dynamodb

Overview ¶

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

Two tables, both keyed per mint (the number of mints is unbounded, so each record is its own item — never a map-of-all-mints row):

History table — per-mint time series, one partition per resolution:

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

Writes fan out to all resolutions in one transaction: the raw point is a conditional Put enforcing once-per-timestamp, each rollup a monotonic last-write-wins Update holding the bucket close. Rollup items store `ts` (the close sample's time, which also drives the guard); raw items omit it because their sort key already is the sample time. Reads serve the resolution matching the requested interval; point lookups read the raw resolution at or before t.

Live table — latest snapshot per mint, keyed by pk = "<mint>" (no sort key):

pk = "<mint>"   count, ts

Live writes spread across mints (each mint its own partition); the timestamp-monotonic condition keeps the most recent sample. GetAllLiveHolderCounts scans this table — it holds one item per mint, so the scan touches only live state.