Materialization Behavior
Choose materialization strategies for Redshift models and semantic layers.
With a validated gateway, this is the build stage. Choose a model kind and understand its Redshift operation before running it.
Materialization behavior per model kind
SQL models compile to Redshift SQL (PostgreSQL-derived).
FULL
Table rebuild / truncate+insert per adapter
VIEW
CREATE VIEW
INCREMENTAL_BY_TIME_RANGE
Windowed delete/merge + insert
INCREMENTAL_BY_UNIQUE_KEY
MERGE / delete+insert on key
INCREMENTAL_BY_PARTITION
Partition-scoped overwrite
Engine-native feature support
After core models work, add Redshift-specific capabilities where they add clear value.
Cluster + database
Depot host, database
Must exist
Schema
Vulcan CREATE SCHEMA / before_all
e.g. bronzev2
Table (dist/sort)
Guarded post_statements
COMPOUND SORTKEY, DISTKEY
View
kind VIEW
No view comments via Vulcan
MERGE / incremental
Model kinds
Validate keys
COPY from S3
Boundary / before_all
Spectrum + IAM outside Vulcan
GRANT
SQL in lifecycle or admin
Adapter supports grants
Python models
.py in project
Validate per image
Redshift Serverless
Depot connection
Same SQL path; tune RPU in AWS
Not supported
Vulcan does not manage these Redshift capabilities:
Cluster creation, resizing, or WLM configuration: no AWS lifecycle exists in Vulcan. Provision through the AWS console, Terraform, or Redshift Serverless admin.
REPLACE TABLEsemantics (SUPPORTS_REPLACE_TABLE = False): expect the FULL-refresh and merge strategies Vulcan emits, not a naiveCREATE OR REPLACE TABLE.Rich view comments (
COMMENT_CREATION_VIEW = UNSUPPORTED): document in the semanticdescriptionorai_contextinstead.The full PostgreSQL extension surface: Redshift supports a subset; validate UDFs against the Redshift SQL reference.
Superuser-only operations: run these as admin outside Vulcan.
Transactions are also limited on Redshift, so design incrementals to be idempotent on retry, and run ANALYZE after large loads (after_all) so the optimizer and BI stay healthy.
Lowercase identifiers
Lifecycle DDL guards
Session tuning in before_all
Use SET statement_timeout, SET query_group (WLM) sparingly, and reset them in after_all.
Semantic models
Use kind: semantic, with explicit dimensions and measures, joins for multi-entity queries, and optional rollups with enable_rollup: true in config. Rollups require the transpiler.
Business metrics
Data quality
Redshift uses the same four layers as other engines: linter, assertions, unit tests (DuckDB locally), audits, and kind: dq checks that run on the cluster after a successful run or plan apply.
Lineage and version rollback
Lineage comes from Vulcan's SQL parse. Cross-schema references need either visible tables or a declaration in inputs.yaml so linting and lineage resolve them.
Before you ship checklist (Data Product developer)
Next step
After models deploy, make data available through Serve, or skip straight to a ready-made shape in Deployment Recipes.
Last updated
Was this helpful?