Where readings live.
WHAT MAKES THIS DATA DISTINCTIVE
It is append-only It is queried by time range It is rarely updated Recent data is queried far more than old
WHAT A TIME-SERIES DATABASE PROVIDES
Efficient storage of timestamped values Compression exploiting regularity Fast range queries Downsampling and retention built in
WHY ORDINARY DATABASES STRUGGLE
Volume, and indexes growing without bound.
WHAT DOWNSAMPLING IS
Storing summaries at lower resolution as data ages.
WHAT TO KEEP
Full resolution recently Hourly or daily summaries for older periods Aggregates indefinitely, where storage allows
WHY
Nobody queries second-level detail from two years ago, and storing it costs continuously.
WHAT RETENTION POLICY MUST DEFINE
How long each resolution is kept.
WHAT TO AUTOMATE
Downsampling and deletion.
WHAT TAGS PROVIDE
Dimensions for filtering: device, site, type, customer.
WHAT TO BE CAREFUL WITH
High-cardinality tags, which degrade performance badly.
WHAT COUNTS AS HIGH CARDINALITY
Values with very many distinct possibilities, such as unique message identifiers.
WHAT TO NEVER USE AS A TAG
Anything unbounded.
WHAT TO MONITOR
Storage growth, and query performance as data accumulates.
