How to migrate from Amazon Redshift to Firebolt Migrating from Amazon Redshift to Firebolt improves performance, scalability, and cost efficiency for analytics workloads. Firebolt’s decoupled architecture removes the limitations of Redshift’s monolithic cluster, allowing faster queries, elastic scaling, and better workload management. Instead of relying on a single cluster, Firebolt uses independent compute engines for workload isolation and cost control. To use these features effectively, plan your migration carefully and adapt your workloads to Firebolt’s architecture rather than simply copying your existing setup. This guide walks you through migrating from Redshift to Firebolt. It starts with an overview of key architectural and schema differences to help you understand Firebolt’s engine-based architecture, workload isolation, indexing, and flexible schema design. Then, it provides step-by-step instructions for exporting data from Redshift, loading it into Firebolt, translating queries, testing and optimizing performance, and setting up automated migration workflows. Differences between Amazon Redshift and Firebolt Differences between Amazon Redshift and Firebolt Topics:
  • Architectural differences – An overview of how to adapt your Redshift workflow to Firebolt’s decoupled compute-storage model, which provides elastic scaling, workload isolation, and advanced indexing.
    • Compute and storage – Scale compute resources for different workloads without resizing clusters or redistributing data.
    • Scalability and elasticity – Use right-sized engines for ingestion, analytics, and ETL without downtime.
    • Workload isolation – Assign dedicated engines to individual workloads to avoid resource contention and improve reliability.
    • Cost model – Control costs with pay-as-you-go billing and flexible engine usage. Pause idle engines and size compute resources according to workload demand.
  • Schema differences – An overview of how to adapt your schema from Redshift’s SORT and DIST keys to Firebolt’s dynamic indexing, denormalization, and flexible JSON handling, which replace manual distribution keys and rigid schemas for faster, simpler queries.
    • Table design – Use fact tables for large datasets and dimension tables for reference data. Optimize queries with columnar storage and primary indexes.
    • Indexing – Replace manual SORTKEY and DISTKEY tuning with dynamic primary and aggregating indexes for faster query performance and automatic data pruning.
    • Denormalization – Reduce joins and query complexity by storing frequently used fields directly in fact tables.
    • Handling JSON – Store JSON as TEXT and query nested fields using JSON functions without rigid schemas.
    • Data types – Convert Redshift data types to Firebolt equivalents. Use TEXT for strings and JSON data, and ensure numeric precision and date compatibility.
Migration steps:
  1. Export data from Redshift – Learn how to efficiently export your Redshift data to Amazon S3 in CSV or Parquet format, format it for Firebolt, and organize it for optimal parallel loading.
  2. Load data into Firebolt – Load exported data into Firebolt using the COPY FROM command, using parallel ingestion and indexing to optimize loading speed and query performance.
  3. Translate queries – Use indexing, denormalization, and JSON functions to improve query efficiency and reduce complexity when migrating to Firebolt.
  4. Performance testing and optimization – Validate data consistency, test query performance, adjust engine configurations, and fine-tune indexes to maximize speed and resource efficiency.
  5. Automate migration – Set up an automated pipeline using tools like Airflow or dbt to streamline data migration, schema deployment, and query adaptation.

​
Architectural differences

Migrating from Amazon Redshift to Firebolt requires adapting to a different architecture. Redshift’s monolithic cluster model, where compute and storage are tightly integrated, limits flexibility and forces full-cluster scaling. In contrast, Firebolt separates compute engines from storage, enabling elastic scaling, workload isolation, and advanced indexing for faster performance, as summarized in the following table:
FeatureRedshiftFireboltRecommendations
Compute and storageMonolithic cluster with tightly coupled compute and storage.Decoupled architecture with virtualized compute engines and separate storage.Use multiple engines to separate workloads and improve flexibility.
Scalability and elasticityResizing requires downtime and offers limited concurrency scaling.Elastic scaling with independent compute and storage.Assign separate engines for ingestion, analytics, and ETL to optimize resource use.
Workload isolationShared cluster causes resource contention.Fully isolated workloads with multiple independent engines.Use dedicated engines to prevent interference between queries.
Cost modelBilled for cluster uptime, with additional charges for concurrency scaling.Pay-as-you-go pricing based on engine runtime.Pause idle engines and right-size compute to control costs.
The following sections outline key differences in compute and storage, scalability and elasticity, workload isolation, and cost model, to help you optimize Firebolt for both speed and efficiency. Firebolt differs from Redshift in compute, scalability, workload isolation, and cost model.

​
Compute and storage

Firebolt differs from Redshift in compute and storage. One of the biggest advantages of migrating to Firebolt is its engine-based architecture, which allows you to tailor your compute resources to specific workloads. Amazon Redshift combines compute and storage into a single cluster, meaning all workloads share the same resources, and scaling requires resizing the entire cluster, even if only compute or storage needs adjustment. In contrast, Firebolt separates compute engines from storage, allowing workloads to run on independent engines that can be resized, paused, or restarted without affecting storage. This prevents resource contention, improves query performance, and enables cost-efficient scaling. Since Firebolt allows dedicated compute engines for different workloads, selecting the right engine TYPE is critical to optimizing performance and cost efficiency. Unlike Redshift, where all workloads run on the same cluster, Firebolt lets you assign separate engines for ingestion, analytics, and transformation tasks. To maximize efficiency, engines should be right-sized based on workload needs, scaled independently of storage, and paused when not in use to reduce costs.
Best practice for compute and storageImpactHow to implement
Scale compute and storage independently.Allows flexible resource allocation without over-provisioning.Increase compute resources for analytics without expanding storage, and scale engines independently based on workload needs.
Firebolt allows scaling compute without affecting storage, unlike Redshift, where storage grows with compute. You can allocate more compute resources for analytics while keeping storage unchanged. Use smaller engines for intermittent workloads. The following code example shows how to scale an engine for high-currency analytics in Firebolt: 
ALTER ENGINE "analytics_engine" SET NODES = 10;

​
Scalability and elasticity

Firebolt differs from Redshift in scalability and elasticity. Firebolt’s architecture allows you to scale compute resources independently and elastically. Unlike Redshift, which requires cluster resizing and downtime, Firebolt engines can be adjusted at any time based on workload demands.
Best practice for scalability and elasticityImpactHow to implement
Use right-sized engines based on workload needs.Optimizes performance and cost by allocating appropriate resources for each workload.Start with small or medium engines for lighter workloads and scale up using ALTER ENGINE for high-concurrency or intensive tasks.
The following table outlines the recommended Firebolt engine configurations for different workloads, ensuring optimal performance, scalability, and cost efficiency:
Workload typeRecommended engine configuration
High-concurrency business intelligence (BI)Use medium to large engines with high parallelism for supporting many concurrent queries efficiently.
ETL and transformationsUse small to medium engines with moderate concurrency, optimized for data ingestion and transformations.
Data scienceUse small engines for iterative, ad-hoc queries or training datasets.
Data-intensive applicationsUse dedicated small engines with consistent SLA for low-latency queries.
Ad-hoc analyticsUse on-demand engines that start quickly and scale based on query complexity.
The following code example shows how to resize an existing engine for higher concurrency in Firebolt: 
ALTER ENGINE analytics_engine SET NODES = 8;

​
Workload isolation

Firebolt differs from Redshift in workload isolation. Unlike Redshift, where all queries run in a shared cluster, Firebolt enables workload isolation by assigning dedicated compute engines for different tasks. This ensures that data loading processes no longer compete with analytical queries, maintaining consistent performance. Allocate dedicated engines for critical workloads such as real-time dashboards or high-volume data transformation pipelines to maximize efficiency. Some key differences include:
  • Firebolt engines support both read and write operations on shared data while maintaining strong consistency across engines, eliminating the need for manual synchronization.
  • Firebolt optimizes workload execution dynamically based on configuration, resource utilization, and query history. This helps balance latency and throughput, ensuring that resources are used efficiently.
Best practice for workload isolationImpactHow to implement
Use separate Firebolt engines for different workloads.Prevents resource contention and improves workload isolation.Assign dedicated engines for ingestion, analytics, and transformations to avoid competition for resources.
The following table provides recommended Firebolt engine configurations for different workloads, ensuring optimal performance, scalability, and cost efficiency:
WorkloadEngine nameConfigurationPurpose
Data ingestioningestion_engineTYPE = S, NODES = 1Continuous data loading.
High-concurrency queriesanalytics_engineTYPE = M, NODES = 6Supports many concurrent queries.
Real-time dashboardsdashboard_engineTYPE = M, NODES = 2Low-latency SLA-sensitive queries.
Ad-hoc analyticsad_hoc_engineTYPE = S, NODES = 2, AUTO_STOPCost-efficient for intermittent use.
The following code example shows how to create different engines with different sizes to separate data loading and analytics workloads in Firebolt: 
CREATE ENGINE "ingestion_engine" WITH
TYPE = "M"
NODES = 2;

CREATE ENGINE "analytics_engine" WITH
TYPE = "L"
NODES = 8;

​
Cost model

Firebolt engines consume credits only when engines are running, so pausing unused engines can significantly reduce costs. In Redshift, charges are based on cluster uptime, even if it is idle. In Firebolt, you can pause an engine without losing access to stored data. Firebolt differs from Redshift in cost model.
Best practice for scalability and elasticityImpactHow to implement
Pause idle engines to reduce costsLowers operational costs while keeping data accessible.Use auto-stop settings to pause unused engines and restart them on demand.
The following code example creates an engine that automatically pauses after 30 minutes of inactivity in Firebolt: 
CREATE ENGINE "adaptive_engine" WITH
TYPE = "L"
AUTO_STOP = 30
AUTO_START = TRUE;

​
Schema differences

Migrating from Amazon Redshift to Firebolt requires adapting your schema to leverage Firebolt’s indexing model, denormalization benefits, and optimized query execution. Unlike Redshift, which relies on SORT and DIST keys for performance tuning, Firebolt automatically optimizes queries using primary, and aggregating indexes. This eliminates the need for manual data distribution and sorting, reducing complexity while improving performance. Additionally, semi-structured data handling differs. Redshift’s SUPER type allows nested JSON storage, while Firebolt stores JSON as TEXT, enabling flexible querying with JSON functions like JSON_VALUE and JSON_EXTRACT. Firebolt’s schema design also favors denormalization, reducing the need for complex joins and improving analytical query speed. Firebolt has multiple schema differences from Redshift. The following table outlines key schema differences between Redshift and Firebolt, highlighting the mechanisms Firebolt uses to optimize data modeling and query performance:
FeatureRedshiftFireboltRecommendations
Table designUses fact and dimension tables with explicit DISTKEY and SORTKEY properties for tuning.Uses fact and dimension tables with indexing for optimized access.Design fact tables for large datasets and dimension tables for reference data.
IndexingUses SORTKEY and DISTKEY for performance tuning, requiring manual selection and optimization.Uses primary and aggregating indexes instead.Use primary indexes for efficient filtering and aggregating indexes for precomputed aggregations.
DenormalizationOften normalized to reduce redundancy and optimize joins.Encourages denormalization to reduce joins and improve query performance.Store frequently joined reference columns in fact tables to minimize joins.
Handling JSONSupports SUPER data type for nested JSON storage, queried with JSON_EXTRACT_PATH_TEXT.Stores JSON as TEXT and uses JSON_VALUE and JSON_POINTER_EXTRACT_TEXT for querying.Store JSON data as TEXT and use JSON functions to extract values on demand.
Data typesSupports a broad range of native types including SUPER.Uses similar types, but JSON is stored as TEXT.Convert SUPER columns to TEXT and adjust data types to match Firebolt-supported types.

​
Table design

Firebolt has a different table design than Redshift. Firebolt’s columnar storage model organizes data in tablets, allowing queries to scan only the necessary portions of a dataset. Unlike row-based storage, columnar storage reduces I/O overhead and improves query performance by efficiently storing, filtering, and aggregating data.
Best practice for table designImpactHow to implement
Store frequently queried large datasets in fact tables.Improves query speed and reduces scan times.Create fact tables for large datasets used in analytical queries.
Use dimension tables for reference data.Simplifies lookups and schema organization.Define dimension tables for static or lookup datasets.
Define primary indexes on frequently filtered columns.Enables efficient pruning and faster filtering.Add primary indexes to columns commonly used in WHERE or JOIN conditions.
The following code example shows a normalized Redshift schema using separate tables for orders and customers: 
CREATE TABLE orders (
    order_id INT PRIMARY KEY,
    customer_id INT,
    order_date DATE,
    total_amount DECIMAL(10,2)
);

CREATE TABLE customers (
    customer_id INT PRIMARY KEY,
    customer_name TEXT
);
In Redshift, retrieving an order with its customer name requires a join as follows: 
SELECT o.order_id, o.order_date, o.total_amount, c.customer_name
FROM orders o
JOIN customers c ON o.customer_id = c.customer_id
WHERE o.order_date >= '2024-01-01';
The following code examples show how to define a fact table and a dimension table with primary indexes on frequently filtered columns in Firebolt: 
CREATE FACT TABLE orders (
    order_id INT,
    customer_id INT,
    order_date DATE,
    total_amount DECIMAL(10,2)
)
PRIMARY INDEX (customer_id);

CREATE DIMENSION TABLE customers (
    customer_id INT,
    customer_name TEXT
)
PRIMARY INDEX (customer_id);
The following query in Firebolt retrieves orders for a specific customer and prunes tablets using the customer_id primary index, scanning only relevant data: 
SELECT o.order_id, o.order_date, o.total_amount, c.customer_name
FROM orders o
JOIN customers c
ON o.customer_id = c.customer_id
WHERE o.customer_id = 12345;

​
Indexing

Firebolt uses dynamic indexing. Amazon Redshift uses SORTKEY and DISTKEY to improve data distribution and query performance. The SORTKEY controls the physical order of stored data, optimizing filtering and range-based queries. The DISTKEY distributes data across cluster nodes to balance workloads. These keys require careful, manual tuning and must be set during table creation. Changing them requires rebuilding tables and reloading data, making schema adjustments rigid and time-consuming. Poor DISTKEY choices in Redshift can lead to data skew, where some nodes store more data than others, creating performance bottlenecks. Additionally, scaling Redshift clusters requires redistributing data based on the defined DISTKEY. This process can result in downtime and degraded performance while workloads rebalance, and often forces additional manual optimization. Firebolt eliminates manual distribution keys and sort keys by using dynamic indexing to optimize query processing. Indexes in Firebolt can be added, modified, or replaced without reloading data. Primary indexes physically organize data to speed up filtering and scanning during query runtime. Aggregating indexes precompute aggregate results, allowing faster retrieval without dynamic calculations. Firebolt’s indexing model works with tablet-based storage, reducing the need for manual tuning tasks such as VACUUM and ANALYZE in Redshift. The decoupled compute-storage architecture allows dynamic data access without requiring distribution keys or physical data redistribution when scaling. Indexes enable the engine to prune irrelevant data and improve query performance automatically. Understanding how indexes and tablet-based storage replace Redshift’s DISTKEY and SORTKEY is essential for successful schema migration. Firebolt’s indexing approach simplifies schema design, reduces maintenance overhead, and delivers faster query performance with greater flexibility.
Best practice for indexingImpactHow to implement
Use a primary index instead of a SORTKEY.Reduces scan times and improves query filtering.Create primary indexes on columns used for filtering and range queries.
Replace DISTKEY with an aggregating index.Speeds up aggregations without manual data distribution.Create aggregating indexes for precomputed sums or counts on frequently grouped columns.
The following code example shows a Redshift schema that creates a sales table that distributes data by customer_id (DISTKEY) for optimized joins, sorts rows by sale_date (SORTKEY) for faster range queries, and defines a customers table with customer_id as the primary key for relational integrity. In Redshift, the primary key is informational only and not enforced at the database level: 
CREATE TABLE sales (
    sale_id INT,
    customer_id INT,
    product_id INT,
    sale_date DATE,
    total_amount DECIMAL(10,2)
) 
DISTSTYLE KEY 
DISTKEY (customer_id) 
SORTKEY (sale_date);

CREATE TABLE customers (
    customer_id INT PRIMARY KEY,
    customer_name TEXT
);
The following code example shows the Firebolt equivalent schema using primary and aggregating indexes: 
CREATE FACT TABLE sales (
    sale_id INT,
    customer_id INT,
    product_id INT,
    sale_date DATE,
    total_amount DECIMAL(10,2)
) 
PRIMARY INDEX (sale_date);

CREATE AGGREGATING INDEX sales_summary 
ON sales (
    customer_id,
    SUM(total_amount)
);

CREATE DIMENSION TABLE customers (
    customer_id INT,
    customer_name TEXT
) 
PRIMARY INDEX (customer_id);
In the previous code example, the following apply:
  • DISTKEY(customer_id) is replaced with the aggregating index (customer_id, SUM(total_amount)). This aggregation runs faster than Redshift because SUM(total_amount) is precomputed in the aggregating index.
  • SORTKEY(sale_date) is replaced with the primary index (sale_date).
In the following example, the DISTKEY in Redshift partitions data across nodes by game_id. Firebolt’s primary index on GameID and PlayerID allows the engine to quickly prune data during query processing, achieving similar optimization without manual distribution. In Redshift, the following schema uses DISTKEY and SORTKEY for the playstats table: 
CREATE TABLE playstats (
    game_id INT,
    player_id INT,
    score BIGINT,
    play_time INT
)
DISTSTYLE KEY
DISTKEY (game_id)
SORTKEY (game_id, player_id);
In Firebolt, the following code example creates a primary index, which is a sparse index, to filter on GameID or PlayerID so that the query scans over only this index: 
CREATE FACT TABLE playstats (
    GameID INT,
    PlayerID INT,
    CurrentScore BIGINT,
    PlayTime INT
) 
PRIMARY INDEX (GameID, PlayerID);
In Firebolt, the following code example uses an aggregating index to return results instead of processing calculations at runtime: 
CREATE AGGREGATING INDEX playtime_agg_idx 
ON PlayStats (
    GameID,
    SUM(CurrentPlayTime)
);

​
Denormalization

Firebolt has denormalized schema. In Redshift, schemas are often highly normalized to reduce redundancy and optimize joins. However, in Firebolt, denormalization improves performance by reducing the need for complex joins and allowing queries to retrieve data more efficiently. Since Firebolt’s indexing model automatically optimizes filtering and aggregations, denormalized schemas can perform better without increasing redundancy-related costs.
Best practice for denormalizationImpactHow to implement
Denormalize frequently joined tables.Reduces query complexity and improves performance.Store commonly joined attributes directly in fact tables.
Avoid excessive joins.Speeds up analytical queries and reduces resource usage.Move frequently accessed reference data into the same table.
Balance redundancy and maintenance.Prevents excessive data duplication while simplifying queries.Only denormalize attributes used in frequent filters or aggregations.
In Redshift, retrieving customer names for each order typically requires joining separate orders and customers tables as follows: 
CREATE TABLE orders (
    order_id INT PRIMARY KEY,
    customer_id INT,
    order_date DATE,
    total_amount DECIMAL(10,2)
);

CREATE TABLE customers (
    customer_id INT PRIMARY KEY,
    customer_name TEXT
);
In Firebolt, denormalization removes the need for this join by storing the customer_name directly in the orders table: 
CREATE FACT TABLE orders (
    order_id INT,
    customer_id INT,
    customer_name TEXT,
    order_date DATE,
    total_amount DECIMAL(10,2)
);
Now, the same query in Firebolt does not require a join, making it simpler and faster: 
SELECT order_id, order_date, total_amount, customer_name
FROM orders
WHERE order_date >= '2024-01-01';
In the previous code example, the customer_name column is stored directly in the orders table, eliminating the need for a join and reducing query complexity.

​
Handling JSON

Firebolt handles JSON differently than Redshift. Firebolt improves the handling of JSON data by storing it as TEXT, allowing dynamic field extraction without predefined schemas. Firebolt includes specialized JSON functions to extract, convert, and manipulate JSON data stored in TEXT columns, enabling flexible and efficient querying without predefined structures. Unlike Redshift’s SUPER type, which requires fixed structures and functions like JSON_EXTRACT_PATH_TEXT, Firebolt enables flexible querying with JSON_VALUE, JSON_EXTRACT, and JSON_POINTER_EXTRACT_TEXT. These functions retrieve only the necessary JSON fields, reducing scan time and improving performance. Firebolt’s approach removes the need for manual schema updates, simplifies query logic, and supports dynamic key lookups without restructuring the data model. Queries are easier to write and maintain because Firebolt avoids the nested function calls required in Redshift. By replacing SUPER with TEXT, users gain a more efficient and flexible way to query JSON data.
Best practice for handling JSONImpactHow to implement
Store JSON data as TEXTEnables flexible querying without predefined schemas.Export JSON columns from Redshift as TEXT and define them as TEXT in Firebolt.
Use JSON_VALUE for simple field extractionSimplifies queries and reduces scan time.Extract single fields directly with JSON_VALUE.
Use JSON_POINTER_EXTRACT_TEXT for nested fieldsAllows access to deeply nested fields in JSON documents.Use JSON_POINTER_EXTRACT_TEXT with JSON pointer expressions.
Use dynamic path construction with CONCAT when neededSupports querying dynamic keys without fixed paths.Build JSON pointer paths using CONCAT for dynamic extraction.
In Redshift, tables may store JSON data using the SUPER type as follows: 
CREATE TABLE events (
    event_id INT PRIMARY KEY,
    event_data SUPER
);
When migrating to Firebolt, export the JSON data from Redshift as TEXT, by unloading to Amazon S3 in JSON or CSV format. Then, define the column in Firebolt as TEXT as follows: 
CREATE FACT TABLE events (
    event_id INT,
    event_data TEXT
);
The previous code example allows you to load JSON content into Firebolt as plain text and query it using Firebolt’s JSON functions. Extracting JSON in Redshift versus Firebolt In Redshift, extracting a JSON field requires fixed paths and multiple function calls for nested structures as follows: 
SELECT JSON_EXTRACT_PATH_TEXT(event_data, 'user_id') AS user_id
FROM events;
For nested JSON fields, Redshift queries become more complex as shown in the following code example: 
SELECT JSON_EXTRACT_PATH_TEXT(JSON_EXTRACT_PATH_TEXT(event_data, 'metadata'), 'user_id') AS user_id
FROM events;
Firebolt simplifies this by allowing direct field extraction with JSON_VALUE as follows: 
SELECT JSON_VALUE(event_data, 'user_id') AS user_id
FROM events;
For nested fields, Firebolt uses a single function call as follows: 
SELECT JSON_VALUE(JSON_POINTER_EXTRACT_TEXT(event_data, '/metadata/user/id')) AS user_id
FROM events;
JSON pointer expressions navigate JSON documents, allowing access to nested elements and arrays by specifying the path to the desired value. Firebolt also supports dynamic key lookups without predefined paths. The following example shows dynamic path construction with CONCAT: 
SELECT JSON_VALUE(JSON_POINTER_EXTRACT_TEXT(event_data, CONCAT('/users/', user_index, '/id'))) AS dynamic_user_id
FROM events;

​
Data types

Firebolt handles JSON differently than Redshift. Redshift and Firebolt support similar data types but with key differences that affect schema design and migration. Redshift uses SUPER for semi-structured data and fixed-length CHAR or VARCHAR columns, while Firebolt stores JSON as TEXT and uses TEXT for variable-length strings without length restrictions. Understanding these differences helps ensure consistent schema conversion and accurate query results when migrating.
Best practice for data typesImpactHow to implement
Use TEXT instead of fixed-length CHAR or VARCHARAvoids truncation issues and simplifies schema management.Define string columns as TEXT without length constraints.
Convert SUPER columns to TEXTEnables flexible JSON handling.Store JSON data as TEXT and query using JSON functions.
Review numeric precision and scalePrevents rounding or overflow errors.Map NUMERIC(p, s) in Redshift to DECIMAL(p, s) in Firebolt.
Validate date and timestamp usageEnsures compatibility for temporal queries.Use DATE and TIMESTAMP consistently, without time zones.
The following table shows common Redshift data types and their Firebolt equivalents to help you adjust column definitions during schema migration:
Redshift typeFirebolt equivalentNotes
SMALLINTSMALLINTIdentical.
INTEGERINT or INTEGERFirebolt uses INT for integer types.
BIGINTBIGINTIdentical.
REALFLOAT4Alias mapping for single-precision floats.
DOUBLE PRECISIONDOUBLEUse DOUBLE for double-precision floats.
NUMERIC(p, s)DECIMAL(p, s)Equivalent; Firebolt uses DECIMAL.
BOOLEANBOOLEANIdentical.
CHAR(n) or VARCHAR(n)TEXTFirebolt uses TEXT with no length restrictions.
DATEDATEIdentical.
TIMESTAMPTIMESTAMPFirebolt supports TIMESTAMP without time zone.
JSONTEXTJSON is stored as TEXT and queried using JSON functions.
ARRAYARRAY(TEXT)Firebolt arrays require explicit type definitions.
Example schema conversion The following example shows how to convert a Redshift table to a Firebolt dimension table with type adjustments and a primary index: In Redshift, the following code example creates a players table with columns for player_id, nickname, registration date, and score using specified data types: 
CREATE TABLE players (
    player_id INTEGER,
    nickname VARCHAR(50),
    registered_on DATE,
    score NUMERIC(10, 2)
);
In Firebolt, the following code example creates a players dimension table with columns for PlayerID, Nickname, registration date, and Score, and defines a primary index on PlayerID to optimize filtering and lookups.: 
CREATE DIMENSION TABLE players (
    PlayerID INT,
    Nickname TEXT,
    RegisteredOn DATE,
    Score DECIMAL(10, 2)
) PRIMARY INDEX PlayerID;

​
Steps to migrate from Redshift to Firebolt

  1. Export data from Redshift
  2. Load data into Firebolt
  3. Translate queries into Firebolt
  4. Validation, testing, and optimization
  5. Automated migration

​
Export data from Redshift

The first step to migrate to Firebolt is to export data from Redshift. Before migrating to Firebolt, export your Redshift data to Amazon S3. Redshift supports efficient parallel unloading of data in various formats such as CSV or Parquet, which Firebolt can load directly. Follow these steps to export data:
  1. Use Redshift’s UNLOAD command to export tables or query results to Amazon S3. The following code example unloads the sales table data in CSV format to the specified Amazon S3 bucket: 
    UNLOAD ('SELECT * FROM sales')
TO 's3://your-redshift-data/sales/'
CREDENTIALS 'aws_access_key_id=your-access-key;aws_secret_access_key=your-secret-key'
PARALLEL OFF
ALLOWOVERWRITE
DELIMITER ',';
  2. Format data for Firebolt as follows:
    • Ensure that exported data is in CSV or Parquet format.
    • Include headers in CSV files for clarity during loading.
    • Use Parquet with Snappy compression for faster loading.
    • Compress CSV files with gzip compression to reduce storage and speed up transfer.
    • Use consistent delimiters and avoid special characters that may cause parsing errors.
    • Validate that exported data types match the expected Firebolt schema.
  3. Organize files in Amazon S3 for parallel loading.
    • Store exported files in Amazon S3 folders using clear prefixes, for example: s3://your-redshift-data/sales/2024/01/.
    • Use multiple small-to-medium sized files of 100 MB to 1 GB each to maximize parallel loading in Firebolt.
    • Follow consistent naming conventions including table names, export dates, or partitions.
    • Clean up any incomplete or partial exports to avoid loading invalid data.

​
Load data into Firebolt

The second step to migrate to Firebolt is to load data into Firebolt. Firebolt loads data directly from AWS S3 using the COPY FROM command, supporting parallel ingestion, automatic schema discovery, metadata handling, and error management. It works with both CSV and Parquet formats, with Parquet recommended for efficient compression. You can use either the SQL Workspace or the Load data wizard to load data. For more information, see Load data. The following code example loads data from Parquet files stored in an S3 bucket into the tournaments table in Firebolt, mapping specific columns from the file to the table’s fields and including metadata such as the source file name and timestamp: 
COPY INTO tournaments (
  tournamentid $1, 
  name $2, 
  gameid $3, 
  totalprizedollars $4, 
  startdatetime $5, 
  enddatetime $6, 
  rulesdefinition $7, 
  source_file_name $SOURCE_FILE_NAME, 
  source_file_timestamp $SOURCE_FILE_TIMESTAMP)
FROM 's3://firebolt-sample-datasets-public-us-east-1/gaming/parquet/tournaments/' 
WITH PATTERN = '*' 
TYPE = PARQUET;
For more information, see Load data.

​
Translate queries into Firebolt

The third step to migrate to Firebolt is to translate queries into Firebolt. Migrating queries from Redshift to Firebolt requires more than just adjusting syntax; it involves utilizing Firebolt’s advanced query optimization features to enhance performance. While the SQL syntax in both platforms is largely similar, Firebolt’s powerful indexing, advanced functions, and query execution insights require careful adaptation of Redshift queries. This section is divided into the following:
  • Window functions – Learn how to translate window functions from Redshift to Firebolt and optimize them using Firebolt’s indexing features for improved performance.
  • Aggregation queries – Discover how to handle aggregation queries, leveraging Firebolt’s aggregating indexes to precompute results and significantly speed up query performance.
  • JSON data – Understand how to work with JSON in Firebolt using its specialized functions, providing more flexibility and efficiency compared to Redshift’s approach.
  • Query plan insights – Gain insights into query plans and how Firebolt’s EXPLAIN (ANALYZE) command helps optimize queries through runtime metrics for optimal performance.

​
Translate window functions into Firebolt

The third step to migrate to Firebolt includes translating window queries into Firebolt. Window functions in Firebolt are compatible with standard SQL, allowing users to compute values across a set of rows related to the current row. Both Redshift and Firebolt support common window functions like ROW_NUMBER, RANK, and SUM with the OVER clause, but Firebolt can enhance performance through indexing, particularly primary indexes, which minimize the data that needs to be processed. The following example shows how to calculate a running total of player scores by game in both Redshift and Firebolt, using the SUM window function. In Redshift, the query to calculate the running total of the currentscore for each playerid within each gamid ordered by playerid, using a window function: 
SELECT gameid, playerid, 
SUM(currentscore) OVER (PARTITION BY gameid ORDER BY playerid) 
AS running_total 
FROM PlayStats;
The same query can be run in Firebolt with no changes, as Firebolt supports standard SQL window functions. Primary indexes help Firebolt engines automatically prune tablets, reducing the data volume scanned by window functions. The following code example creates a PlayStats table in Firebolt with columns for gameid, playerid, and currentscore. It defines a primary index on gameid and playerid to optimize query performance, such as the running total calculation, by efficiently retrieving the relevant rows based on these columns: 
CREATE FACT TABLE PlayStats (
  gameid INT,
  playerid INT,
  currentscore BIGINT
) 
PRIMARY INDEX (gameid, playerid);

​
Translate aggregation queries into Firebolt

The third step to migrate to Firebolt includes translating aggregate queries into Firebolt. Aggregation queries in Firebolt are similar to those in Redshift, but Firebolt offers better performance, especially for large datasets, by utilizing aggregating indexes. Both platforms support common aggregation functions like SUM, COUNT, and AVG, and the GROUP BY clause. Firebolt, however, enhances performance by precomputing and storing aggregated results through aggregating indexes, which reduce the need for runtime calculations. The following example shows how to calculate the total playtime by game in both Redshift and Firebolt using the SUM aggregation function. In Redshift, the following code calculates the total currentplaytime for each gameid in the PlayStats table by grouping the data by gameid and summing the currentplaytime for each group: 
SELECT gameid, SUM(currentplaytime) 
AS total_playtime 
FROM PlayStats 
GROUP BY gameid;
The same query can be run in Firebolt with no changes, as Firebolt supports standard SQL aggregation functions. To optimize performance, create an aggregating index in Firebolt that precomputes the total playtime for each gameid, avoiding runtime aggregation. The following code example creates an aggregating index on the PlayStats table, precomputing the sum of currentplaytime for each gameid, which optimizes query performance by storing the aggregated results for faster retrieval: 
CREATE AGGREGATING INDEX playtime_agg_idx 
ON PlayStats (
  gameid,
  SUM(currentplaytime)
);
With this aggregating index, Firebolt retrieves precomputed results instead of performing the aggregation during query execution, significantly reducing query time.

​
JSON data in Firebolt

The third step to migrate to Firebolt includes translating JSON from Redshift into Firebolt. Firebolt provides a flexible and efficient way to handle JSON data, offering specialized functions to extract, convert, and manipulate JSON stored in TEXT columns. Unlike Redshift’s SUPER data type, which requires predefined structures, Firebolt allows users to work with raw JSON data directly, offering significant advantages in both performance and flexibility. Firebolt’s JSON functions are divided into three main categories:
  • Extract functions – Used to retrieve specific parts of a JSON document.
  • Convert functions – Used to convert JSON data into SQL-compatible types like TEXT or ARRAY.
  • Hybrid functions – Combine extraction and conversion in one step, especially useful for nested JSON elements.
Additionally, Firebolt handles escaped characters using specialized functions. Extract functions Extract functions in Firebolt are designed to extract specific parts of a JSON document while preserving the original data format. Commonly used extract functions include the following: 
SELECT JSON_EXTRACT(event_data, '$.user_id') AS user_id
FROM events;
Convert functions Convert functions are used to convert JSON values into SQL-compatible types, such as TEXT or ARRAY. Key convert functions include the following: 
SELECT JSON_VALUE(event_data, 'user_id') AS user_id
FROM events;
Hybrid functions Hybrid functions, like JSON_POINTER_EXTRACT_TEXT, combine both extraction and conversion, allowing access to nested JSON values. Hybrid functions in Firebolt combine extraction and conversion in a single operation. The following code example extracts the raw JSON value of the id key from the nested metadata object in the event_data JSON column of the events table and returns it as user_id: 
SELECT JSON_POINTER_EXTRACT_TEXT(event_data, '/metadata/user/id') AS user_id
FROM events;
You can also use JSON_POINTER_EXTRACT_TEXT to extract values from nested JSON objects using JSON pointer expressions as follows. The following code example extracts the value of the id key from the nested metadata object in the event_data JSON column of the events table and returns it as user_id: 
SELECT JSON_VALUE(JSON_POINTER_EXTRACT_TEXT(event_data, '/metadata/user/id')) AS user_id
FROM events;
When dealing with JSON keys that contain special characters such as tilde ~ or forward slash /, Firebolt uses the standard JSON Pointer specification for escaping characters: 0 represents the tilde (). ~1 represents the forward slash (/). The following code example extracts values from the event_data JSON column in the events table, accessing keys with the tilde and slash special characters by using escape sequences and returning them as key_with_tilde and key_with_slash: 
SELECT JSON_VALUE(JSON_POINTER_EXTRACT_TEXT(event_data, '/key~0with~0tilde')) AS key_with_tilde,
       JSON_VALUE(JSON_POINTER_EXTRACT_TEXT(event_data, '/key~1with~1slash')) AS key_with_slash
FROM events;

​
Firebolt’s query plan insights

The third step to migrate to Firebolt includes optimizing query plans in Firebolt. Firebolt provides advanced query plan insights that help optimize query performance. Unlike Redshift, which relies on static query plans, Firebolt offers dynamic plans with real-time runtime metrics when using the EXPLAIN (ANALYZE) command. This allows for deeper insights into how queries are run and where performance improvements can be made. In Redshift, the EXPLAIN command provides a static query plan, showing the sequence of operations, estimated costs, and resource allocation. This plan lacks runtime performance metrics and detailed insights into data movement across nodes. In Redshift, the following code example analyzes a join query: 
EXPLAIN 
SELECT eventid, eventname, event.venueid, venuename
FROM event
JOIN venue ON event.venueid = venue.venueid;
The output from running the previous code example in Redshift highlights the sequence of operations and the relative cost estimates of the query: 
XN Hash Join DS_DIST_OUTER  (cost=2.52..58653620.93 rows=8712 width=43)
Hash Cond: ("outer".venueid = "inner".venueid)
->  XN Seq Scan on event  (cost=0.00..87.98 rows=8798 width=23)
->  XN Hash  (cost=2.02..2.02 rows=202 width=22)
->  XN Seq Scan on venue  (cost=0.00..2.02 rows=202 width=22)
In Firebolt, the EXPLAIN command provides two key plans:
  • The logical plan outlines the structure of the query, including projections, filters, and joins.
  • The physical plan reveals how these operations are distributed across nodes, with details on data shuffling and execution strategies.
When paired with the ANALYZE option, Firebolt’s EXPLAIN command adds runtime metrics, such as CPU time, thread time, rows processed, and runtime for each operation, allowing for detailed performance analysis. In Firebolt, the following code example analyzes the same query and provides runtime metrics: 
EXPLAIN (ANALYZE)
SELECT eventid, eventname, event.venueid, venuename
FROM event
JOIN venue ON event.venueid = venue.venueid;
The output from running the previous code example in Firebolt shows output cardinality, thread time, and CPU time for each step: 
[0] [Projection] event.eventid, event.eventname, event.venueid, venue.venuename
 |   [Execution Metrics]: output cardinality = 10000, thread time = 3ms, cpu time = 2ms
 \_[1] [HashJoin] event.venueid = venue.venueid
    |   [Execution Metrics]: output cardinality = 10000, thread time = 10ms, cpu time = 8ms
    \_[2] [StoredTable] event
    |   [Execution Metrics]: output cardinality = 5000, thread time = 5ms, cpu time = 4ms
    \_[3] [StoredTable] venue
        [Execution Metrics]: output cardinality = 5000, thread time = 5ms, cpu time = 4ms

​
Validation, testing, and optimization

The fourth step to migrate to Firebolt is to validate data, test, and optimize engines in Firebolt. Validation and performance testing are critical steps when migrating from Redshift to Firebolt. Before optimizing for performance, validate schema changes and data consistency to ensure that queries and aggregations behave as expected. Once data consistency is confirmed, performance testing helps identify bottlenecks, resource inefficiencies, and areas for improvement.

​
Validate data consistency

The fourth step to migrate to Firebolt includes validating data consistency. Differences in data types, indexing, and table design require careful verification to avoid mismatches and unexpected results after migration. Compare row counts, aggregates, query results, and checksums between Redshift and Firebolt to ensure consistency. Check that counts, aggregates, and indexed performance match across systems. Row counts and aggregates Run queries in both Redshift and Firebolt to compare row counts and aggregated totals: The following code example checks the number of rows in the playstats table in both Redshift and Firebolt: 
SELECT COUNT(*) FROM playstats;
SELECT SUM(currentscore) AS total_score FROM playstats;
Grouped comparisons The following code example checks the aggregate sum in both Redshift and Firebolt: 
SELECT gameid, playerid, SUM(currentscore) AS total_score 
FROM playstats
GROUP BY gameid, playerid;
Validate indexing efficiency with query plans The following code example compares specific fields using GROUP BY in both Redshift and Firebolt: 
SELECT gameid, playerid, SUM(currentscore) AS total_score 
FROM playstats
GROUP BY gameid, playerid;
Checksum validation – Use checksums to compare the integrity of data between Redshift and Firebolt. This method aggregates values and applies an MD5 hash to ensure the data is identical across both systems. The following code example checks a hash value of all columns in Redshift: 
SELECT MD5(STRING_AGG(CONCAT(gameid, ',', playerid, ',', currentscore), '')) AS table_checksum
FROM playstats;
The following code example checks the equivalent hash value in Firebolt: 
SELECT HASH_AGG(*) AS table_checksum
FROM playstats;
Test edge cases – Test how Firebolt handles edge cases like NULL values, large datasets, and complex joins. This ensures that data migration doesn’t introduce errors in rare or complex scenarios. The following code example shows complex joins from multiple tables in Redshift and Firebolt: 
SELECT e.eventid, e.eventname, v.venuename
FROM event e
JOIN venue v ON e.venueid = v.venueid;
The following code example checks how NULL values are handled in both Redshift and Firebolt: 
SELECT COUNT(*) FROM playstats WHERE currentscore IS NULL;
Verify data types The following code example checks data types in Redshift for text and integer columns: 
SELECT DISTINCT column_name, data_type
FROM information_schema.columns
WHERE table_name = 'playstats';
The following code example checks data types in Firebolt: 
DESCRIBE playstats;
Once data consistency is confirmed, check query runtimes, resource utilization, and scalability to ensure optimal query performance: Validate indexing efficiency with query plans – Measure both the time it takes for queries to run and how much CPU, memory, and disk are consumed during query runtime to help identify performance bottlenecks and optimize resource usage. Test the runtime time for common queries, as well as more complex queries involving joins and aggregations. The following code example measures the estimated runtime, memory size used, CPU utilization, and query queue time for a simple query in Redshift: 
EXPLAIN
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM playstats
GROUP BY gameid;

-- Monitor resource usage during query execution
SELECT pid, user_name, start_time, query, db, state, 
       total_queue_time, total_exec_time, memory_size
FROM stv_recents
WHERE query = (SELECT max(query) FROM stv_recents);
The following code example measures the actual runtime, CPU time, thread time, and data processed for each operator in the same query in Firebolt: 
EXPLAIN (ANALYZE)
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM playstats
GROUP BY gameid;
Scalability – Test how Firebolt handles larger datasets or higher concurrency. This includes simulating higher data volumes or increased user loads to test how well Firebolt scales under pressure. Simulating larger datasets also tests engine stability and query planner behavior under stress conditions. The dataset example in this section duplicates rows to simulate large datasets to observe the impact on performance. Change the code to better emulate your actual workloads. The following code uses UNION ALL to increase the size of the playstats table in Redshift and measures the runtime and resource usage: 
EXPLAIN
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM (
    SELECT * FROM playstats
    UNION ALL
    SELECT * FROM playstats  -- Duplicate the rows to simulate a larger dataset
) AS playstats
GROUP BY gameid;
The following code example uses the same approach to duplicate the dataset and measures the runtime and resource utilization in Firebolt: 
EXPLAIN (ANALYZE)
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM (
    SELECT * FROM playstats
    UNION ALL
    SELECT * FROM playstats  -- Duplicate the rows to simulate a larger dataset
) AS playstats
GROUP BY gameid;

​
Performance testing

The fourth step to migrate to Firebolt includes performance testing. Check how queries perform with different engine configurations to help identify the optimal engine size for your workload and ensure that Firebolt is scaled correctly for your needs. In Redshift, engine configurations are set at the cluster level rather than for individual queries, so changing the engine configuration requires resizing the cluster. You can benchmark queries after resizing your cluster to identify the best configuration for your workload. Use EXPLAIN and STV_RECENTS to evaluate query performance and resource consumption. The following code example measures query runtime, resource usage, and queue time in Firebolt. Rerun it using different cluster sizes: 
EXPLAIN
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM playstats
GROUP BY gameid;

​
Engine optimization

The fourth step to migrate to Firebolt includes engine optimization. In Firebolt, you can easily test different engine sizes for your queries by creating engines with varying configurations and compare the query performance. The following code example creates a small and large engine, and then specifies which engine runs the same query to measure performance in Firebolt: 
-- Create a small engine
CREATE ENGINE "small_engine" WITH
TYPE = "S"
NODES = 2;

-- Create a large engine
CREATE ENGINE "large_engine" WITH
TYPE = "L"
NODES = 8;

-- Activate the small engine for running the query
SELECT * FROM system_engines WHERE engine_name = 'small_engine';

EXPLAIN (ANALYZE)
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM playstats
GROUP BY gameid;

-- To run on the large engine, activate that engine and then run the query
SELECT * FROM system_engines WHERE engine_name = 'large_engine';

EXPLAIN (ANALYZE)
SELECT gameid, SUM(currentplaytime) AS total_playtime
FROM playstats
GROUP BY gameid;

​
Automate migration

You can optionally set up an automated migration to Firebolt using either Airflow or dbt. Automating the migration process from Redshift to Firebolt can streamline the transition, reduce manual effort, and ensure consistency. It can also reduce errors, increase consistency, and support synchronizing dynamic datasets. By leveraging modern data engineering tools like dbt (Data Build Tool) and Apache Airflow, you can automate the critical steps of data extraction, schema conversion, query translation, and performance optimization.

​
Automate data migration with Airflow

You can optionally automate your migration using Airflow. You can automate the data migration process from Redshift to Firebolt using Apache Airflow, which helps you create workflows to export data from Redshift, store it in S3, and load it into Firebolt. Before automating your migration with Airflow, you’ll need set aws_access_key_id, aws_secret_access_key, and s3_redshift_data_bucket variables for your migration tasks. For more information, see the Apache Airflow documentation on variables. Use Airflow to run Redshift’s UNLOAD command on a scheduled basis to export data from Redshift tables to Amazon S3 as follows: 
unload_task = SQLExecuteQueryOperator(
    task_id='unload_data_from_redshift',
    sql="""
    UNLOAD ('SELECT * FROM your_table')
    TO '{{ var.value.s3_redshift_data_bucket }}/your_table/'
    CREDENTIALS 'aws_access_key_id={{ var.value.aws_access_key_id }};aws_secret_access_key={{ var.value.aws_secret_access_key }}
    PARALLEL OFF
    DELIMITER ',';
    """,
    conn_id='redshift_connection',
    dag=dag
)
Once the data is in Amazon S3, use another Airflow task to run Firebolt’s COPY FROM command to load the data into your Firebolt tables as follows: 
load_task = FireboltOperator(
    task_id='load_data_to_firebolt',
    sql="""
    COPY INTO your_table
    FROM '{{ var.value.s3_redshift_data_bucket }}/your_table/'
    WITH
    TYPE = CSV
    CREDENTIALS = (AWS_ACCESS_KEY_ID='{{ var.value.aws_access_key_id }}' AWS_SECRET_ACCESS_KEY='{{ var.value.aws_secret_access_key }}');
    """,
    firebolt_conn_id='firebolt_connection',
    dag=dag
)
After migrating the schema and data, test the performance of different Firebolt engine configurations to ensure optimal query performance. Use Airflow to automate performance testing by running queries across various engine sizes and comparing the results to help you identify the best engine size for your workload. The following code example runs a query in Firebolt: 
# Define the performance test function
def run_performance_test(**context):
    """Run performance test on Firebolt."""
    try:
        # Replace the query with the one you want to test
        sql_to_test = "SELECT * FROM your_table"

        sql_to_run = """
        EXPLAIN (ANALYZE)
        SELECT * FROM your_table
        """
        hook = FireboltHook(firebolt_conn_id='firebolt_connection')
        result = hook.get_first(sql_to_run)[0]
        logging.info("Performance test result: %s", result)
    except Exception as e:
        logging.error("Error in performance test: %s", str(e))
        raise

performance_test = PythonOperator(
    task_id='run_performance_test',
    python_callable=run_performance_test,
    provide_context=True,
    dag=dag
)
You can also orchestrate your entire migration pipeline using a DAG (Directed Acyclic Graph) that includes tasks for exporting data, loading it into Firebolt, transforming the schema, and testing performance. The following code example uses a DAG to automate migrating data, transforming schema and testing performance using Airflow: 
from airflow import DAG
from airflow.providers.common.sql.operators.sql import SQLExecuteQueryOperator
from firebolt_provider.operators.firebolt import FireboltOperator
from firebolt_provider.hooks.firebolt import FireboltHook
from airflow.operators.python import PythonOperator
from datetime import datetime, timedelta
import logging

# Define the dbt model execution function
def run_dbt_model(**context):
    """Execute dbt models for schema conversion and data transformation."""
    try:
        # Add your dbt execution logic here
        logging.info("Starting dbt model execution")
        # Example: subprocess.run(['dbt', 'run', '--models', 'your_model'])
        logging.info("dbt model execution completed successfully")
    except Exception as e:
        logging.error("Error in dbt model execution: %s", e)
        raise

# Define the performance test function
def run_performance_test(**context):
    """Run performance test on Firebolt."""
    try:
        # Replace the query with the one you want to test
        sql_to_test = "SELECT * FROM your_table"

        sql_to_run = """
        EXPLAIN (ANALYZE)
        SELECT * FROM your_table
        """
        hook = FireboltHook(firebolt_conn_id='firebolt_connection')
        result = hook.get_first(sql_to_run)[0]
        logging.info("Performance test result: %s", result)
    except Exception as e:
        logging.error("Error in performance test: %s", str(e))
        raise

default_args = {
    'owner': 'airflow',
    'depends_on_past': False,
    'start_date': datetime(2025, 1, 1),
    'email_on_failure': False,
    'email_on_retry': False,
}

dag = DAG(
    'redshift_to_firebolt_migration',
    default_args=default_args,
    schedule_interval='@daily',
    catchup=False,
    tags=['migration', 'redshift', 'firebolt']
)

unload_task = SQLExecuteQueryOperator(
    task_id='unload_data_from_redshift',
    sql="""
    UNLOAD ('SELECT * FROM your_table')
    TO '{{ var.value.s3_redshift_data_bucket }}/your_table/'
    CREDENTIALS 'aws_access_key_id={{ var.value.aws_access_key_id }};aws_secret_access_key={{ var.value.aws_secret_access_key }}
    PARALLEL OFF
    DELIMITER ',';
    """,
    conn_id='redshift_connection',
    dag=dag
)

load_task = FireboltOperator(
    task_id='load_data_to_firebolt',
    sql="""
    COPY INTO your_table
    FROM '{{ var.value.s3_redshift_data_bucket }}/your_table/'
    WITH
    TYPE = CSV
    CREDENTIALS = (AWS_ACCESS_KEY_ID='{{ var.value.aws_access_key_id }}' AWS_SECRET_ACCESS_KEY='{{ var.value.aws_secret_access_key }}');
    """,
    firebolt_conn_id='firebolt_connection',
    dag=dag
)

dbt_task = PythonOperator(
    task_id='run_dbt_for_schema_conversion',
    python_callable=run_dbt_model,
    provide_context=True,
    dag=dag
)


performance_test = PythonOperator(
    task_id='run_performance_test',
    python_callable=run_performance_test,
    provide_context=True,
    dag=dag
)

# Set task dependencies
unload_task >> load_task >> dbt_task >> performance_test
For more information, see Firebolt’s guide to Connecting to Airflow.

​
Automate data migration with dbt

You can optionally automate your migration using dbt. You can use dbt to automate the generation of Firebolt schemas and adjust Redshift queries for Firebolt’s indexing features. Use dbt to convert Redshift’s SORTKEY and DISTKEY into Firebolt’s primary and aggregating indexes. Create dbt models to define Firebolt-compatible schemas and apply them automatically. The following code example uses dbt to create a Firebolt schema: 
-- Redshift to Firebolt schema conversion
CREATE FACT TABLE playstats (
  gameid INT,
  playerid INT,
  currentscore BIGINT
)
PRIMARY INDEX (gameid, playerid);
Use dbt to replace Redshift-specific functions with Firebolt equivalents, like converting SUPER to TEXT for JSON handling as follows: 
-- Redshift JSON function converted to Firebolt
SELECT JSON_VALUE(event_data, 'user_id') AS user_id
FROM events;
For more information, see Firebolt’s documentation to integrate with dbt.