Implementing Time Series Data

1. Using Sorted Sets for Time Series

Schema	Detail
Score	Unix timestamp (ms)
Member	Unique value (e.g. `ts:payload-id`)

ZADD metric:cpu 1716200000000 "65.4|host1"
ZADD metric:cpu 1716200060000 "67.1|host1"

Pattern	Note
Member uniqueness	Include ts in member to avoid collisions

Command	Description
`ZRANGE key from to BYSCORE`	Inclusive range
`ZRANGE key from to BYSCORE LIMIT off cnt`	Paginate

Aggregation	Approach
Sum/avg	Pull range, compute client-side or Lua
Bucketed	HINCRBY into `metric:hour:<ts>`

Strategy	Detail
Periodic job	Aggregate raw → minute → hour → day
Different keys	Separate retention per resolution

Method	Detail
`ZREMRANGEBYSCORE`	Drop entries older than threshold
Per-bucket EXPIRE	Auto-drop with TTL

Pattern	Detail
Per-hour key	`metric:cpu:2026051912` with TTL 7 days

Technique	Detail
Pipeline ZADD	Batch metric writes
Local buffer + flush	Reduce server load

Command	Description
`TS.CREATE key RETENTION ms LABELS k v`	Create series
`TS.ADD key ts value`	Insert sample (use `*` for now)
`TS.RANGE key from to AGGREGATION avg 60000`	Bucketed query
`TS.MRANGE from to FILTER label=value`	Multi-series query
`TS.CREATERULE src dst AGGREGATION sum bucketMs`	Downsampling rule