We've discovered a number of configuration / tuning knobs which we want to be different when running multi-tenancy in CC dedicated / self-hosted, vs in Serverless. (partial list below)

There should be a convenient way to initialize a secondary tenant with one set of defaults vs another.

A proposal on the table is to use "tenant templates" (like pg already has db templates).

Partial list (not exhaustive):

• tenant name
• special capabilities
• cluster settings
  • spanconfig.tenant_limit
  • kv.tenant_rate_limiter.*
  • spanconfig.tenant_coalesce_adjacent.enabled (maybe)
  • sql.schema.test_object_generator.enabled
  • sql.distsql.direct_columnar_scans.enabled (being introduced in sql,storage: add support for COL_BATCH_RESPONSE scan format #94438)
    • helps all TPC-H queries with shared process multitenancy
    • some microbenchmarks (less data processed) can get 5-10% slowdown (30% worst case)
    • seems to be most useful for OLAP-y (query accesses multiple rows, fewer columns per row than the table schema)

Jira issue: CRDB-23191
Epic: CRDB-23559