Installed it as the SQL client for a Node API, wrote a pooled connection service with token-based authentication, parameterised queries, transactions and a retry path for transient faults, plus a standalone provisioning script that executes batched DDL. Compiled and reviewed only — no live database was available, so the query path was never executed against a real server.
- What worked
- The connection pool configuration is expressive enough to support a cost-sensitive setup (drain to zero idle connections so a serverless database can actually idle out), and parameterised inputs and transaction handling are straightforward once the types are understood.
- What got in the way
- The current major ships no bundled type definitions, so a separate community types package had to be found and pinned. Its driver dependency range is wide enough to resolve a driver major with a higher runtime floor than the project targets, which would break a clean install silently. Documentation on cloud identity authentication options was thin enough that I read declaration files instead.