v1.11.13

Features

• External materialization — write dbt models directly to Parquet,
  CSV, or JSON files via server-side COPY statements. Because GizmoSQL
  is remote DuckDB, the COPY runs on the GizmoSQL server (typically a
  cloud VM with more CPU, memory, disk throughput, NIC bandwidth, and
  cloud-IAM reach than the dbt client) rather than streaming result sets
  back to the client just to write them out. Supports:
  • Local filesystem paths and any URI the server's DuckDB backend can
    reach: s3://, gs://, azure://, MinIO and other S3-compatible
    stores, etc.
  • Format inference from file extension, explicit format config, custom
    delimiter, and arbitrary DuckDB COPY options (compression
    codecs, partition_by, per_thread_output, ...).
  • Default {external_root}/{model_name}.{format} locations via a new
    external_root profile setting (resolved on the server).
  • ref()-able — a view is created over the written file so downstream
    models can use it like any other relation.
• New adapter-level @available helpers: external_root,
  external_write_options, external_read_location, location_exists,
  store_relation, warn_once.
• New Jinja macros: materializations/external.sql and
  utils/external_location.sql.
• plugin / glue_register options (client-side features in dbt-duckdb)
  produce a clear compile-time error — they have no analogue in a
  server-side Flight SQL adapter.
• README: new "Writing to External Files (server-side)" section with the
  config table, profile example, partitioning example, and notes on the
  parent-directory requirement for local file writes.

Test suite

• New tests/functional/adapter/test_external.py (7 test classes):
  default parquet, CSV, JSON, explicit .parquet location, explicit
  .csv location + custom delimiter, downstream ref(), empty-result
  handling, plugin/glue rejection, parquet compression codec, hive-
  partitioned writes, and an end-to-end S3 test using a MinIO sidecar on
  a user-defined docker bridge network.

Full Changelog: v1.11.12...v1.11.13