Redis Development Guidelines: Key Design, Command Usage, and Best Practices

Key Design

Key Name Design

Use a business name or database name as a prefix to avoid key collisions, separating parts with colons, e.g., ugc:video:1 or business:table:id .

Readability and Manageability

Choose readable prefixes and keep keys concise; long keys increase memory usage. Example of shortening:

user:{uid}:friends:messages:{mid} can be simplified to u:{uid} .

Simplicity

Avoid special characters such as spaces, newlines, quotes, or escape characters.

Value Design

Avoid Big Keys

Keep string values under 10KB; limit hash, list, set, and sorted‑set elements to no more than 5,000. Large collections (e.g., a list with two million items) should be avoided.

For non‑string big keys, use incremental deletion with hscan , sscan , or zscan instead of DEL , and be aware of expiration‑triggered deletions.

Select Appropriate Data Types

Use the most suitable Redis data structure and configure memory encoding (e.g., ziplist) while balancing memory savings and performance.

Bad example (using separate strings for each field):

set user:1:name tom
set user:1:age 19
set user:1:favor football

Good example (using a hash):

hmset user:1 name tom age 19 favor football

Control Key Lifecycle

Set expiration times with EXPIRE (consider staggering expirations to avoid spikes). For non‑expiring data, monitor IDLETIME closely.

Command Usage

Beware O(N) Commands

Commands like HGETALL , LRANGE , SMEMBERS , ZRANGE , and SINTER traverse N elements; use their scan counterparts ( HSCAN , SSCAN , ZSCAN ) when N can be large.

Prohibited Commands

Avoid using KEYS , FLUSHALL , FLUSHDB in production; disable them via Redis rename mechanism or replace with safe alternatives like SCAN .

Selective Use of SELECT

Redis databases are weakly isolated; many clients have limited support, and using multiple databases can cause interference in a single‑threaded environment.

Batch Operations for Efficiency

Native batch commands such as MGET and MSET .

Use pipelines for non‑native batch operations, but keep batch size reasonable (e.g., ≤500 items, depending on payload size).

Note that native batch commands are atomic, while pipelines are not; pipelines can combine different commands but require client and server support.

Limit Use of Transactions

Redis transactions lack rollback and, in clustered mode, all keys in a transaction must reside in the same slot (use hashtags to co‑locate keys if needed).

Lua Scripting in Cluster Mode

All keys used in a Lua script must be passed via the KEYS array, and they must belong to the same slot; otherwise the script fails with an error.

MONITOR Command

Use MONITOR only when necessary and avoid long‑running sessions.

By following these guidelines, developers can minimize Redis‑related issues, improve performance, and maintain a clean, manageable data store.