Redshift Guide

Amazon Redshift Guide

Redshift is NOT PostgreSQL (read first)

Redshift speaks PostgreSQL’s wire protocol and shares much of its surface syntax, so LLMs assume PostgreSQL behavior carries over — it frequently does not. Divergences span system tables (pg_catalog is incomplete), DDL (no indexes, no sequences), functions (string_agg, SUBSTR on tables, leader-node-only functions), types (a text column becomes VARCHAR(256)), and comparison semantics (trailing blanks, unenforced constraints). Assume divergence and verify against the reference below — do not answer from PostgreSQL habit. Common PostgreSQL→Redshift divergences are in references/redshift-sql-syntax.md.

Works best with the AWS MCP server (opens in a new tab) — it runs the AWS CLI and Redshift Data API calls below in a sandboxed, audit-logged environment. All guidance here is plain AWS CLI and SQL and works without it.

STEP 0: Serverless or Provisioned?

Establish this before answering — APIs, system tables, and capabilities differ. Take it from the question when it says which one; ask when it does not. SELECT version() does not identify it.

• Serverless — identified by a workgroup (and namespace). Data API calls take --workgroup-name; the user says “workgroup”/”Serverless”.
• Provisioned — identified by a cluster. Data API calls take --cluster-identifier; the user says “cluster”.

Critical Facts

• SHOW commands are the primary metadata interface — SHOW DATABASES, SHOW SCHEMAS, SHOW TABLES, SHOW COLUMNS, SHOW TABLE, SHOW VIEW. Do NOT default to pg_catalog or information_schema. → Load references/redshift-sql-metadata.md for metadata/discovery questions and any “relation does not exist” report — it has the diagnostic flow.
• SYS_ views are the preferred system views — they work everywhere. STL_, STV_, SVL_, and SVCS_ are provisioned single-AZ only, and some SVV_ views are unsupported on Serverless. → Load references/redshift-sql-metadata.md for any system-view or monitoring question.
• sys_load_error_detail for COPY debugging (not stl_load_errors, which is provisioned single-AZ only).
• DATEADD/DATEDIFF — unit-first argument order: DATEADD(day, -30, GETDATE()), DATEDIFF(day, start, end).
• APPROXIMATE COUNT(DISTINCT col) — Redshift-specific, ~2% error, much faster than exact COUNT(DISTINCT) on large datasets.
• MERGE … REMOVE DUPLICATES — simplified dedup when source and target have identical schemas.
• COPY should use IAM_ROLE (the namespace role, not the caller role) + supports MANIFEST for explicit file lists + MAXERROR for error tolerance.
• SUBSTR() is leader-node-only — works on literals but errors on table columns (SUBSTR() function is not supported (Hint: use SUBSTRING instead)). Use SUBSTRING() on columns.
• UNIQUE / PRIMARY KEY / FOREIGN KEY are informational only — NOT enforced (duplicate rows are accepted with no error). Optimizer hints; enforce integrity in the application or via MERGE. NOT NULL IS enforced.
• SHOW VIEW <schema.name> returns the definition of a regular view, materialized view, or late-binding view. MV freshness: SVV_MV_INFO (is_stale).
• TOP N and LIMIT N both work (TOP N PERCENT does not). A text column becomes VARCHAR(256) — use VARCHAR(max) or explicit length.
• Iceberg tables use CREATE TABLE ... USING ICEBERG (not STORED AS ICEBERG, not TABLE_FORMAT=ICEBERG).
• Datashares support read and write operations — consumers can write once the producer grants write privileges. Treat “permission denied” on a datashare write as a missing grant, not an unsupported operation. → Load references/redshift-sql-metadata.md for requirements and limits.

Safety Guardrails

BLOCK: DROP DATABASE, DELETE without WHERE, publicly-accessible=true, GRANT ALL ON ALL WARN then confirm: RESIZE, RESTORE, VACUUM on large tables, ALTER PASSWORD, WLM config change Confirm: CREATE, GRANT specific, COPY, UNLOAD

Security Considerations

Apply these defaults when generating anything that connects, loads, or exports. Details are in the reference files noted.

• In transit: the Data API is HTTPS-only. For JDBC/ODBC set the require_ssl parameter and connect with sslmode=verify-full so the server certificate is checked.
• At rest: keep cluster/namespace encryption enabled, and add ENCRYPTED KMS_KEY_ID '<arn>' to UNLOAD — it writes query results to S3, outside Redshift’s own encryption. → references/redshift-sql-ddl-copy.md
• Credentials: prefer SecretArn (Secrets Manager) or IAM Identity Center; DbUser is acceptable because it issues temporary credentials. Never place database passwords in code, environment variables, or SQL text. → references/redshift-sql-recipes-load-api.md
• Least privilege: scope the namespace IAM_ROLE to the specific bucket and prefix (s3:GetObject on arn:aws:s3:::<bucket>/<prefix>/*), not s3:* or a managed full-access policy, and condition its trust policy on both aws:SourceArn (the cluster/namespace ARN) and aws:SourceAccount — SourceArn alone still allows another resource in the account to assume it. Grant per-object privileges rather than GRANT ALL ON ALL.
• Audit: CloudTrail records redshift-data:* API calls but not the SQL executed; enable Redshift audit logging (useractivitylog, connectionlog, userlog) for that. Both capture query text and user activity, so encrypt every destination in use: the CloudWatch Logs group (aws logs associate-kms-key), the CloudTrail trail (SSE-KMS), and the audit-log S3 bucket (SSE-S3 — audit logging to S3 supports only S3-managed keys, not KMS). Serverless only supports sending audit logs to CloudWatch.
• Network: keep PubliclyAccessible=false and connect over a VPC endpoint. Do not open port 5439 to 0.0.0.0/0 or ::/0 — scope inbound rules to specific CIDRs or to a referencing security group.
• Sensitive data: Data API results persist for 24h and sys_load_error_detail can echo fragments of rejected rows, so treat statement IDs and load-error output as sensitive.
• Further reading: Security in Amazon Redshift (opens in a new tab) for the full guidance behind these defaults.

Routing Table

MANDATORY: When a question matches a row below, you MUST load and read the referenced file BEFORE answering.

Ask whether the target is provisioned or Serverless before giving troubleshooting steps — unless the question already says which one, in which case use that and do not re-confirm.

Data API Quick Reference

→ Load references/redshift-sql-recipes-load-api.md before answering ANY Data API, COPY-error, or async-query question. It carries the bounded poll loop, the HasResultSet and ResourceNotFoundException handling, the per-target parameters, and the auth options.

Data API calls are async by default — use long polling (--wait-time-seconds, 1–30) rather than blind sleeps, and keep a bounded loop for work that can exceed 30s. Serverless takes --workgroup-name, provisioned takes --cluster-identifier.