Redis — Data Types & Commands¶
Every key has one type. A command for another type fails with WRONGTYPE Operation against a key holding the wrong kind of value. Keys are created on the first write and removed when the last element is removed (an empty list, set or hash does not exist).
| Type | Think of it as | Typical use |
|---|---|---|
| String | Bytes up to 512 MB, or a number | Cache entries, counters, flags, tokens, locks |
| Hash | Small dict of string fields | Objects, sessions, per-user settings |
| List | Linked list, push / pop at both ends | Simple queues, recent items |
| Set | Unordered unique strings | Tags, "already processed" IDs, unique visitors |
| Sorted set | Unique members ordered by a float score | Leaderboards, priority queues, sliding windows |
| Stream | Append-only log of field-value entries | Event log, work queue with consumer groups |
| JSON | Nested document, JSONPath access | Documents with partial updates (built into Redis 8) |
Command examples below are redis-cli sessions; Python equivalents are in 04 — Python Client.
Strings¶
SET user:42:name "Ann" -- OK
GET user:42:name -- "Ann"
GET missing -- (nil)
SET session:abc "{...}" EX 1800 -- expires in 1800 s (PX for ms, EXAT / PXAT for a Unix time)
SET lock:job "token-1" NX PX 30000 -- OK: only if the key does not exist
SET lock:job "token-2" NX PX 30000 -- (nil): key exists, nothing changed
SET cfg:mode "v2" XX -- only if the key already exists
SET cfg:mode "v3" GET -- set and return the old value
GETEX session:abc EX 1800 -- read and refresh the TTL (sliding session)
GETDEL otp:42 -- read once and delete (one-time codes)
MSET a 1 b 2 -- several keys in one round trip
MGET a b c -- 1) "1" 2) "2" 3) (nil)
Counters¶
INCR page:views -- 1 (a missing key counts as 0)
INCRBY page:views 5 -- 6
DECR stock:sku-1 -- can go below 0 — check the result
INCRBYFLOAT balance:42 1.5 -- "1.5"
INCR user:42:name -- ERR value is not an integer or out of range
INCR is atomic: 100 clients doing INCR at the same time always end with 100. Reading a number in Python, adding 1 and writing it back loses updates — see 07 — Race Conditions.
Hashes¶
HSET user:42 name Ann email [email protected] visits 0 -- 3 (fields added)
HGET user:42 name -- "Ann"
HMGET user:42 name missing -- 1) "Ann" 2) (nil)
HINCRBY user:42 visits 1 -- 1
HGETALL user:42 -- all fields and values
HDEL user:42 email
HEXISTS user:42 email -- 0
HLEN user:42
All values are strings — visits comes back as "1", not an integer. Nested objects need JSON in a field or the JSON type.
Field-level expiry (Redis 7.4+) sets a TTL on single fields:
HEXPIRE user:42 60 FIELDS 1 visits -- 1) 1
HTTL user:42 FIELDS 2 visits name -- 1) 60 2) -1
HGETDEL user:42 FIELDS 1 name -- read and delete a field (Redis 8.0+)
Lists¶
RPUSH queue:emails a b c -- 3: push to the tail
LPOP queue:emails -- "a": pop from the head (FIFO with RPUSH)
LRANGE queue:emails 0 -1 -- all elements: "b" "c"
LLEN queue:emails -- 2
LPUSH recent:user:1 item-9 -- newest first
LTRIM recent:user:1 0 99 -- keep the last 100 (capped list)
BLPOP queue:emails 5 -- block up to 5 s waiting for an element
LMOVE queue:emails queue:processing LEFT RIGHT -- atomically move to an "in progress" list
A list is a simple queue: when a worker pops a job and crashes, the job is gone. LMOVE into a processing list, or a stream with consumer groups, lets you recover unfinished jobs.
Sets¶
SADD tags:post:1 redis python redis -- 2: duplicates are ignored
SMEMBERS tags:post:1 -- order is not defined
SISMEMBER tags:post:1 redis -- 1
SCARD tags:post:1 -- 2
SREM tags:post:1 python
SADD tags:post:2 python pytest
SINTER tags:post:1 tags:post:2 -- common members
SUNION tags:post:1 tags:post:2
SPOP raffle:tickets -- remove and return a random member
Use a set with SADD for "have I seen this ID?" checks: SADD returns 1 for a new member and 0 for a known one, atomically.
Sorted Sets¶
Members are unique, ordered by score; equal scores are ordered by member name.
ZADD lb 100 ann 250 bob 180 eve -- 3
ZINCRBY lb 50 ann -- "150"
ZRANGE lb 0 -1 REV WITHSCORES -- bob 250, eve 180, ann 150
ZRANGE lb 0 9 REV -- top 10
ZREVRANK lb ann -- 2 (0-based position from the top)
ZSCORE lb bob -- "250"
ZRANGE lb 100 200 BYSCORE -- members with 100 <= score <= 200
ZREMRANGEBYSCORE lb -inf 160 -- drop everything below 160
ZCARD lb
ZADD lb GT 120 ann -- update only if the new score is greater
Scores are 64-bit floats: exact for integers up to 2^53. With a timestamp as the score, a sorted set becomes a time index — the basis of the sliding-window rate limiter in 03.
Streams¶
A stream is an append-only log. Each entry has an ID (<ms timestamp>-<seq>) and field-value pairs. Consumer groups share the work: each entry goes to one consumer in the group and stays pending until it is acknowledged.
XADD orders * id 1 status new -- "1790789269433-0" (* = auto ID)
XADD orders MAXLEN ~ 100000 * id 2 status new -- trim to about 100k entries
XLEN orders
XRANGE orders - + COUNT 10 -- read by ID range, no group
XGROUP CREATE orders billing $ MKSTREAM -- group reads new entries only ($); 0 = from the start
XREADGROUP GROUP billing worker-1 COUNT 10 BLOCK 5000 STREAMS orders > -- > = never delivered
XACK orders billing 1790789269434-1 -- done
XPENDING orders billing -- delivered but not acked
XAUTOCLAIM orders billing worker-2 60000 0-0 -- take over entries idle for 60 s
XINFO GROUPS orders -- consumers, pending, lag
Compare with Pub/Sub and lists in 03 — Pub/Sub vs Streams.
JSON¶
Redis 8 includes the JSON type (earlier versions need the RedisJSON module or Redis Stack; Valkey has a separate module).
JSON.SET doc:1 $ '{"user":{"name":"Ann","tags":["qa"]},"visits":0}'
JSON.GET doc:1 $.user.name -- ["Ann"]
JSON.NUMINCRBY doc:1 $.visits 1 -- [1]
JSON.ARRAPPEND doc:1 $.user.tags '"python"'
If the server may be plain Redis 7 or Valkey, store JSON as a string (SET key '{"..."}') and parse it in the app — this works everywhere.
Other Types in Brief¶
| Type | Commands | Use |
|---|---|---|
| Bitmap (on a string) | SETBIT, GETBIT, BITCOUNT |
Daily active flags per user ID |
| HyperLogLog | PFADD, PFCOUNT |
Approximate unique counts in 12 KB |
| Geo (on a sorted set) | GEOADD, GEOSEARCH |
"Stores within 5 km" |
| Probabilistic, time series, vector sets | BF.*, TS.*, VADD / VSIM |
Bundled in Redis 8 |
TTL & Expiry Rules¶
SET k v EX 100 -- TTL 100
TTL k -- 100
SET k v2 -- plain SET removes the TTL
TTL k -- -1
SET k v3 EX 100
SET k v4 KEEPTTL -- change the value, keep the TTL
EXPIRE k 60 NX -- only if there is no TTL yet (Redis 7.0+; also XX, GT, LT)
PERSIST k -- remove the TTL
TTL missing -- -2
| Rule | Consequence |
|---|---|
SET without EX / KEEPTTL clears the TTL |
A cache refresh written as SET key value creates a key that never expires |
INCR, HSET, RPUSH, SADD keep the existing TTL |
A counter created by INCR has no TTL until you add one |
EXPIRE with 0 or a negative value deletes the key |
Useful to expire a key "now" in tests |
| TTL is on the whole key (except hash fields in 7.4+) | You cannot expire one list element or set member |
RENAME moves the TTL with the value |
|
| Expired keys are removed lazily on access plus by a background sampler | An expired key is never returned, but memory is freed a bit later |
| TTLs use the server clock | Changing the clock in your test process does not affect a real Redis |
The last rule matters for testing: the app computes values like "window number" on the client, but expiry happens on the server. 07 — Testing TTL shows how to test both.