dynamodb

Overview ¶

Package dynamodb implements the exchange.Store interface on top of DynamoDB.

Data model (single table, no secondary indexes). Items are grouped into one partition per time resolution:

pk = "rates#raw"     sk = <nanos>          rates             (every sample)
pk = "rates#hour"    sk = <hour-start>     ts, rates         (close per hour)
pk = "rates#day"     sk = <day-start>      ts, rates         (close per day)
pk = "rates#week"    sk = <week-start>     ts, rates         (close per week, Sunday-start)
pk = "rates#month"   sk = <month-start>    ts, rates         (close per month)

Every item carries the full symbol->rate map for that instant; rollup items hold the map from the most recent sample in their bucket (the "close"). The sort key is the bucket-start time as unix-nanos (a Number), which sorts in chronological order. Rollup items also store `ts`, the close sample's actual time, which both drives the monotonic last-write-wins guard and is the timestamp returned for the point. Raw items omit `ts` because their sort key already is the sample time.

Writes fan out to all resolutions in a single transaction: the raw point is a conditional Put that enforces the once-per-timestamp contract, and each rollup is a monotonic last-write-wins Update. Reads serve the resolution matching the requested interval (see resolutionForInterval): the caller chooses an interval appropriate to the range and the store honors it directly. Point lookups always read the raw resolution.