DuckDB iceberg_rest CTAS table materialization fails with __dbt_tmp already exists

[dataders]

Summary

DuckDB iceberg_rest catalog attach works far enough for dbt seed, but dbt run table materialization fails during CTAS temp relation creation.

This blocks using an Iceberg REST catalog as a normal dbt model destination in the catalogs.yml v2 DuckDB flow.

related: duckdb/duckdb-iceberg#595

Repro

Using the dbt-fusion binary built from the dbt-labs/fs catalogs.yml v2 stack:

• dbt-labs/fs#9420
• dbt-labs/fs#9587

Configure an Iceberg REST catalog against local Lakekeeper:

catalogs:
  - name: iceberg_demo
    type: iceberg_rest
    table_format: iceberg
    config:
      duckdb:
        endpoint: "http://localhost:8181/catalog"
        warehouse: "demo"
        authorization_type: "NONE"
        access_delegation_mode: "NONE"
        attach_as: "iceberg_demo"

Then add a table model targeting that catalog:

{{ config(
    materialized='table',
    database='iceberg_demo'
) }}

select
  4 as demo_order,
  'iceberg_rest' as catalog_name,
  'lakekeeper + minio' as storage_type

Run:

RUST_LOG=debug /path/to/fs/target/debug/dbt run \
  --project-dir /path/to/jaffle-sandbox \
  --profiles-dir /path/to/jaffle-sandbox \
  --target catalog_showcase \
  --target-path /tmp/jaffle-target-9587 \
  --log-path /tmp/jaffle-logs-9587 \
  --select iceberg_rest_catalog_demo

Actual

The model fails while creating the dbt temp relation:

Invalid Configuration Error: Table iceberg_rest_catalog_demo__dbt_tmp already exists

Expected

materialized='table' should create or replace the target Iceberg table, or the adapter/materialization path should use an Iceberg-safe temp relation flow.

Notes

• The same Iceberg REST catalog can be written with dbt seed --full-refresh.
• The jaffle-sandbox demo uses a seed for the Iceberg REST catalog as a temporary workaround.
• Related feature issue: DuckDB: add initial catalogs.yml v2 support for glue + iceberg_rest #14524
• Related implementation PRs: dbt-labs/fs#9420 and dbt-labs/fs#9587

[dataders]

Demo PR with Iceberg seed workaround: dbt-labs/jaffle-sandbox#1

[dataders]

Update:

• Fusion/fs draft fix: https://github.com/dbt-labs/fs/pull/9658
• jaffle-sandbox demo now uses dbt run for all four catalog models: Add catalogs v2 Iceberg and DuckLake demo jaffle-sandbox#1
• Upstream dbt-duckdb heads-up draft: [DO NOT MERGE (yet)] Avoid Iceberg CTAS for table materialization duckdb/dbt-duckdb#725

Verified the jaffle-sandbox demo with dbt run --select tag:catalog_demo: built-in DuckDB, local DuckLake, remote MotherDuck DuckLake, and local Iceberg REST all succeeded.

[dataders]

resolved (at least in dbt Core v2) via duckdb/duckdb-iceberg#974