Thread by @antirez

Redis 5 thread! What's new? Why I call it Redis 5? Lets start from this last question. Redis future versions will be called Redis 5, Redis 6, Redis 7 and so forth. We release every 1 / 1.5 years more or less, there is no point in saying 3.2 or 4.4.

However versioning schema will not change, so like we did for 4.0 having 4.0.1, 4.0.2, we'll do the same for 5.0.x. Also as usually release candidates of 5 start at version 4.9.101 for RC1, 4.9.102 for RC2 and so forth. So this is "Redis 5" without too many other attributes.

What's new in Redis 5?

1) The star of the release is... the new streams data type with the consumer groups. If you don't know what they are we have a good introduction here that will show you the important parts: redis.io/topics/streams…

Redis Streams were also covered here: brandur.org/redis-streams even if back then when @brandur wrote this great blog post, consumer groups were totally lacking from the implementation, so the new story is a bit more complex, but the new features are useful!

2) Redis modules have two new set of APIs: timers and cluster. The Cluster API is not about "Redis Cluster", is a cluster message bus abstraction so that you can write distributed systems with Redis. I'm re-writing Disque on top of those new APIs, as a Redis module.

3) RDB was modified in order to include LFU/LRU metadata alongside with keys. This means that if you restart an instance with RDB, the eviction process will not start at random, but will still use clues from the last execution.

4) Redis 5 moves away from the redis-trib.rb Ruby script to manage the cluster. Now the cluster manager is inside redis-cli itself. Try redis-cli --cluster help, to get more info. It is faster, does not require any dependency (like Redis...), and we are going to improve it more.

5) Sorted sets are now like lists! They have pop-style and block operations: ZPOPMAX ZPOPMIN BZPOPMAX BZPOPMIN. They are documented: redis.io/commands/bzpop… and redis.io/commands/zpopm…

6) The active defragmentation feature now works better, even if currently cannot be used with Streams (just a matter of the next RC release...). It was improved in many ways, but is currently disabled because we upgraded to Jemalloc 5.1, and we need to re-write the hooks.

7) HyperLogLogs now use a better and faster algorithm. Now they are implemented in terms of this paper: arxiv.org/abs/1702.0128 (the PR was submitted by the author of the paper in person :-D

8) Better memory reporting capabilities. We have now different favors of memory usage and fragmentation reporting that use the allocator internals, and are used both for INFO output and for MEMORY DOCTOR.

9) Now many commands that have sub commands, have a special HELP command! So instead of going nuts remembering, you just type "OBJECT HELP" for instance, and it shows you the subcommands available. Usability is important and programmers do not deserve terrible APIs.

10) Many other random things: speed improvements, Jemalloc was updated, a number of fixes (many backported into Redis 4). And so forth.

Thanks to @RedisLabs for sponsoring my work.
A special thank you to the following contributors.

Fabio Nicotra
Soloestoy
Itamar Haber
Oran Agra
Dvir Volk
dejun.xdj
Guy Benoish
Charsyam
Otmar Ertl
Jan-Erik Rediger
Spinlock

(Ordered by number of commits).