Cheat Sheet
Redis Commands
Commands grouped by data type with the complexity from the official reference and the gotcha that bites in production. Read the complexity column first: it tells you what is safe on a big key.
11
Groups
Redis docs
Complexity source
redis-cli
Options checked
01
Chapter 1 - How to read this sheet
Complexity Is the Point
Redis runs commands on one thread, so one slow command delays every client. The Complexity column is copied from the command reference (redis.io/docs/latest/commands, data file commands.json in the redis/docs repository, checked 2026-10-31). N is almost always the size of the key or the number of items returned. Anything O(N) on a large key is a latency risk.
| Complexity | Meaning for you |
|---|---|
| O(1) | Safe at any size |
| O(log N) | Safe; sorted sets and similar |
| O(N) with N = items returned | Safe if you page (COUNT, LIMIT); dangerous when you ask for everything |
| O(N) with N = size of the key or keyspace | Avoid on big data in production (KEYS, SMEMBERS, HGETALL, LRANGE 0 -1) |
02
Chapter 2 - Strings and generic keys
Strings, Counters, Key Management
Strings hold text, numbers or bytes (up to the string size limit in the docs). Counters use INCR.
| Command | What it does | Complexity | Gotcha |
|---|---|---|---|
SET k v [EX s | PX ms] [NX | XX] [GET] [KEEPTTL] | Set a value with optional expiry and conditions | O(1) | SET clears an existing TTL unless KEEPTTL; NX + EX is the basic lock primitive; GET option since 6.2 |
GET k | Read a string | O(1) | Returns nil for a missing key; WRONGTYPE error for other types |
SETNX, SETEX, PSETEX | Older set variants | O(1) | Marked deprecated in the command reference; use SET options |
MSET / MGET | Set / get many keys | O(N), N = number of keys | In Redis Cluster all keys must share one hash slot |
INCR / INCRBY / DECR | Atomic integer counter | O(1) | Value must be an integer string; creates the key at 0 if missing |
INCRBYFLOAT | Atomic float add | O(1) | Float rounding applies |
APPEND k v | Append to a string | O(1) amortised | Good for small appends |
GETDEL k | Read and delete atomically | O(1) | Since 6.2 |
GETEX k EX s | Read and change expiry | O(1) | Since 6.2; sliding-expiry cache pattern |
STRLEN / GETRANGE / SETRANGE | Length / substring / overwrite part | O(1) / O(N) returned / O(1) or O(M) copied | Offsets are in bytes |
EXISTS k [k ...] | Count how many keys exist | O(N), N = keys checked | Passing the same key twice counts twice |
DEL k [k ...] | Delete keys | O(N) keys; O(M) per big collection | Deleting a huge collection blocks the server |
UNLINK k [k ...] | Delete without blocking (memory freed in another thread) | O(1) per key | Since 4.0; prefer over DEL for big keys |
TYPE k | Data type of the key | O(1) | Returns none if missing |
RENAME a b | Rename a key | O(1) | Overwrites b; in Cluster both keys need the same slot |
COPY src dst | Copy a key | O(N) for collections | Since 6.2 |
SCAN cursor [MATCH p] [COUNT n] [TYPE t] | Iterate keys without blocking | O(1) per call, O(N) for a full pass | May return duplicates; a key present throughout is returned at least once; loop until the cursor is 0 |
KEYS pattern | List all matching keys | O(N) over the whole keyspace | Never in production; use SCAN |
OBJECT ENCODING k / OBJECT FREQ k | Internal encoding / access frequency (LFU policies) | O(1) | OBJECT FREQ needs an LFU maxmemory-policy |
MEMORY USAGE k | Bytes a key uses | O(N), N = samples | Since 4.0; includes overhead |
03
Chapter 3 - Hashes and lists
Hashes (objects) and Lists (queues)
Hashes model an object with fields. Lists are linked sequences, good at the two ends.
| Command | What it does | Complexity | Gotcha |
|---|---|---|---|
HSET k f v [f v ...] | Set one or more fields | O(1) per pair | Multiple pairs accepted since 4.0 (replaces HMSET) |
HGET k f, HMGET k f1 f2 | Read one / several fields | O(1) / O(N) fields | Missing field returns nil |
HGETALL k | All fields and values | O(N), N = hash size | Avoid on big hashes; use HSCAN |
HDEL k f, HLEN k, HKEYS k | Remove field, count, list names | O(N) removed / O(1) / O(N) | HKEYS is also O(N) of the whole hash |
HINCRBY k f n | Atomic counter in a field | O(1) | Per-object counters without read-modify-write |
HSCAN k cursor [MATCH p] [COUNT n] | Iterate fields | O(1) per call | Same duplicate caveat as SCAN |
HEXPIRE k seconds FIELDS n f ... | Expire individual hash fields | O(N), N = fields | Hash field expiration listed since 7.4.0 in the reference |
LPUSH / RPUSH k v ... | Add to head / tail | O(1) per element | Order of several values: LPUSH a b c leaves c first |
LPOP / RPOP k [count] | Remove from head / tail | O(N), N = elements returned | The count form needs 6.2+ |
LRANGE k start stop | Read a range | O(S+N) | LRANGE k 0 -1 returns everything: avoid on long lists |
LINDEX k i, LSET k i v | Read / write by index | O(N) to the index (O(1) at either end) | Middle access is slow on long lists |
LLEN k | Length | O(1) | |
LTRIM k start stop | Keep only a range | O(N) removed | LPUSH + LTRIM 0 99 = capped recent list |
LREM k count v | Remove matching values | O(N+M) | Scans the list |
LMOVE src dst LEFT RIGHT | Atomically move an element between lists | O(1) | Since 6.2; reliable-queue building block (replaces RPOPLPUSH) |
BLPOP k [k ...] timeout | Blocking pop | O(N) keys | Timeout 0 blocks forever; blocking commands hold the client connection |
04
Chapter 4 - Sets and sorted sets
Sets, Sorted Sets, Probabilistic Types
Sets: unique members. Sorted sets: unique members ordered by a score (leaderboards, delay queues, rate-limit windows).
| Command | What it does | Complexity | Gotcha |
|---|---|---|---|
SADD / SREM k m ... | Add / remove members | O(1) per add / O(N) removed | Returns the number actually added |
SISMEMBER k m, SMISMEMBER k m1 m2 | Membership test | O(1) / O(N) tested | SMISMEMBER since 6.2 |
SMEMBERS k | All members | O(N), N = set size | Use SSCAN for big sets |
SCARD k | Size | O(1) | |
SINTER / SUNION / SDIFF | Set algebra | O(N*M) worst case / O(N) / O(N) | Large sets block; ...STORE variants save the result |
SPOP / SRANDMEMBER k [count] | Remove / read random members | O(1) without count, O(N) with | |
ZADD k [NX|XX] [GT|LT] [CH] score m | Add or update a member's score | O(log N) per item | GT / LT since 6.2; CH counts changed members |
ZRANGE k start stop [BYSCORE|BYLEX] [REV] [LIMIT o c] [WITHSCORES] | Range by rank, score or lexicographic order | O(log N + M) | The unified form since 6.2; older ZRANGEBYSCORE and ZREVRANGE are deprecated in the reference |
ZRANK k m, ZSCORE k m | Rank / score of a member | O(log N) / O(1) | Rank is 0-based |
ZINCRBY k n m | Add to a score | O(log N) | Atomic leaderboard update |
ZCARD / ZCOUNT k min max | Size / count in a score range | O(1) / O(log N) | Use ( for exclusive bounds: (5 |
ZREM k m | Remove members | O(M log N) | |
ZPOPMIN / ZPOPMAX k [count] | Pop lowest / highest scores | O(log N * M) | Basis of a simple priority or delay queue |
ZREMRANGEBYRANK / ...BYSCORE | Trim a sorted set | O(log N + M) | Sliding-window rate limiting: remove old scores |
PFADD / PFCOUNT | HyperLogLog: approximate distinct count | O(1) | Fixed small memory; approximate by design |
SETBIT / GETBIT / BITCOUNT | Bit arrays | O(1) / O(1) / O(N) | Compact flags per user id |
GEOADD / GEOSEARCH | Geospatial index on a sorted set | O(log N) per item / O(N+log M) | Radius searches cost grows with results |
05
Chapter 5 - Streams, pub/sub, scripting
Streams, Pub/Sub, Transactions and Scripts
Streams are an append-only log with consumer groups (a lightweight Kafka-like model).
| Command | What it does | Complexity | Gotcha |
|---|---|---|---|
XADD s * f v [MAXLEN ~ n] | Append an entry (* = auto id) | O(1), O(N) when trimming | ~ trims approximately and is cheaper |
XLEN s, XRANGE s - + COUNT n | Length / read a range | O(1) / O(N) returned | Always pass COUNT |
XREAD COUNT n BLOCK ms STREAMS s id | Read new entries | See the reference (depends on streams) | $ means "only entries after now" |
XGROUP CREATE s g $ [MKSTREAM] | Create a consumer group | Depends on subcommand | Group needs an existing stream or MKSTREAM |
XREADGROUP GROUP g c COUNT n STREAMS s > | Read as a group member | O(M) returned | > means undelivered entries; an id such as 0 replays your pending ones |
XACK s g id | Acknowledge an entry | O(1) per id | Unacked entries stay in the pending list |
XPENDING s g | Inspect unacknowledged entries | O(N) returned | Find stuck consumers |
XAUTOCLAIM s g c min-idle 0 | Take over idle pending entries | O(1) with small COUNT | Reply gained an element in 7.0 |
XTRIM / XDEL | Trim by length or id / delete entries | O(N) evicted / O(1) each | XDEL removes entries by id; use XTRIM to bound stream size |
PUBLISH ch msg, SUBSCRIBE ch | Fire-and-forget messaging | O(N+M) / O(N) | Subscribers that are offline miss messages; use streams for durability |
SPUBLISH / SSUBSCRIBE | Sharded pub/sub for Cluster | O(N) subscribers | Since 7.0 |
MULTI ... EXEC | Queue commands and run them as one atomic block | Depends on commands | No rollback: a command that errors at run time does not undo the others |
WATCH k ... MULTI ... EXEC | Optimistic locking | O(1) per key | If a watched key changes, EXEC returns nil: retry |
EVAL script numkeys k ... arg ..., EVALSHA | Run a Lua script atomically | Depends on the script | Declare keys in KEYS, not in the script text; long scripts block |
FUNCTION LOAD, FCALL | Redis Functions (server-side libraries) | O(1) load, depends on function | Since 7.0; persisted with the dataset |
06
Chapter 6 - TTL and eviction
Expiry and Memory Limits
Expiry removes keys by time; eviction removes keys when memory is full. Different mechanisms.
| Command | What it does | Complexity | Gotcha |
|---|---|---|---|
EXPIRE k s [NX|XX|GT|LT] | Set a TTL in seconds | O(1) | Options since 7.0; PEXPIRE for milliseconds, EXPIREAT for a Unix time |
TTL k / PTTL k | Remaining time | O(1) | -1 = no expiry, -2 = key does not exist |
PERSIST k | Remove the TTL | O(1) | |
EXPIRETIME k | Absolute expiry time | O(1) | Since 7.0 |
| Overwriting a key | SET removes the TTL; INCR, LPUSH and similar keep it | n/a | Re-apply the TTL after rewrites |
CONFIG SET maxmemory 2gb | Memory limit | O(N) parameters | Without a limit Redis can grow until the OS kills it |
maxmemory-policy noeviction | Writes fail with an error when full | n/a | Default policy per the eviction docs |
allkeys-lru / allkeys-lfu / allkeys-random | Evict any key by recency / frequency / randomly | n/a | The docs call allkeys-lru a good default when you have no other preference; LRU is approximate (sampling) |
volatile-lru / volatile-lfu / volatile-random / volatile-ttl | Evict only keys that have a TTL | n/a | With no TTL keys they behave like noeviction |
allkeys-lrm / volatile-lrm | Least recently modified policies | n/a | Listed in the current docs; not in older releases, check your version |
INFO memory, INFO stats | used_memory, evicted_keys, expired_keys | O(1) | Rising evicted_keys means the cache is too small |
07
Chapter 7 - Persistence and replication
RDB, AOF, Replicas
Persistence choices checked against the Redis persistence documentation.
| Item | What it does | Complexity | Gotcha |
|---|---|---|---|
RDB (save <seconds> <changes>) | Periodic point-in-time snapshot to dump.rdb | n/a | You can lose writes since the last snapshot |
BGSAVE | Snapshot in a child process | O(1) to start | Fork can pause the main thread on very large datasets; SAVE blocks and is O(N) |
LASTSAVE | Time of the last successful save | O(1) | Use it to confirm snapshots happen |
AOF (appendonly yes) | Log of every write, replayed on start | n/a | Since 7.0 AOF is split into several files in the appenddirname directory |
appendfsync everysec | Fsync once per second (the suggested default) | n/a | You can lose about one second of writes |
appendfsync always | Fsync on every write batch | n/a | Safest and slowest |
appendfsync no | Let the OS flush | n/a | Fastest, least safe |
BGREWRITEAOF | Compact the AOF in the background | O(1) to start | Rewrite uses disk and memory while running |
WAIT n timeout, WAITAOF | Block until replicas or local AOF acknowledge the writes | O(1) | WAITAOF since 7.2; acknowledgement still is not a full consensus guarantee |
REPLICAOF host port / REPLICAOF NO ONE | Make this node a replica / promote it | O(1) | Old name SLAVEOF |
ROLE, INFO replication | Show role, offsets, lag | O(1) | Compare master_repl_offset with replica offsets |
FAILOVER | Coordinated switchover to a replica | O(1) | Since 6.2 |
08
Chapter 8 - Cluster and operations
Cluster, Diagnostics, Admin
Cluster spreads keys over 16384 hash slots (stated in the cluster documentation); a key's slot is a hash of its name, and {tag} parts force related keys into one slot.
| Command | What it does | Complexity | Gotcha |
|---|---|---|---|
CLUSTER INFO | Cluster state, slots assigned, known nodes | O(1) | cluster_state:ok is what you want |
CLUSTER NODES | All nodes with ids, roles, slot ranges, flags | O(N) nodes | fail and handshake flags show trouble |
CLUSTER SHARDS | Shard-oriented view (since 7.0) | O(N) nodes | Replaces CLUSTER SLOTS, which is deprecated in the reference |
CLUSTER KEYSLOT key | Which slot a key maps to | O(N) bytes in key | Check hash-tag usage |
CLUSTER COUNTKEYSINSLOT slot | Keys in a slot | O(1) | Useful during resharding |
CLUSTER MEET ip port | Introduce a node | O(1) | Usually done by redis-cli --cluster |
CLUSTER FAILOVER | Manual replica takeover (run on the replica) | O(1) | Safer than killing the master |
MOVED / ASK errors | Key lives on another node / slot is migrating | n/a | Use a cluster-aware client or redis-cli -c |
INFO [section] | Server metrics | O(1) | Sections: server, clients, memory, persistence, stats, replication, cpu, keyspace |
DBSIZE | Number of keys in the selected database | O(1) | |
SLOWLOG GET 10 | Recent slow commands | O(N) returned | Threshold set by slowlog-log-slower-than |
LATENCY DOCTOR | Human-readable latency report | O(1) | Needs the latency monitor threshold configured |
MEMORY DOCTOR | Memory problem hints | O(1) | |
CLIENT LIST / CLIENT KILL | List / disconnect clients | O(N) clients | Look for idle or blocked clients |
MONITOR | Stream every command received | n/a | Heavy; reduces throughput, never leave it running; also exposes data |
CONFIG GET / CONFIG SET / CONFIG REWRITE | Read, change, persist configuration | O(N) params / O(1) | CONFIG SET is lost on restart unless rewritten |
FLUSHDB / FLUSHALL [ASYNC] | Delete all keys in one DB / all DBs | O(N) | No undo; many deployments restrict it with ACLs |
ACL SETUSER, AUTH, HELLO | Users, authentication, protocol negotiation | O(N) rules / O(N) passwords / O(1) | HELLO 3 switches to RESP3 |
SHUTDOWN [NOSAVE|SAVE] | Stop the server | O(N) when saving | Default saves if persistence is on |
09
Chapter 9 - redis-cli
redis-cli Options
Option names checked against redis-cli.c in the Redis source (unstable branch); redis-cli --help shows your build's list.
| Option | What it does | Gotcha |
|---|---|---|
-h host -p port | Connect to a host and port | Defaults to 127.0.0.1:6379 |
-u redis://user:pass@host:6379/0 | Connect with a URI | Passwords on the command line show in the process list |
-a password / --user name --pass pw / --askpass | Authenticate | --askpass prompts so nothing is left in history |
-n 2 | Select a database | Cluster mode has database 0 only |
-c | Follow MOVED and ASK redirects | Needed for Cluster |
--tls, --insecure | Use TLS; skip certificate verification | --insecure only for tests |
-3 | Use RESP3 | Output format changes |
--raw / --no-raw | Raw or formatted output | Raw is the default when output is not a terminal |
--csv, --json | Machine-readable output | Handy for scripts |
-x | Read the last argument from stdin | For setting binary or multi-line values |
-r 10 -i 1 cmd | Repeat a command 10 times, 1 second apart | -r -1 repeats forever |
--scan --pattern "user:*" [--count n] | Iterate keys with SCAN | Safe alternative to KEYS |
--bigkeys, --memkeys, --keystats | Sample the keyspace for the largest keys / memory use / statistics | Uses SCAN; --memkeys and --keystats exist in newer builds |
--hotkeys | Find hot keys | Works only with an LFU eviction policy |
--latency, --latency-history, --latency-dist | Measure latency (live, over time, as a distribution) | Measures from the client, includes the network |
--intrinsic-latency 5 | Measure system latency on the machine running it | Run on the Redis host, not remotely |
--stat | Rolling server statistics | A quick top for Redis |
--rdb file | Download an RDB snapshot from the server | Triggers a save on the server |
--pipe | Mass insert from a stream of protocol text | Much faster than one command per line |
--cluster create|check|reshard ... | Cluster administration | Manage with care; --cluster-yes auto-confirms |
--eval script.lua keys , args | Run a Lua script file | A comma separates keys from arguments |
10
Chapter 10 - Common mistakes
Redis Mistakes
⚠ The ones that cost real time
- **
KEYS *in production.** O(N) over the keyspace, blocks everyone. UseSCAN. - **
HGETALL,SMEMBERS,LRANGE 0 -1on big keys.** Page withHSCAN,SSCAN, ranges. - **Caching without a TTL and without
maxmemory.** The instance grows until it is killed. Set both. - **Assuming
SETkeeps the TTL.** It clears it unless you passKEEPTTL. - Using pub/sub where messages must not be lost. Offline subscribers miss them; use Streams.
- **Treating
MULTI/EXEClike a database transaction.** There is no rollback. - Multi-key commands across slots in Cluster. Use hash tags like
{user:1}:cart. - **Big deletes with
DEL.** UseUNLINK. - **
MONITORleft running.** It slows the server and leaks data. - **One lock key with
SETNXand no expiry.** A crashed holder blocks everyone forever; useSET k v NX EX 30and a unique value.
11
Chapter 11 - Go deeper
Long-form pages and what was checked
| Topic | Page |
|---|---|
| Redis data model, persistence and clustering in depth | Redis |
| Caching strategies and invalidation | Caching Patterns |
| Choosing a database | NoSQL Fundamentals |
| Redis as a Spring cache | Caching & Resilience |
| Cache performance | Performance: Caching |
Version note
Checked on 2026-10-31: every complexity string and "since" version in this sheet against the machine-readable command reference (commands.json in the redis/docs repository, which backs redis.io/docs/latest/commands); deprecation marks for SETNX, SETEX, ZRANGEBYSCORE, ZREVRANGE and CLUSTER SLOTS from the same file; eviction policies and the noeviction policy from the Redis eviction documentation (the current docs also list the lrm policies, which are newer); persistence behaviour (RDB, appendfsync modes, multi-part AOF since 7.0) from the persistence documentation; redis-cli option names from redis-cli.c (unstable). No Redis server was available locally, so no command was executed. The 16384-slot figure and the hash-tag rule are standard Redis Cluster facts not re-fetched for this sheet.