy your snapshot into an authorized S3 bucket of your choice in the same Region and then grant cross-account bucket permissions to the other account.

ElastiCache provides storage space for one snapshot free of charge for each active ElastiCache cache. Additional storage will be charged based on the space used by the snapshots with $0.085/GB every month (same price in all Regions). Data transfer for using the snapshots is free of charge.

When you delete an ElastiCache cache, your manual snapshots are retained. You will also have an option to create a final snapshot before the cache is deleted. Automatic cache snapshots are not retained.

Yes, a backup taken from any version of ElastiCache for Redis OSS can be restored to any version of ElastiCache for Valkey. For more information about restoring backups, see the ElastiCache documentation.

The engine within ElastiCache is fully compatible with Valkey and Redis OSS but also comes with enhancements that improve performance, robustness, and stability. Some of the enhancements include:

• More usable memory: You can now safely allocate more memory for your application without risking increased swap usage during syncs and snapshots.
• Improved synchronization: More robust synchronization under heavy load and when recovering from network disconnections. Additionally, syncs are faster as both the primary and replicas no longer use the disk for this operation.
• Smoother failovers: In the event of a failover, your shard now recovers faster as replicas no longer flush their data to do a full resync with the primary.

The enhanced engine is fully compatible with Valkey or Redis OSS, thus you can enjoy its improved robustness and stability without the need to make any changes to your application code.

There is no additional charge for using the enhanced engine.

Encryption in transit, encryption at rest, Valkey AUTH, and Role-Based Access Control (RBAC) are features you can select when creating your ElastiCache cache. If you enabled in-transit encryption, you can choose to use AUTH or RBAC for additional security and access control.

Encryption at rest provides mechanisms to guard against unauthorized access of your data. When enabled, it encrypts the following aspects:

• Disk during sync, backup, and swap operations
• Backups stored in Amazon S3

ElastiCache offers default (service-managed) encryption at rest, as well as the ability to use your own symmetric customer-managed AWS KMS keys in AWS KMS. Visit at-rest encryption to learn more.

The encryption in transit feature facilitates encryption of communications between clients and ElastiCache, as well as between the servers (primary and read replicas). Read more about ElastiCache in-transit encryption.

No, ElastiCache manages certification expiration and renewal behind the scenes. No user action is necessary for ongoing certificate maintenance.

No, there are no additional costs for using encryption.

Global Datastore is a feature of ElastiCache that provides fully managed, fast, reliable and security-focused cross-Region replication. With Global Datastore, you can write to your cache in one Region and have the data available for read in up to two other cross-Region replica clusters, thereby enabling low-latency reads and disaster recovery across Regions.

Designed for real-time applications with a global footprint, Global Datastore typically replicates data across Regions in under one second, increasing the responsiveness of your applications by providing geolocal reads closer to end users. In the unlikely event of Regional degradation, one of the healthy cross-Region replica caches can be promoted to become the primary with full read and write capabilities. Once initiated, the promotion typically completes in less than a minute, allowing your applications to remain available.

Global Datastore is supported on ElastiCache version 7.2 for Valkey and ElastiCache version 5.0.6 onward for Redis OSS.

You can replicate to up to two secondary Regions within a Global Datastore. The caches in secondary Regions can be used to serve low-latency local reads and for disaster recovery in the unlikely event of Regional degradation.

You can set up a Global Datastore by using an existing cache or creating a new cache to be used as a primary. You can create a Global Datastore with just a few steps in the ElastiCache Management Console or by downloading the latest AWS SDK or AWS CLI. There is support for Global Datastore in AWS CloudFormation.

No, ElastiCache doesn’t automatically promote a secondary cluster in the event when a primary cluster (Region) is degraded. You can manually initiate the failover by promoting a secondary cluster to become a primary. The fail over and promotion of a secondary cluster typically completes in less than one minute.

ElastiCache doesn’t provide an SLA for RPO and RTO. The RPO varies based on replication lag between Regions and depends on network latency between Regions and cross-Region network traffic congestion. The RPO of Global Datastore is typically under one second, so the data written in a primary Region is available in secondary Regions within one second. The RTO of Global Datastore is typically under a minute. Once a failover to a secondary cluster is initiated, ElastiCache typically promotes the secondary to full read and write capabilities in under a minute.

ElastiCache does not charge any premium to use Global Datastore. You pay for the primary and secondary caches in your Global Datastore and for the cross-Region data transfer traffic.

Though there is no precise answer for this question, with ElastiCache, you don't need to worry about getting the number of nodes exactly right, as you can quickly add or remove nodes later. You can also use ElastiCache Serverless to simplify running a highly available Memcached cache. The following two inter-related aspects can be considered for the choice of your initial configuration:

• The total memory required for your data to achieve your target cache-hit rate, and
• The number of nodes required to maintain acceptable application performance without overloading the database backend in the event of node failures.

The amount of memory required is dependent upon the size of your data set and the access patterns of your application. To improve fault tolerance, once you have a rough idea of the total memory required, divide that memory into enough nodes such that your application can survive the loss of one or two nodes. For example, if your memory requirement is 13 GB, you might want to use two cache.m4.large nodes instead of using one cache.m4.xlarge node. It is important that other systems such as databases will not be overloaded if the cache-hit rate is temporarily reduced during failure recovery of one or more nodes. Please refer to the ElastiCache User Guide for more details.

Yes. When creating a cluster or adding nodes to an existing cluster, you can choose the AZs for the new nodes. You can either specify the requested number of nodes in each AZ or select Spread Nodes Across Zones. If the cluster is in Amazon VPC, nodes can only be placed in AZs that are part of the selected cache subnet group. For additional details, please see ElastiCache VPC documentation.

You can run a maximum of 300 nodes per Region. If you need more nodes, please fill in the ElastiCache Limit Increase Request form.

The service will detect the node failure and react with the following automatic steps:

• ElastiCache will repair the node by acquiring new service resources and will then redirect the node's existing DNS name to point to the new service resources. For Amazon VPC installations, ElastiCache will ensure that both the DNS name and the IP address of the node remain the same when nodes are recovered in case of failure. For non–Amazon VPC installations, ElastiCache will ensure that the DNS name of a node is unchanged; however, the underlying IP address of the node can change.
• If you associated an SNS topic with your cluster, when the new node is configured and ready to be used, ElastiCache will send an SNS notification to let you know that node recovery occurred. This allows you to optionally arrange for your applications to force the Memcached client library to attempt to reconnect to the repaired nodes. This could be important, as some Memcached libraries will stop using a server (node) indefinitely if they encounter communication errors or timeouts with that server.

You could add more nodes to your existing Memcached Cluster by using the Add Node option on the Nodes tab for your Cache Cluster in the console or calling the ModifyCacheCluster API.

ElastiCache is ideally suited as a front end for AWS services like Amazon RDS and DynamoDB, providing extremely low latency for high-performance applications and offloading some of the request volume while these services provide long-lasting data durability. The service can also be used to improve application performance in conjunction with Amazon EC2 and Amazon EMR.

Memcached client libraries are available for many, if not all of the popular programming languages. If you encounter any issues with specific Memcached clients when using ElastiCache, please engage us in the ElastiCache community forum.

ElastiCache does not require specific client libraries and works with existing Memcached client libraries without recompilation or application relinking (Memcached 1.4.5 and later). Examples include libMemcached (C) and libraries based on it (for instance, PHP, Perl, Python), spyMemcached (Java) and fauna (Ruby).

Auto Discovery is a feature that saves developers time and effort, while reducing the complexity of their applications. Auto Discovery enables automatic discovery of cache nodes by clients when they are added to or removed from an ElastiCache cluster. Previously, to handle cluster membership changes, developers had to manually update the list of cache node endpoints. Depending on how the client application is architected, typically a client initialization (by shutting down the application and restarting it) is needed, resulting in downtime. Through Auto Discovery, ElastiCache removes this complexity. With Auto Discovery, in addition to being backwards protocol compliant with the Memcached protocol, ElastiCache provides clients with information on cache cluster membership. A client capable of processing the additional information reconfigures itself, without any initialization, to use the most current nodes of an ElastiCache cluster.

An ElastiCache cluster can be created with nodes that are addressable through named endpoints. With Auto Discovery the ElastiCache cluster is also given a unique configuration endpoint, which is a DNS Record that is valid for the lifetime of the cluster. This DNS Record contains the DNS Names of the nodes that belong to the cluster. ElastiCache will ensure that the configuration endpoint always points to at least one such target node. A query to the target node then returns endpoints for all the nodes of the cluster in question. After this, you can connect to the cluster nodes just as before and use the Memcached protocol commands such as get, set, incr, and decr. For more details, see the documentation. To use Auto Discovery, you will need a client capable of Auto Discovery. Auto Discovery clients for .Net , Java and PHP are available for download from the ElastiCache console. Upon initialization, the client will automatically determine the current members of the ElastiCache cluster using the configuration endpoint. When you make changes to your cache cluster by adding or removing nodes, or if a node is replaced upon failure, the Auto Discovery client automatically determines the changes, and you do not need to initialize your clients manually.

To get started, download the ElastiCache Cluster Client by selecting the Download ElastiCache Cluster Client link on the ElastiCache console. Before you can download, you must have an ElastiCache account; if you do not already have one, you can sign up from the ElastiCache detail page. After you download the client, you can begin setting up and activating your ElastiCache cluster by visiting the ElastiCache console. More details can be found in the documentation.

Yes, you can stop using Auto Discovery at any time. You can disable Auto Discovery by specifying the mode of operation during the ElastiCache Cluster client initialization. Also, since ElastiCache continues to support Memcached, you can use any Memcached protocol-compliant client as before.

To take advantage of Auto Discovery, a client capable of Auto Discovery must be used to connect to an ElastiCache Cluster. ElastiCache currently supports clients capable of Auto Discovery for .Net, Java, and PHP. These can be downloaded from the ElastiCache console. You can create clients for any other languages by building upon the popular Memcached clients available.

You can take any Memcached Client Library and add support for Auto Discovery. If you would like to add or modify your own client to enable Auto Discovery, please refer to the Auto Discovery command set documentation.

Yes, ElastiCache is still Memcached protocol compliant and does not require you to change your clients. However, to take advantage of the Auto Discovery feature, we enhanced the Memcached client capabilities. If you choose not to use the ElastiCache Cluster Client, you can continue to use your own clients or modify your own client library to understand the Auto Discovery command set.
 

Yes, the same ElastiCache cluster can be connected through a client capable of Auto Discovery and the traditional Memcached client at the same time. ElastiCache remains 100% Memcached compliant.

Yes, you can stop using Auto Discovery at any time. You can disable Auto Discovery by specifying the mode of operation during the ElastiCache Cluster client initialization. Also, since ElastiCache continues to support Memcached, you can use any Memcached protocol-compliant client as before.

ElastiCache allows you to control if and when the Memcached protocol-compliant software powering your cluster is upgraded to new versions supported by ElastiCache. This provides you with the flexibility to maintain compatibility with specific Memcached versions, test new versions with your application before deploying in production, and perform version upgrades on your own terms and timelines. Version upgrades involve some compatibility risk; thus they will not occur automatically and must be initiated by you. This approach to software patching puts you in the driver's seat of version upgrades, but still offloads the work of patch application to ElastiCache. You can learn more about version management by reading the FAQs that follow. Alternatively, you can refer to the ElastiCache User Guide. While Engine Version Management functionality is intended to give you as much control as possible over how patching occurs, we can patch your cluster on your behalf if we determine there is any security vulnerability in the system or cache software.

You can specify any currently supported version (minor or major) when creating a new cluster. If you wish to initiate an upgrade to a supported engine version release, you can do so using the Modify option for your cluster. Specify the version you wish to upgrade to in the Cache Engine Version field. The upgrade will then be applied on your behalf either immediately (if the Applied Immediately option is checked) or during the next scheduled maintenance window for your cluster.

Yes. You can do so by creating a new cluster with the new engine version. You can point your development or staging application to this cluster, test it, and decide whether or not to upgrade your original cluster.

We plan to support additional Memcached versions for ElastiCache, both major and minor. The number of new version releases supported in a given year will vary based on the frequency and content of the Memcached version releases and the outcome of a thorough vetting of the release by our engineering team.

You can upgrade your existing Memcached cluster by using the Modify process. When upgrading from an older version of Memcached to Memcached version 1.4.33 or newer, please ensure that your existing parameter max_chunk_size values satisfies conditions needed for slab_chunk_max parameter. Please review upgrade prerequisites.

ElastiCache provides enhanced I/O threads that deliver significant improvements to throughput and latency at scale through multiplexing, presentation layer offloading, and more. Enhanced I/O threads improve performance by using more cores for processing I/O and dynamically adjusting to the workload. ElastiCache improves the throughput of TLS-enabled clusters by offloading encryption to the same enhanced I/O threads. ElastiCache (Redis OSS) version 7.0 introduced enhanced I/O multiplexing, that combines many client requests into a single channel and improves thread efficiency.

In ElastiCache version 7.1 and above for Redis OSS , we extended the enhanced I/O threads functionality to also handle the presentation layer logic. Enhanced I/O threads not only read client input, but also parses the input into a binary command format, which is then forwarded to the main thread for execution, to provide performance gains. With ElastiCache version 7.1 for Redis OSS, you can achieve up to 100% more throughput and 50% lower P99 latency, compared to prior version. On r7g.4xlarge or larger, you can achieve over 1 million requests per second (RPS) per node.

ElastiCache provides two different sets of metrics to measure the CPU utilization of your cache depending on your cache deployment choice. When using ElastiCache Serverless, you can monitor the CPU utilization with the ElastiCache Processing Units (ECPU) metric. The number of ECPUs consumed by your requests depends on the vCPU time taken and the amount of data transferred. Each read and write, like the Redis OSS GET and SET commands or the Memcached get and set commands, requires 1 ECPU for each kilobyte (KB) of data transferred. Some Redis OSS commands that operate on in-memory data structures can consume more vCPU time than a GET or SET command. ElastiCache calculates the number of ECPUs consumed based on the vCPU time taken by the command compared to a baseline of the vCPU time taken by a SET or GET command. If your command takes additional vCPU time and transfers more data than the baseline of 1 ECPU, then ElastiCache calculates the ECPUs required based on the higher of the two dimensions.

When designing your own cluster, you can monitor EngineCPUUtilization and CPUUtilization. The CPUUtilization metric measures the CPU utilization for the instance (node), and the EngineCPUUtilization metric measures the utilization at the engine process level. You need the EngineCPUUtilization metric in addition to the CPUUtilization metric, as the main Redis OSS process is single threaded and uses just one CPU of the multiple CPU cores available on an instance. Therefore, the CPUUtilization metric does not provide precise visibility into the CPU utilization rates at the engine process level. We recommend that you use both the CPUUtilization and EngineCPUUtilization metrics together to get a detailed understanding of CPU utilization for your Redis OSS clusters.

Both sets of metrics are available in all AWS Regions, and you can access these metrics using Amazon CloudWatch or in the console. Additionally, we recommend that you visit the documentation to learn about useful metrics for performance monitoring.

Read replicas serve two purposes in Redis OSS:

• Failure handling
• Read scaling

When you run a cache with a read replica, the primary serves both writes and reads. The replica serves exclusively read traffic and is also available as a warm standby in the event that the primary becomes impaired.

With ElastiCache Serverless, read replicas are automatically maintained by the service. When designing your own cache, there are a variety of scenarios where deploying one or more read replicas for a given primary node might make sense. Common reasons for deploying a read replica include:

• Scaling beyond the compute or I/O capacity of a single primary node for read-heavy workloads: This excess read traffic can be directed to one or more read replicas.
• Serving read traffic while the primary is unavailable: If your primary node cannot take I/O requests (for example, due to I/O suspension for backups or scheduled maintenance), you can direct read traffic to your read replicas. For this use case, keep in mind that the data on the read replica might be stale, since the primary instance is unavailable. The read replica can also be used warmed up to restart a failed primary.
• Data protection scenarios: In the unlikely event of primary node failure or that the AZ in which your primary node resides becomes unavailable, you can promote a read replica in a different AZ to become the new primary. 

You can connect to a read replica just as you would connect to a primary cache node. If you have multiple read replicas, it is up to your application to determine how read traffic will be distributed amongst them. Here are more details:

• Redis (cluster mode disabled) OSS clusters, use the individual node endpoints for read operations. (In the API/CLI these are referred to as read endpoints.)
• Redis OSS (cluster mode enabled) clusters, use the cluster's configuration endpoint for all operations. You can still read from individual node endpoints. (In the API and CLI these are referred to as read endpoints.)

ElastiCache allows you to create up to five (5) read replicas for a given primary cache node.

In the event of a failover, any associated and available read replicas should automatically resume replication once failover has completed (acquiring updates from the newly promoted read replica).

Updates to a primary cache node will automatically be replicated to any associated read replicas. However, with Redis OSS asynchronous replication technology, a read replica can fall behind its primary cache node for a variety of reasons. Typical reasons include:

• Write I/O volume to the primary cache node exceeds the rate at which changes can be applied to the read replica.
• Network partitions or latency between the primary cache node and a read replica.

Read replicas are subject to the strengths and weaknesses of Redis OSS replication. If you are using read replicas, you should be aware of the potential for lag between a read replica and its primary cache node or “inconsistency.” ElastiCache emits a metric to help you understand the inconsistency.

A read replica is billed as a standard cache node and at the same rates. Just like a standard cache node, the rate per cache node hour for a read replica is determined by the cache node class of the read replica: visit the ElastiCache pricing page for up-to-date pricing. You are not charged for the data transfer incurred in replicating data between your primary cache node and read replica. Billing for a read replica begins as soon as the read replica has been successfully created (when the status is listed as active). The read replica will continue being billed at standard ElastiCache cache node hour rates until you issue a command to delete it.

Initiated failover is supported by ElastiCache so that you can resume cache operations as quickly as possible. When failing over, ElastiCache flips the DNS record for your cache node to point at the read replica, which is in turn promoted to become the new primary. We encourage you to follow best practices and implement cache node connection retry at the application layer. Typically, start to finish, steps one to five below complete within six minutes.

These are the automatic failover events, listed in order of occurrence:

1. Replication group message: Test Failover API called for node group
2. Cache cluster message: Failover from primary node to replica node completed
3. Replication group message: Failover from primary node to replica node completed
4. Cache cluster message: Recovering cache nodes
5. Cache cluster message: Finished recovery for cache nodes

No, your read replica can only be provisioned in the same or different AZ of the same Region as your cache node primary. You can, however, use the Global Datastore to work with fully managed, fast, reliable, and security-focused replication across AWS Regions. Using this feature, you can create cross-Region read replica clusters for ElastiCache to enable low-latency reads and disaster recovery across AWS Regions.

Yes. You can gain visibility into the location of the current primary by using the console or DescribeCacheClusters API.

Yes. You can add or remove a read replica across one or more shards in a cluster environment. The cluster continues to stay online and serve incoming I/O during this operation.

Multi-AZ is a feature that allows you to run in a more highly available configuration when designing your own ElastiCache cache. All ElastiCache Serverless caches are automatically run in a Multi-AZ configuration. An ElastiCache replication group consists of a primary and up to five read replicas. If Multi-AZ is enabled, then at least one replica is required per primary. During certain types of planned maintenance, or in the unlikely event of an ElastiCache node failure or AZ failure, ElastiCache will automatically detect the failure of a primary, select a read replica, and promote it to become the new primary. ElastiCache also propagates the DNS changes of the promoted read replica, so if your application is writing to the primary node endpoint, no endpoint change will be needed.

The main benefits of running your ElastiCache in Multi-AZ mode are enhanced availability and a smaller need for administration. When running ElastiCache in a Multi-AZ configuration, your caches are eligible for the 99.99% availability SLA. If an ElastiCache primary node failure occurs, the impact on your ability to read and write to the primary is limited to the time it takes for automatic failover to complete. When Multi-AZ is enabled, ElastiCache node failover is automatic and requires no administration. 

You can use Multi-AZ if you are using ElastiCache and have a replication group consisting of a primary node and one or more read replicas. If the primary node fails, ElastiCache will automatically detect the failure, select one from the available read replicas, and promote it to become the new primary. ElastiCache will propagate the DNS changes of the promoted replica so that your application can keep writing to the primary endpoint. ElastiCache will also spin up a new node to replace the promoted read replica in the same AZ of the failed primary. In case the primary failed due to temporary AZ disruption, the new replica will be launched once that AZ has recovered.

Yes. Note that placing both the primary and the replicas in the same AZ will not make your ElastiCache replication group resilient to an AZ disruption. Additionally, having replicas in the same AZ as the primary will not be allowed if Multi-AZ is turned on.

ElastiCache will fail over to a read replica in the event of any of the following:

• Loss of availability in primary’s AZ
• Loss of network connectivity to primary
• Compute unit failure on primary

If there is more than one read replicas, the read replica with the smaller asynchronous replication lag to the primary will be promoted.

Yes, ElastiCache will create an event to inform you that automatic failover occurred. You can use the DescribeEvents API to return information about events related to your ElastiCache node, or select the Events section in the ElastiCache Management Console.

AZs are engineered to provide low latency network connectivity to other AZs in the same Region. You should consider architecting your application and other AWS resources with redundancy across multiple AZs so your application will be resilient in the event of an AZ disruption.

For more information about Multi-AZ, see ElastiCache documentation.

Creating snapshots can be useful in case of data loss caused by node failure, as well as the unlikely event of a hardware failure. Another common reason to use backups is for archiving purposes. Snapshots are stored in Amazon S3.

When a backup is initiated, ElastiCache will take a snapshot of a specified cache that can later be used for recovery or archiving. You can initiate a backup any time you choose or set a recurring daily backup with retention period of up to 35 days. When you choose a snapshot to restore, a new ElastiCache cache will be created and populated with the snapshot’s data. ElastiCache snapshots are compatible with the Redis OSS RDB file format.

You can use the Backup and Restore feature through the console, ElastiCache APIs, and AWS CLI. You can deactivate and reactivate the feature any time you choose.

Backup and Restore creates snapshots on a per-cache basis. Users can specify which ElastiCache cache to back up through the console, AWS CLI, or the ElastiCache API. We recommend users enable backup on one of the cache’s read replicas, minimizing any potential impact on the primary. When using ElastiCache Serverless, backups are automatically conducted against read replicas.

Yes, you can export your ElastiCache snapshots to an authorized S3 bucket in the same Region as your cache.

Yes. You must first copy your snapshot into an authorized S3 bucket of your choice in the same Region and then grant cross-account bucket permissions to the other account.

ElastiCache provides storage space for one snapshot free of charge for each active ElastiCache cache. Additional storage will be charged based on the space used by the snapshots with $0.085/GB every month (same price in all Regions). Data transfer for using the snapshots is free of charge.

When you delete an ElastiCache cache, your manual snapshots are retained. You will also have an option to create a final snapshot before the cache is deleted. Automatic cache snapshots are not retained.

The engine within ElastiCache is fully compatible with Redis OSS but also comes with enhancements that improve performance, robustness, and stability. Some of the enhancements include:

• More usable memory: You can now safely allocate more memory for your application without risking increased swap usage during syncs and snapshots.
• Improved synchronization: More robust synchronization under heavy load and when recovering from network disconnections. Additionally, syncs are faster as both the primary and replicas no longer use the disk for this operation.
• Smoother failovers: In the event of a failover, your shard now recovers faster as replicas no longer flush their data to do a full resync with the primary.
• TLS offload and IO multiplexing: ElastiCache is designed to better use available CPU resources by handling certain network-related processes on dedicated threads.

No. The enhanced engine is fully compatible with Redis OSS, thus you can enjoy its improved robustness and stability without the need to make any changes to your application code.

There is no additional charge for using the enhanced engine.

Encryption at rest provides mechanisms to guard against unauthorized access of your data. When enabled, it encrypts the following aspects:

• Disk during sync, backup, and swap operations
• Backups stored in Amazon S3

ElastiCache offers default (service-managed) encryption at rest, as well as the ability to use your own symmetric customer-managed AWS KMS keys in AWS KMS. Visit at-rest encryption to learn more.

The encryption in transit feature facilitates encryption of communications between clients and ElastiCache, as well as between the servers (primary and read replicas). Read more about ElastiCache in-transit encryption.

Encryption in transit, encryption at rest, Redis OSS AUTH, and Role-Based Access Control (RBAC) are features you can select when creating your ElastiCache cache. If you enabled in-transit encryption, you can choose to use Redis OSS AUTH or RBAC for additional security and access control.

No, ElastiCache manages certification expiration and renewal behind the scenes. No user action is necessary for ongoing certificate maintenance.

No, there are no additional costs for using encryption.

Global Datastore is a feature of ElastiCache that provides fully managed, fast, reliable and security-focused cross-Region replication. With Global Datastore, you can write to your cache in one Region and have the data available for read in up to two other cross-Region replica clusters, thereby enabling low-latency reads and disaster recovery across Regions.

Designed for real-time applications with a global footprint, Global Datastore typically replicates data across Regions in under one second, increasing the responsiveness of your applications by providing geolocal reads closer to end users. In the unlikely event of Regional degradation, one of the healthy cross-Region replica caches can be promoted to become the primary with full read and write capabilities. Once initiated, the promotion typically completes in less than a minute, allowing your applications to remain available.

You can replicate to up to two secondary Regions within a Global Datastore. The caches in secondary Regions can be used to serve low-latency local reads and for disaster recovery in the unlikely event of Regional degradation.

Global Datastore is supported on ElastiCache for Redis 5.0.6 onward.

No, ElastiCache doesn’t automatically promote a secondary cluster in the event when a primary cluster (Region) is degraded. You can manually initiate the failover by promoting a secondary cluster to become a primary. The fail over and promotion of a secondary cluster typically completes in less than one minute.

Global Datastore uses encryption in transit for cross-Region traffic to keep your data more secure. Additionally, you can also encrypt your primary and secondary caches using encryption at rest to keep your data better secured. Each primary and secondary cache can have a separate customer-managed AWS KMS key for encryption at rest.

ElastiCache doesn’t provide an SLA for RPO and RTO. The RPO varies based on replication lag between Regions and depends on network latency between Regions and cross-Region network traffic congestion. The RPO of Global Datastore is typically under one second, so the data written in a primary Region is available in secondary Regions within one second. The RTO of Global Datastore is typically under a minute. Once a failover to a secondary cluster is initiated, ElastiCache typically promotes the secondary to full read and write capabilities in under a minute.

ElastiCache doesn’t provide an SLA for RPO and RTO. The RPO varies based on replication lag between Regions and depends on network latency between Regions and cross-Region network traffic congestion. The RPO of Global Datastore is typically under one second, so the data written in a primary Region is available in secondary Regions within one second. The RTO of Global Datastore is typically under a minute. Once a failover to a secondary cluster is initiated, ElastiCache typically promotes the secondary to full read and write capabilities in under a minute.

Data tiering provides a new price-performance option by using lower-cost solid state drives (SSDs) in each cluster node in addition to storing data in memory. It is ideal for workloads that regularly access up to 20% of their overall dataset and for applications that can tolerate additional latency when accessing data on SSD. ElastiCache R6gd nodes with memory and SSDs have nearly 5x more total storage capacity and can help you achieve over 60% savings in price when running at maximum use compared to ElastiCache R6g nodes with memory only.

Data tiering works by automatically and transparently moving the least recently used items from memory to locally attached NVMe SSDs when available memory capacity is completely consumed. When an item that moves to SSD is subsequently accessed, ElastiCache moves it back to memory asynchronously before serving the request.

Data tiering is designed to have minimal impact on application performance. Assuming 500-byte string values, you can expect an additional 300µs latency on average for requests to data stored on SSD compared to requests to data in memory.

ElastiCache supports data tiering for ElastiCache for Redis OSS versions 6.2 and above.

ElastiCache supports data tiering on clusters using R6gd nodes.

All Valkey and Redis OSS commands and most ElastiCache features are supported when using data tiering. For a list of features that are not supported on clusters using data tiering, see the documentation.

There are no additional costs for using data tiering besides the node’s hourly cost. Nodes with data tiering are available with on-demand pricing and as reserved nodes. For pricing, see the ElastiCache pricing page.