Redis.

Redis supports these core data types, each with its own commands:

• Strings: text, numbers or binary data, up to 512MB.
• Hashes: field-value pairs, like an object.
• Lists: strings in insertion order.
• Sets: unique strings with no order.
• Sorted sets, or ZSets: unique strings ordered by a score.
• Bitmaps: a string used as an array of bits.
• HyperLogLog: a probabilistic structure that estimates cardinality, the number of unique items.
• Geospatial indexes: longitude and latitude points you search by radius.
• Streams: append-only logs of events, used as message queues.

Redis 8 adds JSON, time series and probabilistic types such as Bloom filters. It also adds vector sets for similarity search.

SET mykey "hello"          # String
HSET user:1 name "John"    # Hash
LPUSH mylist "item1"       # List
SADD myset "member1"       # Set
ZADD myzset 1 "item1"      # Sorted Set

Redis keeps its data in memory as keys holding data structures, while a relational database keeps tables on disk and answers SQL queries. Redis is far faster for simple reads and writes, and a relational database gives you joins, a fixed schema and full transactions.

Redis suits caching, sessions and real-time work such as counters and leaderboards. A relational database suits complex business logic and data with many relationships. Many systems use both. The relational database stays the source of truth, and Redis holds a fast copy of the data read most often.

Redis saves its data to disk as RDB snapshots or an AOF log, or both. A pure cache can turn persistence off.

Redis Database (RDB) saves a point-in-time snapshot at set intervals. Redis forks a child process, a copy of itself, which writes one compact binary file. That file suits backups, disaster recovery and fast restarts. After a crash you lose the writes since the last snapshot, often several minutes.

Append Only File (AOF) logs every write operation, and Redis replays it at startup. The default appendfsync everysec risks about one second of writes, and always syncs every write but is slow. The file is larger and Redis restarts from it more slowly, so Redis rewrites it in the background.

With both on, Redis restores from the AOF, the more complete copy. The AOF starts from a base file, a snapshot the logged writes follow. By default that base uses the RDB format, so it loads fast.

# RDB configuration
# Save if at least 1 key changed in 900 seconds
save 900 1
# Save if at least 10 keys changed in 300 seconds
save 300 10

# AOF configuration
appendonly yes
# Options: always, everysec, no
appendfsync everysec