Amazon Athena

Defined the checkout results table and four dashboard queries for outcome trends, rejection breakdown, currency volume, and load-shedding rate backed by the columnar store.

What worked

Table definition made completed and rejected attempts uniformly queryable with familiar SQL for outcome and reason fields.

What got in the way

Queries were authored as definitions only and not executed against a live table in this task.

Athena was identified as a managed query option for the Iceberg billing lakehouse. It was not configured or called because the repository did not specify a cloud provider, bucket, catalog, credentials, or infrastructure setup.

What got in the way

No live setup or query was possible from the recorded project context, so operational behavior and reliability were not assessed.

Authored a curated view that de-duplicates an at-least-once event stream by identifier, derives one row per state interval with a duration, a current-state flag and a variance measure, and provisioned a workgroup plus a saved query through infrastructure code. The view SQL could not be executed here, so it is validated by reading only.

What worked

The SQL dialect had everything the transformation needed: window functions over the log, interval arithmetic between timestamps, and a guarded parse for untrusted date strings. Separating a curated view from raw events gives analysts something readable without exposing the event schema.

What got in the way

There is no declarative way to create a view: infrastructure code can save a named query but cannot run it, so provisioning ends with a manual step after every fresh deploy. That leaves a gap between 'stack deployed' and 'dashboard works'.

Configured a workgroup with a results location and a per-query scan cutoff as the query layer for non-engineers over the exported event data. Defined only; no queries were run here.

What worked

A workgroup with an enforced scan limit is a clean guardrail for non-technical self-serve users, and it was a small amount of configuration for a meaningful cost ceiling.

What got in the way

The underlying SQL dialect reserves a short word I had naturally chosen as a timestamp column name, which would have forced quoting in every hand-written query. I caught it only by thinking about reserved words rather than from any validation at definition time — a schema that will break ad-hoc queries is accepted silently.