Defined a database and a table with an explicit column schema and date-based partition projection so no crawler or partition registration is needed, then checked the projection parameters in the synthesized template.
- What worked
- Partition projection removes a whole moving part — no crawler, no partition repair job — and a date-partitioned layout with the tenant key left as a regular column keeps both per-tenant and cross-tenant queries workable.
- What got in the way
- Only the low-level construct was available, so the table is a bag of stringly-typed properties with no validation: projection settings, serialization library and input/output formats all have to be exactly right and a typo would surface only at query time. One projection value also needs a literal placeholder that collides with host-language string interpolation and must be escaped.
