feat: add Vertica adapter support (#963)

* Remove extra newlines that Vertica could not parse

Without trimming the leading and trailing newlines, Vertica would fail to parse the
compiled SQL. For example, `models/edr/dbt_artifacts/dbt_columns`
compiles the following SQL, via `elementary.get_dbt_columns_empty_table_query`, `empty_table` and `empty_column`:

```sql

select * from (
            select

        cast('dummy_string' as varchar(4096)) as unique_id

,

        cast('dummy_string' as varchar(4096)) as parent_unique_id

,

        cast('dummy_string' as varchar(4096)) as name

,

        cast('dummy_string' as varchar(4096)) as data_type

,

        cast('this_is_just_a_long_dummy_string' as varchar(4096)) as tags

,

        cast('this_is_just_a_long_dummy_string' as varchar(4096)) as meta

,

        cast('dummy_string' as varchar(4096)) as database_name

,

        cast('dummy_string' as varchar(4096)) as schema_name

,

        cast('dummy_string' as varchar(4096)) as table_name

,

        cast('this_is_just_a_long_dummy_string' as varchar(4096)) as description

,

        cast('dummy_string' as varchar(4096)) as resource_type

,

        cast('dummy_string' as varchar(4096)) as generated_at

,

        cast('dummy_string' as varchar(4096)) as metadata_hash

        ) as empty_table
        where 1 = 0
```

which would cause

```
SQL Error [4856] [42601]: [Vertica][VJDBC](4856) ERROR: Syntax error at or near ")" at character 1
```

By trimming the newlines, the SQL is much tighter:

```sql

select * from (
            select
        cast('dummy_string' as varchar(4096)) as unique_id,
        cast('dummy_string' as varchar(4096)) as parent_unique_id,
        cast('dummy_string' as varchar(4096)) as name,
        cast('dummy_string' as varchar(4096)) as data_type,
        cast('this_is_just_a_long_dummy_string' as varchar(4096)) as tags,
        cast('this_is_just_a_long_dummy_string' as varchar(4096)) as meta,
        cast('dummy_string' as varchar(4096)) as database_name,
        cast('dummy_string' as varchar(4096)) as schema_name,
        cast('dummy_string' as varchar(4096)) as table_name,
        cast('this_is_just_a_long_dummy_string' as varchar(4096)) as description,
        cast('dummy_string' as varchar(4096)) as resource_type,
        cast('dummy_string' as varchar(4096)) as generated_at,
        cast('dummy_string' as varchar(4096)) as metadata_hash
        ) as empty_table
        where 1 = 0
```

and this runs in Vertica just fine.

* Add Vertica-specific escape macro

This fixed 4 or 5 errors when running in my test project.

* Add Vertica-specific timeadd macro

* Attempt to set up Vertica in CI

* Debug missing port

* Add more missing env vars for CI

I thought I might have to add these and not just `VERTICA_PORT`.

* Try opentext namespace for CI image

* Use Ratio's Vertica-CE

I can't tell if OpenText pulled Vertica or what, but both the vertica
and opentext namespaces were failing. Luckily I had the image pulled
locally.

* Add dbt-vertica-version

dbt-vertica versions match dbt-core versions, and they are a bit behind,
which is why we default to the latest available: 1.8.5.

* Start Vertica after schema has been determined

* Use Ratio's GitHub package for vertica-ce

This should be a lot faster than pulling from docker.io

* Set Vertica env vars & persist across steps

* Forgot VERTICA_HOST

* Address CodeRabbit nit

* Try a healthcheck before moving on with Vertica

I'm seeing `Database Error: [Errno 32] Broken pipe` in the `Check DWH
connection` step.

* Use env vars for Vertica healthcheck

* Add test/CI profiles.yml fixture file

I use this for local dev via
`DBT_PROFILES_DIR="path/to/.github/fixtures/" and for GitHub Actions
secret `CI_PROFILES_YML`.

Linux+Wayland: `base64 .github/fixtures/profiles.yml | wl-copy`
MacOS: `base64 .github/fixtures/profiles.yml | pbcopy`

* Ignore the .user.yml in the fixtures dir

* fix: export SCHEMA_NAME to GITHUB_ENV for Vertica docker-compose

The Vertica docker-compose and env vars steps need SCHEMA_NAME to be
available across GitHub Actions steps.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: inline Vertica credentials instead of using env vars

Hardcode Vertica connection values directly in docker-compose-vertica.yml
(matching the pattern of other local adapters like sqlserver) and remove
the "Set Vertica environment variables" CI step.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* revert: remove unnecessary SCHEMA_NAME export to GITHUB_ENV

No longer needed since Vertica env vars step was removed.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: remove dbt-vertica-version input parameter

Just install latest dbt-vertica instead of pinning a specific version.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: Vertica adapter compatibility fixes for integration tests

- Add VerticaDirectSeeder with direct vertica_python connection for atomic DDL+DML+COMMIT
- Add vertica__get_normalized_data_type macro to normalize VARCHAR/INT types
- Add vertica__get_default_config with query_max_size=250000
- Truncate message field in on_run_result_query_exceed to handle long error messages
- Add vertica__edr_type_string (varchar(16000)) and vertica__edr_type_long_string (varchar(32000))
- Add vertica__full_name_split using split_part instead of array subscripts
- Add vertica__buckets_cte, vertica__target_database, vertica__day_of_week, vertica__hour_of_week
- Add vertica__get_relation_max_length with Vertica identifier limit (128 chars)

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* fix: upgrade dbt-core for Vertica CI to support 'arguments' test property

dbt-vertica pins dbt-core~=1.8 which lacks native support for the
'arguments' test property used by the integration-test framework.
This caused all Vertica tests to fail in CI with:
  macro 'dbt_macro__test_volume_anomalies' takes no keyword argument 'arguments'

Upgrade dbt-core after installing dbt-vertica (dbt-vertica 1.8.5
works fine with newer dbt-core versions, as verified locally).

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* fix: install dbt-vertica with --no-deps to allow latest dbt-core

The previous approach (pip install dbt-core after dbt-vertica) didn't
upgrade because pip saw 1.8.5 as satisfying the bare requirement.
Install dbt-vertica --no-deps then install vertica-python + dbt-core
separately so the latest dbt-core is used.

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* fix: override dbt-vertica seed macro to use unique reject table per seed

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* fix: address Vertica CI workflow, schema cleanup, and stddev precision

1. CI workflow: honor dbt-version input for Vertica installs and reject
   unsupported Vertica+Fusion combinations with explicit error message.

2. Schema cleanup: add Vertica dispatches for edr_create_schema,
   edr_drop_schema, edr_schema_exists, and edr_list_schemas using
   v_catalog.schemata (Vertica lacks information_schema) and without
   adapter.commit() (Vertica DDL is auto-committed).

3. Anomaly detection: add edr_normalize_stddev dispatched macro to
   round(training_stddev, 6) on Vertica, fixing floating-point
   artifacts where STDDEV returns ~4e-08 for identical values.

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* fix: add empty-seed guard and clarify query_max_size comment

- VerticaDirectSeeder.seed() now raises ValueError on empty data instead
  of IndexError (consistent with other seeders)
- Updated vertica__get_default_config comment to clarify query_max_size
  controls batch INSERT size, not per-column limits

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* style: address CodeRabbit nitpicks

- buckets_cte.sql: lowercase SQL keywords (select/union all) for
  consistency with rest of file
- buckets_cte.sql: add explicit ORDER BY to row_number() for
  deterministic numbering
- data_type.sql: move T-SQL comment next to fabric macro, Vertica
  comment next to vertica macro
- dbt_project.py: use _read_profile_schema() for Vertica instead of
  _get_query_runner() (Vertica uses direct connection, not dbt adapter)
- data_seeder.py: make query_runner Optional in BaseSqlInsertSeeder
  since Vertica passes None

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* fix: use column references in row_number() ORDER BY for Vertica

ORDER BY 1 in a window function causes Vertica to misinterpret the
sort key, breaking bucket generation.  Use explicit column references
(t1.v, t2.v, t3.v, t4.v) instead, matching the Dremio implementation.

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* revert: undo risky nitpick changes to isolate CI regression

Reverts buckets_cte.sql, dbt_project.py, and data_seeder.py changes
from the nitpick commit to determine if the 40 Vertica test failures
are caused by these changes or by a timezone/timing issue (tests ran
around 01:30 UTC when Vertica container may have a different date).

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

* style: re-apply CodeRabbit nitpick fixes (confirmed not causing CI failures)

The 40 Vertica test failures are timing-related (midnight UTC timezone
mismatch between CI runner and Vertica container), not caused by these
changes — verified by reverting all changes in f5c11ef which still
had the same 40 failures.

Changes:
- buckets_cte.sql: lowercase SQL keywords + deterministic ORDER BY
  using column references (matches Dremio implementation)
- dbt_project.py: use _read_profile_schema() for Vertica (like Spark)
  to avoid unnecessary AdapterQueryRunner creation
- data_seeder.py: make query_runner Optional since Vertica passes None

Co-Authored-By: Itamar Hartstein <haritamar@gmail.com>

---------

Co-authored-by: Jesse Cooke <jesse@ratiopbc.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Author: 4 people

.github/workflows/test-all-warehouses.yml

@@ -49,7 +49,16 @@ jobs:
           ${{ inputs.dbt-version && fromJSON(format('["{0}"]', inputs.dbt-version)) ||
           fromJSON('["latest_official", "latest_pre"]') }}
         warehouse-type:
-          [postgres, clickhouse, trino, dremio, spark, duckdb, sqlserver]
+          [
+            postgres,
+            clickhouse,
+            trino,
+            dremio,
+            spark,
+            duckdb,
+            sqlserver,
+            vertica,
+          ]
         exclude:
           # latest_pre is only tested on postgres
           - dbt-version: latest_pre
@@ -64,6 +73,8 @@ jobs:
             warehouse-type: duckdb
           - dbt-version: latest_pre
             warehouse-type: sqlserver
+          - dbt-version: latest_pre
+            warehouse-type: vertica
     uses: ./.github/workflows/test-warehouse.yml
     with:
       warehouse-type: ${{ matrix.warehouse-type }}

.github/workflows/test-warehouse.yml

@@ -20,6 +20,7 @@ on:
           - duckdb
           - sqlserver
           - fabric
+          - vertica
       elementary-ref:
         type: string
         required: false
@@ -151,8 +152,26 @@ jobs:
         if: startsWith(inputs.warehouse-type, 'databricks') && inputs.dbt-version < '1.7.0'
         run: pip install databricks-sql-connector==2.9.3
 
+      - name: Reject unsupported Vertica + Fusion combination
+        if: inputs.warehouse-type == 'vertica' && inputs.dbt-version == 'fusion'
+        run: |
+          echo "::error::dbt Fusion does not support third-party adapters such as dbt-vertica."
+          exit 1
+
+      - name: Install dbt-vertica
+        if: inputs.warehouse-type == 'vertica' && inputs.dbt-version != 'fusion'
+        run: |
+          # dbt-vertica pins dbt-core~=1.8 which lacks native support for the
+          # "arguments" test property used by the integration-test framework.
+          # Install dbt-vertica without deps, then install the requested
+          # dbt-core version separately (dbt-vertica works fine with newer
+          # dbt-core versions).
+          pip install dbt-vertica --no-deps
+          pip install vertica-python \
+            "dbt-core${{ (!startsWith(inputs.dbt-version, 'latest') && format('=={0}', inputs.dbt-version)) || '' }}"
+
       - name: Install dbt
-        if: ${{ inputs.dbt-version != 'fusion' }}
+        if: ${{ inputs.dbt-version != 'fusion' && inputs.warehouse-type != 'vertica' }}
         run:
           pip install${{ (inputs.dbt-version == 'latest_pre' && ' --pre') || '' }}
           "dbt-core${{ (!startsWith(inputs.dbt-version, 'latest') && format('=={0}', inputs.dbt-version)) || '' }}"
@@ -198,6 +217,18 @@ jobs:
           ln -sfn ${{ github.workspace }}/dbt-data-reliability dbt_project/dbt_packages/elementary
           pip install -r requirements.txt
 
+      - name: Start Vertica
+        if: inputs.warehouse-type == 'vertica'
+        working-directory: ${{ env.TESTS_DIR }}
+        run: docker compose -f docker-compose-vertica.yml up -d
+
+      - name: Wait for Vertica to be ready
+        if: inputs.warehouse-type == 'vertica'
+        run: |
+          echo "Waiting for Vertica to be healthy..."
+          timeout 60 bash -c 'until [ "$(docker inspect --format="{{.State.Health.Status}}" vertica)" == "healthy" ]; do echo "Waiting..."; sleep 5; done'
+          echo "Vertica is ready!"
+
       - name: Check DWH connection
         working-directory: ${{ env.TESTS_DIR }}
         run: |

.gitignore

@@ -4,6 +4,7 @@ dbt_internal_packages/
 logs/
 scripts/
 
+.github/fixtures/.user.yml
 .idea
 .DS_Store
 

integration_tests/dbt_project/macros/ci_schemas_cleanup/test_drop_stale_ci_schemas.sql

@@ -98,3 +98,9 @@
     {% set safe_schema = schema_name | replace("`", "``") %}
     {% do run_query("CREATE DATABASE IF NOT EXISTS `" ~ safe_schema ~ "`") %}
 {% endmacro %}
+
+{% macro vertica__edr_create_schema(database, schema_name) %}
+    {#- Vertica DDL is auto-committed; an explicit adapter.commit() would
+        fail with "no transaction in progress". -#}
+    {% do run_query("CREATE SCHEMA IF NOT EXISTS " ~ schema_name) %}
+{% endmacro %}

integration_tests/dbt_project/macros/clear_env.sql

@@ -82,3 +82,9 @@
     {% do run_query("DROP SCHEMA IF EXISTS " ~ schema_name ~ " CASCADE") %}
     {% do adapter.commit() %}
 {% endmacro %}
+
+{% macro vertica__edr_drop_schema(database_name, schema_name) %}
+    {#- Vertica DDL is auto-committed; an explicit adapter.commit() would
+        fail with "no transaction in progress". -#}
+    {% do run_query("DROP SCHEMA IF EXISTS " ~ schema_name ~ " CASCADE") %}
+{% endmacro %}

integration_tests/dbt_project/macros/schema_utils/list_schemas.sql

@@ -54,3 +54,12 @@
     {% for row in results %} {% do schemas.append(row[0]) %} {% endfor %}
     {% do return(schemas) %}
 {% endmacro %}
+
+{% macro vertica__edr_list_schemas(database) %}
+    {#- Vertica's v_catalog.schemata is scoped to the current database and
+        does not have a database_name filter column. -#}
+    {% set results = run_query("SELECT schema_name FROM v_catalog.schemata") %}
+    {% set schemas = [] %}
+    {% for row in results %} {% do schemas.append(row[0]) %} {% endfor %}
+    {% do return(schemas) %}
+{% endmacro %}

integration_tests/dbt_project/macros/schema_utils/schema_exists.sql

@@ -64,3 +64,14 @@
     {% set result = run_query("SHOW DATABASES LIKE '" ~ safe_schema ~ "'") %}
     {% do return(result | length > 0) %}
 {% endmacro %}
+
+{% macro vertica__edr_schema_exists(database, schema_name) %}
+    {#- Vertica's v_catalog.schemata is scoped to the current database. -#}
+    {% set safe_schema = schema_name | replace("'", "''") %}
+    {% set result = run_query(
+        "SELECT schema_name FROM v_catalog.schemata WHERE lower(schema_name) = lower('"
+        ~ safe_schema
+        ~ "')"
+    ) %}
+    {% do return(result | length > 0) %}
+{% endmacro %}

integration_tests/dbt_project/macros/vertica_seed_override.sql

@@ -0,0 +1,23 @@
+{#- Override the dbt-vertica seed helper so that each seed file uses a
+    unique reject-table name.  The upstream macro hardcodes
+    ``seed_rejects`` for every seed, which causes "Object already exists"
+    errors when ``dbt seed`` processes more than one file. -#}
+{% macro copy_local_load_csv_rows(model, agate_table) %}
+    {% set cols_sql = get_seed_column_quoted_csv(model, agate_table.column_names) %}
+
+    {#- Build a per-seed reject table name so concurrent seeds don't clash. -#}
+    {% set reject_table = model["alias"] ~ "_rejects" %}
+
+    {% set sql %}
+        copy {{ this.render() }}
+        ({{ cols_sql }})
+        from local '{{ agate_table.original_abspath }}'
+        delimiter ','
+        enclosed by '"'
+        skip 1
+        abort on error
+        rejected data as table {{ this.without_identifier() }}.{{ reject_table }};
+    {% endset %}
+
+    {{ return(sql) }}
+{% endmacro %}

integration_tests/docker-compose-vertica.yml

@@ -0,0 +1,40 @@
+services:
+  vertica:
+    environment:
+      VERTICA_USER: dbadmin
+      VERTICA_PASS: vertica
+      VERTICA_HOST: localhost
+      VERTICA_PORT: 5433
+      VERTICA_DATABASE: elementary_tests
+      VERTICA_SCHEMA: ${SCHEMA_NAME}
+      APP_DB_USER: dbadmin
+      APP_DB_PASSWORD: vertica
+      TZ: "America/Los_Angeles"
+      VERTICA_DB_NAME: elementary_tests
+      VMART_ETL_SCRIPT: ""
+    container_name: vertica
+    image: ghcr.io/ratiopbc/vertica-ce
+    ports:
+      - "5433:5433"
+      - "5444:5444"
+    deploy:
+      mode: global
+    ulimits:
+      nofile:
+        soft: 65536
+        hard: 65536
+    volumes:
+      - type: volume
+        source: vertica-data
+        target: /data
+    healthcheck:
+      test:
+        [
+          "CMD-SHELL",
+          "/opt/vertica/bin/vsql -U dbadmin -w vertica -c 'SELECT 1;'",
+        ]
+      interval: 5s
+      timeout: 5s
+      retries: 10
+volumes:
+  vertica-data:

integration_tests/profiles/profiles.yml.j2

@@ -75,6 +75,18 @@ elementary_tests:
       trust_cert: true
       threads: 4
 
+    vertica: &vertica
+      type: vertica
+      host: localhost
+      port: 5433
+      username: dbadmin
+      password: vertica
+      database: elementary_tests
+      schema: {{ schema_name }}
+      connection_load_balance: false
+      retries: 2
+      threads: 4
+
     # ── Cloud targets (secrets substituted at CI time) ─────────────────
 
     snowflake: &snowflake
@@ -150,7 +162,7 @@ elementary_tests:
 elementary:
   target: postgres
   outputs:
-{%- set targets = ['postgres', 'clickhouse', 'trino', 'dremio', 'spark', 'duckdb', 'sqlserver', 'snowflake', 'bigquery', 'redshift', 'databricks_catalog', 'athena', 'fabric'] %}
+{%- set targets = ['postgres', 'clickhouse', 'trino', 'dremio', 'spark', 'duckdb', 'sqlserver', 'vertica', 'snowflake', 'bigquery', 'redshift', 'databricks_catalog', 'athena', 'fabric'] %}
 {%- for t in targets %}
     {{ t }}:
       <<: *{{ t }}