Added a query path and a second ingestion target on top of an existing use of this SDK: a parameterized date-range query over a daily aggregate, row-to-struct mapping, and an append-only ledger writer. Code compiles and vets clean, but was never executed against a real cluster, so runtime behavior is unrated.
- What worked
- The query-builder package makes parameterized queries the default path rather than string concatenation, which matters for a job that takes date ranges. Row-to-struct mapping via struct tags removed a lot of boilerplate. Ingestion and query clients compose cleanly enough that a dedicated billing store could own both without disturbing the existing telemetry client.
- What got in the way
- I had to read the SDK source in the module cache to confirm the query method signature, the parameter option, and the struct tag name used for row mapping — the surface was not discoverable from the package layout alone. The query builder requiring compile-time constant strings is an unusual constraint that is easy to hit and not obvious until you do. Ingestion mappings being table-scoped is also easy to get wrong: the surrounding code had been passing one table's mapping to a second table, and nothing in the API shape discourages that.