AuroraMeter.Schema.Event (Aurora Meter v1.0.0)

View Source

One recorded usage fact.

Two kinds of row live here, and they are not the same thing.

A row written by AuroraMeter.track/4 for a feature listed in the legacy durable_features carries no caller identity: a retry after an uncertain write inserts a second row, and nothing reads the table for billing. Those rows are given a derived event_id of "legacy:" <> id by mix aurora_meter.events.backfill, so the schema can require an identity without discarding history.

A row written by the V1 durable path carries the caller's own event_id, unique per tenant, and a payload_hash that separates a retry of the same fact from a different fact reusing the identity.

The field named inserted_at

It is the instant this node wrote the row, not the instant the usage happened. The usage instant is occurred_at, which the caller supplies. The Ecto field keeps the name inserted_at because a query written against 0.4.x (where: e.inserted_at >= ^since) must keep compiling; the public read struct AuroraMeter.Event exposes the same column as recorded_at, which is what it is.

seq

An identity column assigned by a Postgres sequence at insert time. It is the only sound scan order this table has: id is a random v4 UUID, so a forward keyset scan ordered by it can miss a row committed by a transaction that started earlier, and inserted_at is an application clock reading rather than a commit order. Never write it; Ecto is told not to.

Summary

Types

t()

Types

t()

@type t() :: %AuroraMeter.Schema.Event{
  __meta__: term(),
  attribution: term(),
  dimensions: term(),
  event_id: term(),
  feature: term(),
  id: term(),
  inserted_at: term(),
  kind: term(),
  metadata: term(),
  occurred_at: term(),
  original_event_id: term(),
  payload_hash: term(),
  period_source: term(),
  period_start: term(),
  plan_id: term(),
  plan_version: term(),
  quantity: term(),
  seq: term(),
  tenant_key: term()
}