Major version behavior and compatibility differences with Redis OSS

MAJOR versions are for API incompatible changes

MINOR versions are for new functionality added in a backwards-compatible way

PATCH versions are for backwards-compatible bug fixes and non-functional changes

SCRIPT LOAD and SCRIPT FLUSH are no longer propagated to replicas. If you need to have some durability for scripts, we recommend you consider using Redis OSS functions.

Pubsub channels are now blocked by default for new ACL users.

STRALGO command was replaced with the LCS command.

The format for ACL GETUSER has changed so that all fields show the standard access string pattern. If you had automation using ACL GETUSER, you should verify that it will handle either format.

The ACL categories for SELECT, WAIT, ROLE, LASTSAVE, READONLY, READWRITE, and ASKING have changed.

The INFO command now shows command stats per sub-command instead of in the top level container commands.

The return values of LPOP, RPOP, ZPOPMIN and ZPOPMAX commands have changed under certain edge cases. If you use these commands, you should check the release notes and evaluate if you are impacted.

The SORT and SORT_RO commands now require access to the entire keyspace in order to use the GET and BY arguments.

The ACL flags of the TIME, ECHO, ROLE, and LASTSAVE commands were changed. This may cause commands that were previously allowed to be rejected and vice versa.

When upgrading from Redis OSS 6.0, the ordering of key/value pairs returned from a map response to a lua script are changed. If your scripts use redis.setresp() or return a map (new in Redis OSS 6.0), consider the implications that the script may break on upgrades.

The maximum number of allowed databases has been decreased from 1.2 million to 10 thousand. The default value is 16, and we discourage using values much larger than this as we’ve found performance and memory concerns.

Set AutoMinorVersionUpgrade parameter to yes, and ElastiCache (Redis OSS) will manage the minor version upgrade through self-service updates. This will be handled through standard customer-notification channels via a self-service update campaign. For more information, see Self-service updates in ElastiCache.

Scripts are by replicated by effects instead of re-executing the script on the replica. This generally improves performance but may increase the amount of data replicated between primaries and replicas. There is an option to revert back to the previous behavior that is only available in ElastiCache (Redis OSS) 5.0.

If you are upgrading from Redis OSS 4.0, some commands in LUA scripts will return arguments in a different order than they did in earlier versions. In Redis OSS 4.0, Redis OSS would order some responses lexographically in order to make the responses deterministic, this ordering is not applied when scripts are replicated by effects.

Inn Redis OSS 5.0.3 and above, ElastiCache (Redis OSS) will offload some IO work to background cores on instance types with more than 4 VCPUs. This may change the performance characteristics Redis OSS and change the values of some metrics. For more information, see Which Metrics Should I Monitor? to understand if you need to change which metrics you watch.

Slow log now logs an additional two arguments, the client name and address. This change should be backwards compatible unless you are explicitly relying on each slow log entry containing 3 values.

The CLUSTER NODES command now returns a slight different format, which is not backwards compatible. We recommend that clients don’t use this command for learning about the nodes present in a cluster, and instead they should use CLUSTER SLOTS.

There are no compatibility changes to call out for this version.

Starting in Redis OSS 2.8.22, Redis OSS AOF is no longer supported in ElastiCache (Redis OSS). We recommend using MemoryDB when data needs to be persisted durably.

Starting in Redis OSS 2.8.22, ElastiCache (Redis OSS) no longer supports attaching replicas to primaries hosted within ElastiCache. While upgrading, external replicas will be disconnected and they will be unable to reconnect. We recommend using client-side caching, made available in Redis OSS 6.0 as an alternative to external replicas.

The TTL and PTTL commands now return -2 if the key does not exist and -1 if it exists but has no associated expire. Redis OSS 2.6 and previous versions used to return -1 for both the conditions.

SORT with ALPHA now sorts according to local collation locale if no STORE option is used.