Best practices when enabling in-transit encryption - Amazon ElastiCache (Redis OSS)
Before enabling in-transit encryption: make sure you have proper DNS records handlingDuring the in-transit encryption: pay attention to when the migration process finishes
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Best practices when enabling in-transit encryption

Before enabling in-transit encryption: make sure you have proper DNS records handling

Note

We are changing and deleting old endpoints during this process. Incorrect usage of the endpoints can result in the Redis OSS client using old and deleted endpoints that will prevent it from connecting to the cluster.

While the cluster is being migrated from no-TLS to TLS-preferred, the old per-node DNS records are kept and the new per-node DNS records are being generated in a different format. TLS-enabled clusters use a different format of DNS records than non-TLS-enabled clusters. ElastiCache will keep both DNS records when a cluster is configured in encryption mode: Preferred so that Applications and other Redis OSS clients can switch between them. The following changes in the DNS records take place during the TLS-migration process:

Description of the changes in the DNS records that take place when enabling in-transit encryption

For CME clusters

When a cluster is set to ‘transit encryption mode: preferred’:

  • The original cluster endpoints for non-TLS enabled cluster will remain active. There will be no downtime when cluster is re-configured form TLS encryption mode ‘none’ to ‘preferred’.

  • New TLS Redis OSS endpoints will be generated when cluster is set to TLS-preferred mode. These new endpoints will resolve to the same IPs as the old ones (non-TLS).

  • The new TLS Redis OSS configuration endpoint will be exposed in the ElastiCache Console and in the response to describe-replication-group API.

When a cluster is set to ‘transit encryption mode: required’:

  • Old non-TLS enabled endpoints will be deleted. There will be no downtime of TLS cluster endpoints.

  • You can retrieve a new cluster-configuration-endpoint from ElastiCache Console or from the describe-replication-group API.

For CMD clusters with Automatic Failover enabled or Automatic Failover disabled

When replication group is set to ‘transit encryption mode: preferred’:

  • The original primary endpoint and reader endpoint for non-TLS enabled cluster will remain active.

  • New TLS primary and reader endpoints will be generated when cluster is set to TLS Preferred mode. This new endpoints will resolve to the same IP(s) as the old ones (non-TLS).

  • The new primary endpoint and reader endpoint will be exposed in the ElastiCache Console and in the response to the describe-replication-group API.

When replication group is set to ‘transit encryption mode: required’:

  • Old non-TLS primary and reader endpoints will be deleted. There will be no downtime of TLS cluster endpoints.

  • You can retrieve new primary and reader endpoints from ElastiCache Console or from the describe-replication-group API.

The suggested usage of the DNS records

For CME clusters

  • Use the cluster configuration endpoint instead of per-node DNS records in your application’s code. Using per-node DNS names directly is not recommended because they might change when adding or removing shards.

  • Don't hardcode cluster configuration endpoint in your application as it will change during this process.

  • Having the cluster configuration endpoint hardcoded in your application is a bad practice since it can be changed during this process. After the in-transit encryption is completed, query the cluster configuration endpoint with the describe-replication-group API (as demonstrated above (in bold)) and use the DNS you get in response from this point on.

For CMD clusters with Automatic Failover enabled

  • Use the primary endpoint and reader endpoint instead of the per-node DNS names in your application’s code since the old per-node DNS names are deleted and new ones are generated when migrating the cluster from no-TLS to TLS-preferred. Using per-node DNS names directly is not recommended because you might add replicas to your cluster in the future. Also, when Automatic Failover is enabled, the roles of the primary cluster and replicas are changed automatically by the ElastiCache service, using the primary endpoint and reader endpoint is suggested to help you keep track of those changes. Lastly, using the reader endpoint will help you distribute your reads from the replicas equally between the replicas in the cluster.

  • Having the primary endpoint and reader endpoint hardcoded in your application is a bad practice since it can be changed during the TLS migration process. After the migration change to TLS-preferred is completed, query the primary endpoint and reader endpoint endpoint with the describe-replication-group API and use the DNS you get in response from this point on. This way you will be able to keep track of changes in endpoints in a dynamic way.

For CMD cluster with Automatic Failover disabled

  • Use the primary endpoint and reader endpoint instead of the per-node DNS names in your application’s code. When Automatic Failover is disabled, scaling, patching, failover, and other procedures that are managed automatically by the ElastiCache service when Automatic Failover is enabled are done by you instead. This makes it easier for you to manually keep track of the different endpoints. Since the old per-node DNS names are deleted and new ones are generated when migrating the cluster from no-TLS to TLS-preferred, do not use the per-node DNS names directly. This is mandatory so that clients will be able to connect to the cluster during the TLS-migration. Also, you’ll benefit from evenly spreading the reads between the replicas when using the reader endpoint and keep track of the DNS-records when adding or deleting replicas form the cluster.

  • Having the cluster configuration endpoint hardcoded in your application is a bad practice since it can be changed during the TLS migration process.

During the in-transit encryption: pay attention to when the migration process finishes

Change of transit encryption mode is not immediate and can take some time. This is especially true for large clusters. Only when the cluster finishes the migration to TLS-preferred is it able to accept and serve both TCP and TLS connections. Therefore, you should not create clients that will try to establish TLS connections to the cluster until the in-transit encryption is completed.

There are several ways to get notified when the in-transit encryption is completed successfully or failed: (Not shown in the code example above):

  • Using the SNS service to get a notification when the encryption is completed

  • Using the describe-events API that will emit an event when the encryption is completed

  • Seeing a message in the ElastiCache Console that the encryption is completed

You can also implement logic in your application to know if the encryption is completed. In the example above, we saw several ways to ensure the cluster finishes the migration:

  • Waiting until the migration process starts (the cluster status changes to “modifying“), and waiting until the modification is finished (the cluster status changes back to “available“)

  • Asserting that the cluster has transit_encryption_enabled set to True by querying the describe-replication-group API.

After enabling in-transit encryption: make sure the clients you use are configured properly

While the cluster is in TLS-preferred mode, your application should open TLS connections to the cluster and only use those connections. This way your application will not experience downtime when enabling in-transit encryption. You can make sure that there are no clearer TCP connections to the Redis OSS engine using the Redis OSS info command under the SSL section.

# SSL
ssl_enabled:yes
ssl_current_certificate_not_before_date:Mar 20 23:27:07 2017 GMT
ssl_current_certificate_not_after_date:Feb 24 23:27:07 2117 GMT
ssl_current_certificate_serial:D8C7DEA91E684163
tls_mode_connected_tcp_clients:0   (should be zero)
tls_mode_connected_tls_clients:100