You’re hitting a known Redshift issue: the Redshift driver refuses a single prepared statement that contains multiple commands (

begin; delete …; insert …; commit;

). Elementary’s

delete_and_insert

logic is currently compiling exactly that pattern for Redshift, which produces the error you’re seeing. [GitHub issue; Slack thread] From the shared issues/threads, there are three practical options: 1. Short‑term: disable artifacts autoupload in the post‑hook This stops the failing

dbt_models

/

dbt_columns

writes during

on-run-end

and lets your runs succeed, while you handle artifacts separately. [GitHub issue]

# dbt_project.yml
vars:
  elementary:
    disable_run_results: true
    disable_dbt_artifacts_autoupload: true
    disable_dbt_invocation_autoupload: true

Some teams even temporarily disabled Elementary entirely on prod targets:

models:
  elementary:
    +schema: elementary
    enabled: "{{ target.name in ['beta'] }}"

2. Recommended workaround: override the Redshift macro Community members fixed this by overriding the Redshift implementation of

get_delete_and_insert_queries

so it returns two separate statements (delete + insert) instead of one multi‑command transaction. [Slack thread; Linen recap] Create a macro in your project, e.g. `macros/redshift_get_delete_and_insert_queries.sql`:

{% macro redshift__get_delete_and_insert_queries(relation, insert_relation, delete_relation, delete_column_key) %}
    {% set queries = [] %}

    {% if delete_relation %}
        {% set delete_query %}
            delete from {{ relation }}
            where
            {{ delete_column_key }} is null
            or {{ delete_column_key }} in (select {{ delete_column_key }} from {{ delete_relation }});
        {% endset %}
        {% do queries.append(delete_query) %}
    {% endif %}

    {% if insert_relation %}
        {% set insert_query %}
            insert into {{ relation }} select * from {{ insert_relation }};
        {% endset %}
        {% do queries.append(insert_query) %}
    {% endif %}

    {% do return(queries) %}
{% endmacro %}

If you’re on dbt Cloud or want to be explicit, add a dispatch rule so your project’s macro overrides Elementary’s: [Slack thread]

# dbt_project.yml
dispatch:
  - macro_namespace: elementary
    search_order: ['<YOUR_PROJECT_NAME>', 'elementary']

Replace YOUR_PROJECT_NAME with the

name

from your

dbt_project.yml

. This makes Elementary emit two separate statements (delete, then insert) instead of a single

begin … delete … insert … commit

block, which avoids the Redshift “multiple commands” limitation. 3. Run Elementary models/artifacts in a separate job As an additional mitigation, you can decouple Elementary from your main runs and execute its models (including artifacts) in a separate dbt job, so failures there don’t break core pipelines. The maintainers explicitly recommend running artifacts outside

on-run-end

when this issue appears. [GitHub issue] ------------------------- The knowledge sources show that a fix was released in Elementary 0.21.0, but multiple users on Redshift still reported the same error even after upgrading, and the macro override above is the documented working workaround. [GitHub issue; Slack follow‑up] 👍 0 button 👎 0 button ℹ️ Tip: Mention @elementary.ai in the thread for follow-ups or DM the app for 1-to-1 help.