Skip to content

configuration.md

Latest commit

 

History

History
1536 lines (1077 loc) · 62 KB

configuration.md

File metadata and controls

1536 lines (1077 loc) · 62 KB

Configuration guide

SQLMesh's behavior is determined by three things: a project's files (e.g., models), user actions (e.g., creating a plan), and how SQLMesh is configured.

This page describes how SQLMesh configuration works and discusses the aspects of SQLMesh behavior that can be modified via configuration.

The configuration reference page contains concise lists of all configuration parameters and their default values.

Configuration files

NOTE: SQLMesh project configurations have the following two requirements:

  1. A config.yaml or config.py file must be present in the project's folder.
  2. That configuration file must contain a default SQL dialect for the project's models in the model_defaults dialect key.

SQLMesh configuration parameters can be set as environment variables, in a configuration file in the ~/.sqlmesh folder, and in the configuration file within a project folder.

The sources have the following order of precedence:

  1. Environment variable (e.g., SQLMESH__MODEL_DEFAULTS__DIALECT). [HIGHEST PRECEDENCE]
  2. config.yaml or config.py in the ~/.sqlmesh folder.
  3. config.yaml or config.py in a project folder. [LOWEST PRECEDENCE]

!!! note To relocate the .sqlmesh folder, set the SQLMESH_HOME environment variable to your preferred directory path.

File type

You can specify a SQLMesh configuration in either YAML or Python.

YAML configuration is simpler, and we recommend it for most projects. Python configuration is more complex, but it enables functionality that YAML does not support.

Because Python configuration files are evaluated by Python when SQLMesh reads them, they support dynamic parameters based on the computational environment in which SQLMesh is running.

For example, Python configuration files enable use of third-party secrets managers for storing passwords and other sensitive information. They also support user-specific parameters such as automatically setting project defaults based on which user account is running SQLMesh.

YAML

YAML configuration files consist of configuration keys and values. Strings are not quoted, and some keys are "dictionaries" that contain one or more sub-keys.

For example, the default_gateway key specifies the default gateway SQLMesh should use when executing commands. It takes a single, unquoted gateway name as its value:

default_gateway: local

In contrast, the gateways key takes dictionaries as values, and each gateway dictionary contains one or more connection dictionaries. This example specifies the my_gateway gateway with a Snowflake connection:

gateways:
  my_gateway:
    connection:
      type: snowflake
      user: <username>
      password: <password>
      account: <account>

Gateway dictionaries can contain multiple connection dictionaries if different SQLMesh components should use different connections (e.g., SQLMesh tests should run in a different database than SQLMesh plans). See the gateways section for more information on gateway configuration.

Python

Python configuration files consist of statements that import SQLMesh configuration classes and a configuration specification using those classes.

At minimum, a Python configuration file must:

  1. Create an object of the SQLMesh Config class named config
  2. Specify that object's model_defaults argument with a ModelDefaultsConfig() object specifying the default SQL dialect for the project's models

For example, this minimal configuration specifies a default SQL dialect of duckdb and uses the default values for all other configuration parameters:

from sqlmesh.core.config import Config, ModelDefaultsConfig

config = Config(
    model_defaults=ModelDefaultsConfig(dialect="duckdb"),
)

Python configuration files may optionally define additional configuration objects and switch between the configurations when issuing sqlmesh commands. For example, if a configuration file contained a second configuration object my_second_config, you could create a plan using that config with sqlmesh --config my_second_config plan.

Different Config arguments accept different object types. Some, such as model_defaults, take SQLMesh configuration objects. Others, such as default_gateway, take strings or other Python object types like dictionaries.

SQLMesh's Python configuration components are documented in the sqlmesh.core.config module's API documentation.

The config sub-module API documentation describes the individual classes used for the relevant Config arguments:

See the notifications guide for more information about user and notification specification.

Environment variables

All software runs within a system environment that stores information as "environment variables."

SQLMesh can access environment variables during configuration, which enables approaches like storing passwords/secrets outside the configuration file and changing configuration parameters dynamically based on which user is running SQLMesh.

You can specify environment variables in the configuration file or by storing them in a .env file.

.env files

SQLMesh automatically loads environment variables from a .env file in your project directory. This provides a convenient way to manage environment variables without having to set them in your shell.

Create a .env file in your project root with key-value pairs:

# .env file
SNOWFLAKE_PW=my_secret_password
S3_BUCKET=s3://my-data-bucket/warehouse
DATABASE_URL=postgresql://user:pass@localhost/db

# Override specific SQLMesh configuration values
SQLMESH__DEFAULT_GATEWAY=production
SQLMESH__MODEL_DEFAULTS__DIALECT=snowflake

See the overrides section for a detailed explanation of how these are defined.

The rest of the .env file variables can be used in your configuration files with {{ env_var('VARIABLE_NAME') }} syntax in YAML or accessed via os.environ['VARIABLE_NAME'] in Python.

Custom dot env file location and name

By default, SQLMesh loads .env files from each project directory. However, you can specify a custom path using the --dotenv CLI flag directly when running a command:

sqlmesh --dotenv /path/to/custom/.env plan

!!! note The --dotenv flag is a global option and must be placed before the subcommand (e.g. plan, run), not after.

Alternatively, you can export the SQLMESH_DOTENV_PATH environment variable once, to persist a custom path across all subsequent commands in your shell session:

export SQLMESH_DOTENV_PATH=/path/to/custom/.custom_env
sqlmesh plan
sqlmesh run

Important considerations:

  • Add .env to your .gitignore file to avoid committing sensitive information
  • SQLMesh will only load the .env file if it exists in the project directory (unless a custom path is specified)
  • When using a custom path, that specific file takes precedence over any .env file in the project directory.

Configuration file

This section demonstrates using environment variables in YAML and Python configuration files.

The examples specify a Snowflake connection whose password is stored in an environment variable SNOWFLAKE_PW.

=== "YAML"

Specify environment variables in a YAML configuration with the syntax `{{ env_var('<ENVIRONMENT VARIABLE NAME>') }}`. Note that the environment variable name is contained in single quotes.

Access the `SNOWFLAKE_PW` environment variable in a Snowflake connection configuration like this:

```yaml linenums="1"
gateways:
  my_gateway:
    connection:
      type: snowflake
      user: <username>
      password: {{ env_var('SNOWFLAKE_PW') }}
      account: <account>
```

=== "Python"

Python accesses environment variables via the `os` library's `environ` dictionary.

Access the `SNOWFLAKE_PW` environment variable in a Snowflake connection configuration like this:

```python linenums="1"
import os
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    SnowflakeConnectionConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            connection=SnowflakeConnectionConfig(
                user=<username>,
                password=os.environ['SNOWFLAKE_PW'],
                account=<account>,
            ),
        ),
    }
)
```

Default target environment

The SQLMesh plan command acts on the prod environment by default (i.e., sqlmesh plan is equivalent to sqlmesh plan prod).

In some organizations, users never run plans directly against prod - they do all SQLMesh work in a development environment unique to them. In a standard SQLMesh configuration, this means they need to include their development environment name every time they issue the plan command (e.g., sqlmesh plan dev_tony).

If your organization works like this, it may be convenient to change the plan command's default environment from prod to each user's development environment. That way people can issue sqlmesh plan without typing the environment name every time.

The SQLMesh configuration user() function returns the name of the user currently logged in and running SQLMesh. It retrieves the username from system environment variables like USER on MacOS/Linux or USERNAME on Windows.

Call user() inside Jinja curly braces with the syntax {{ user() }}, which allows you to combine the user name with a prefix or suffix.

The example configuration below constructs the environment name by appending the username to the end of the string dev_. If the user running SQLMesh is tony, the default target environment when they run SQLMesh will be dev_tony. In other words, sqlmesh plan will be equivalent to sqlmesh plan dev_tony.

=== "YAML"

Default target environment is `dev_` combined with the username running SQLMesh.

```yaml
default_target_environment: dev_{{ user() }}
```

=== "Python"

Default target environment is `dev_` combined with the username running SQLMesh.

Retrieve the username with the `getpass.getuser()` function, and combine it with `dev_` in a Python f-string.

```python linenums="1" hl_lines="1 17"
import getpass
import os
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    SnowflakeConnectionConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect="duckdb"),
    gateways={
        "my_gateway": GatewayConfig(
            connection=DuckDBConnectionConfig(),
        ),
    },
    default_target_environment=f"dev_{getpass.getuser()}",
)
```

Overrides

Environment variables have the highest precedence among configuration methods, as noted above. They will automatically override configuration file specifications if they follow a specific naming structure.

The structure is based on the names of the configuration fields, with double underscores __ between the field names. The environment variable name must begin with SQLMESH__, followed by the YAML field names starting at the root and moving downward in the hierarchy.

For example, we can override the password specified in a Snowflake connection. This is the YAML specification contained in our configuration file, which specifies a password dummy_pw:

gateways:
  my_gateway:
    connection:
      type: snowflake
      user: <username>
      password: dummy_pw
      account: <account>

We can override the dummy_pw value with the true password real_pw by creating the environment variable. This example demonstrates creating the variable with the bash export function:

$ export SQLMESH__GATEWAYS__MY_GATEWAY__CONNECTION__PASSWORD="real_pw"

After the initial string SQLMESH__, the environment variable name components move down the key hierarchy in the YAML specification: GATEWAYS --> MY_GATEWAY --> CONNECTION --> PASSWORD.

Configuration types

A SQLMesh project configuration is hierarchical and consists of root level parameters within which other parameters are defined.

Conceptually, we can group the root level parameters into the following types. Each type links to its table of parameters in the SQLMesh configuration reference page:

  1. Project - configuration options for SQLMesh project directories.
  2. Environment - configuration options for SQLMesh environment creation/promotion, physical table schemas, and view schemas.
  3. Gateways - configuration options for how SQLMesh should connect to the data warehouse, state backend, and scheduler.
  4. Gateway/connection defaults - configuration options for what should happen when gateways or connections are not all explicitly specified.
  5. Model defaults - configuration options for what should happen when model-specific configurations are not explicitly specified in a model's file.
  6. Debug mode - configuration option for SQLMesh to print and log actions and full backtraces.

Configuration details

The rest of this page provides additional detail for some of the configuration options and provides brief examples. Comprehensive lists of configuration options are at the configuration reference page.

Cache directory

By default, the SQLMesh cache is stored in a .cache directory within your project folder. You can customize the cache location using the cache_dir configuration option:

=== "YAML"

```yaml linenums="1"
# Relative path to project directory
cache_dir: my_custom_cache

# Absolute path
cache_dir: /tmp/sqlmesh_cache

```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig

config = Config(
    model_defaults=ModelDefaultsConfig(dialect="duckdb"),
    cache_dir="/tmp/sqlmesh_cache",
)
```

The cache directory is automatically created if it doesn't exist. You can clear the cache using the sqlmesh clean command.

Table/view storage locations

SQLMesh creates schemas, physical tables, and views in the data warehouse/engine. Learn more about why and how SQLMesh creates schema in the "Why does SQLMesh create schemas?" FAQ.

The default SQLMesh behavior described in the FAQ is appropriate for most deployments, but you can override where SQLMesh creates physical tables and views with the physical_schema_mapping, environment_suffix_target, and environment_catalog_mapping configuration options.

You can also override what the physical tables are called by using the physical_table_naming_convention option.

These options are in the environments section of the configuration reference page.

Physical table schemas

By default, SQLMesh creates physical schemas for a model with a naming convention of sqlmesh__[model schema].

This can be overridden on a per-schema basis using the physical_schema_mapping option, which removes the sqlmesh__ prefix and uses the regex pattern you provide to map the schemas defined in your model to their corresponding physical schemas.

This example configuration overrides the default physical schemas for the my_schema model schema and any model schemas starting with dev:

=== "YAML"

```yaml linenums="1"
physical_schema_mapping:
  '^my_schema$': my_new_schema,
  '^dev.*': development
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    physical_schema_mapping={
        "^my_schema$": "my_new_schema",
        '^dev.*': "development"
    },
)
```

This config causes the following mapping behaviour:

Model name Default physical location Resolved physical location
my_schema.my_table sqlmesh__my_schema.table_<fingerprint> my_new_schema.table_<fingerprint>
dev_schema.my_table sqlmesh__dev_schema.table_<fingerprint> development.table_<fingerprint>
other.my_table sqlmesh__other.table_<fingerprint> sqlmesh__other.table_<fingerprint>

This only applies to the physical tables that SQLMesh creates - the views are still created in my_schema (prod) or my_schema__<env>.

Disable environment-specific schemas

SQLMesh stores prod environment views in the schema in a model's name - for example, the prod views for a model my_schema.users will be located in my_schema.

By default, for non-prod environments SQLMesh creates a new schema that appends the environment name to the model name's schema. For example, by default the view for a model my_schema.users in a SQLMesh environment named dev will be located in the schema my_schema__dev as my_schema__dev.users.

Show at the table level instead

This behavior can be changed to append a suffix at the end of a table/view name instead. Appending the suffix to a table/view name means that non-prod environment views will be created in the same schema as the prod environment. The prod and non-prod views are differentiated by non-prod view names ending with __<env>.

For example, if you created a dev environment for a project containing a model named my_schema.users, the model view would be created as my_schema.users__dev instead of the default behavior of my_schema__dev.users.

Config example:

=== "YAML"

```yaml linenums="1"
environment_suffix_target: table
```

=== "Python"

The Python `environment_suffix_target` argument takes an `EnvironmentSuffixTarget` enumeration with a value of `EnvironmentSuffixTarget.TABLE`, `EnvironmentSuffixTarget.CATALOG` or `EnvironmentSuffixTarget.SCHEMA` (default).

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig, EnvironmentSuffixTarget

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    environment_suffix_target=EnvironmentSuffixTarget.TABLE,
)
```

!!! info "Default behavior" The default behavior of appending the suffix to schemas is recommended because it leaves production with a single clean interface for accessing the views. However, if you are deploying SQLMesh in an environment with tight restrictions on schema creation then this can be a useful way of reducing the number of schemas SQLMesh uses.

Show at the catalog level instead

If neither the schema (default) nor the table level are sufficient for your use case, you can indicate the environment at the catalog level instead.

This can be useful if you have downstream BI reporting tools and you would like to point them at a development environment to test something out without renaming all the table / schema references within the report query.

In order to achieve this, you can configure environment_suffix_target like so:

=== "YAML"

```yaml linenums="1"
environment_suffix_target: catalog
```

=== "Python"

The Python `environment_suffix_target` argument takes an `EnvironmentSuffixTarget` enumeration with a value of `EnvironmentSuffixTarget.TABLE`, `EnvironmentSuffixTarget.CATALOG` or `EnvironmentSuffixTarget.SCHEMA` (default).

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig, EnvironmentSuffixTarget

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    environment_suffix_target=EnvironmentSuffixTarget.CATALOG,
)
```

Given the example of a model called my_schema.users with a default catalog of warehouse this will cause the following behavior:

  • For the prod environment, the default catalog as configured in the gateway will be used. So the view will be created at warehouse.my_schema.users
  • For any other environment, eg dev, the environment name will be appended to the default catalog. So the view will be created at warehouse__dev.my_schema.users
  • If a model is fully qualified with a catalog already, eg finance_mart.my_schema.users, then the environment catalog will be based off the model catalog and not the default catalog. In this example, the view will be created at finance_mart__dev.my_schema.users

!!! warning "Caveats" - Using environment_suffix_target: catalog only works on engines that support querying across different catalogs. If your engine does not support cross-catalog queries then you will need to use environment_suffix_target: schema or environment_suffix_target: table instead. - Automatic catalog creation is not supported on all engines even if they support cross-catalog queries. For engines where it is not supported, the catalogs must be managed externally from SQLMesh and exist prior to invoking SQLMesh.

Physical table naming convention

Out of the box, SQLMesh has the following defaults set:

  • environment_suffix_target: schema
  • physical_table_naming_convention: schema_and_table
  • no physical_schema_mapping overrides, so a sqlmesh__<model schema> physical schema will be created for each model schema

This means that given a catalog of warehouse and a model named finance_mart.transaction_events_over_threshold, SQLMesh will create physical tables using the following convention:

# <catalog>.sqlmesh__<schema>.<schema>__<table>__<fingerprint>

warehouse.sqlmesh__finance_mart.finance_mart__transaction_events_over_threshold__<fingerprint>

This deliberately contains some redundancy with the model schema as it's repeated at the physical layer in both the physical schema name as well as the physical table name.

This default exists to make the physical table names portable between different configurations. If you were to define a physical_schema_mapping that maps all models to the same physical schema, since the model schema is included in the table name as well, there are no naming conflicts.

Table only

Some engines have object name length limitations which cause them to silently truncate table and view names that exceed this limit. This behaviour breaks SQLMesh, so we raise a runtime error if we detect the engine would silently truncate the name of the table we are trying to create.

Having redundancy in the physical table names does reduce the number of characters that can be utilised in model names. To increase the number of characters available to model names, you can use physical_table_naming_convention like so:

=== "YAML"

```yaml linenums="1"
physical_table_naming_convention: table_only
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig, TableNamingConvention

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    physical_table_naming_convention=TableNamingConvention.TABLE_ONLY,
)
```

This will cause SQLMesh to omit the model schema from the table name and generate physical names that look like (using the above example):

# <catalog>.sqlmesh__<schema>.<table>__<fingerprint>

warehouse.sqlmesh__finance_mart.transaction_events_over_threshold__<fingerprint>

Notice that the model schema name is no longer part of the physical table name. This allows for slightly longer model names on engines with low identifier length limits, which may be useful for your project.

In this configuration, it is your responsibility to ensure that any schema overrides in physical_schema_mapping result in each model schema getting mapped to a unique physical schema.

For example, the following configuration will cause data corruption:

physical_table_naming_convention: table_only
physical_schema_mapping:
  '.*': sqlmesh

This is because every model schema is mapped to the same physical schema but the model schema name is omitted from the physical table name.

MD5 hash

If you still need more characters, you can set physical_table_naming_convention: hash_md5 like so:

=== "YAML"

```yaml linenums="1"
physical_table_naming_convention: hash_md5
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig, TableNamingConvention

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    physical_table_naming_convention=TableNamingConvention.HASH_MD5,
)
```

This will cause SQLMesh generate physical names that are always 45-50 characters in length and look something like:

# sqlmesh_md5__<hash of what we would have generated using 'schema_and_table'>

sqlmesh_md5__d3b07384d113edec49eaa6238ad5ff00

# or, for a dev preview
sqlmesh_md5__d3b07384d113edec49eaa6238ad5ff00__dev

This has a downside that now it's much more difficult to determine which table corresponds to which model by just looking at the database with a SQL client. However, the table names have a predictable length so there are no longer any surprises with identfiers exceeding the max length at the physical layer.

Virtual Data Environment Modes

By default, Virtual Data Environments (VDE) are applied across both development and production environments. This allows SQLMesh to reuse physical tables when appropriate, even when promoting from development to production.

However, users may prefer their production environment to be non-virtual. The non-exhaustive list of reasons may include:

  • Integration with third-party tools and platforms, such as data catalogs, may not work well with the virtual view layer that SQLMesh imposes by default
  • A desire to rely on time travel features provided by cloud data warehouses such as BigQuery, Snowflake, and Databricks

To mitigate this, SQLMesh offers an alternative 'dev-only' mode for using VDE. It can be enabled in the project configuration like so:

=== "YAML"

```yaml linenums="1"
virtual_environment_mode: dev_only
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config

config = Config(
    virtual_environment_mode="dev_only",
)
```

'dev-only' mode means that VDE is applied only in development environments. While in production, model tables and views are updated directly and bypass the virtual layer. This also means that physical tables in production will be created using the original, unversioned model names. Users will still benefit from VDE and data reuse across development environments.

Please note the following tradeoffs when enabling this mode:

  • All data inserted in development environments is used only for preview and will not be reused in production
  • Reverting a model to a previous version will be applied going forward and may require an explicit data restatement

!!! warning Switching the mode for an existing project will result in a complete rebuild of all models in the project. Refer to the Table Migration Guide to migrate existing tables without rebuilding them from scratch.

Environment view catalogs

By default, SQLMesh creates an environment view in the same catalog as the physical table the view points to. The physical table's catalog is determined by either the catalog specified in the model name or the default catalog defined in the connection.

It can be desirable to create prod and non-prod virtual layer objects in separate catalogs instead. For example, there might be a "prod" catalog that contains all prod environment views and a separate "dev" catalog that contains all dev environment views.

Separate prod and non-prod catalogs can also be useful if you have a CI/CD pipeline that creates environments, like the SQLMesh Github Actions CI/CD Bot. You might want to store the CI/CD environment objects in a dedicated catalog since there can be many of them.

!!! info "Virtual layer only" Note that the following setting only affects the virtual layer. If you need full segregation by catalog between environments in the physical layer as well, see the Isolated Systems Guide.

To configure separate catalogs, provide a mapping from regex patterns to catalog names. SQLMesh will compare the name of an environment to the regex patterns; when it finds a match it will store the environment's objects in the corresponding catalog.

SQLMesh evaluates the regex patterns in the order defined in the configuration; it uses the catalog for the first matching pattern. If no match is found, the catalog defined in the model or the default catalog defined on the connection will be used.

Config example:

=== "YAML"

```yaml linenums="1"
environment_catalog_mapping:
  '^prod$': prod
  '^dev.*': dev
  '^analytics_repo.*': cicd
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    environment_catalog_mapping={
        '^prod$': 'prod',
        '^dev.*': 'dev',
        '^analytics_repo.*': 'cicd',
    },
)
```

With the example configuration above, SQLMesh would evaluate environment names as follows:

  • If the environment name is prod, the catalog will be prod.
  • If the environment name starts with dev, the catalog will be dev.
  • If the environment name starts with analytics_repo, the catalog will be cicd.

!!! warning This feature is mutually exclusive with environment_suffix_target: catalog in order to prevent ambiguous mappings from being defined. Attempting to specify both environment_catalog_mapping and environment_suffix_target: catalog will raise an error on project load

Note: This feature is only available for engines that support querying across catalogs. At the time of writing, the following engines are NOT supported:

Regex Tips
  • If you are less familiar with regex, you can use a tool like regex101 to help you build your regex patterns.
    • LLMs, like ChatGPT, can help with generating regex patterns. Make sure to validate the suggestion in regex101.
  • If you are wanting to do an exact word match then surround it with ^ and $ like in the example above.
  • If you want a catch-all at the end of your mapping, to avoid ever using the model catalog or default catalog, then use .* as the pattern. This will match any environment name that hasn't already been matched.

Auto-categorize model changes

SQLMesh compares the current state of project files to an environment when sqlmesh plan is run. It detects changes to models, which can be classified as breaking or non-breaking.

SQLMesh can attempt to automatically categorize the changes it detects. The plan.auto_categorize_changes option determines whether SQLMesh should attempt automatic change categorization. This option is in the plan section of the configuration reference page.

Supported values:

  • full: Never prompt the user for input, instead fall back to the most conservative category (breaking) if the category can't be determined automatically.
  • semi: Prompt the user for input only if the change category can't be determined automatically.
  • off: Always prompt the user for input; automatic categorization will not be attempted.

Example showing default values:

=== "YAML"

```yaml linenums="1"
plan:
  auto_categorize_changes:
    external: full
    python: off
    sql: full
    seed: full
```

=== "Python"

The Python `auto_categorize_changes` argument takes `CategorizerConfig` object. That object's arguments take an `AutoCategorizationMode` enumeration with values of `AutoCategorizationMode.FULL`, `AutoCategorizationMode.SEMI`, or `AutoCategorizationMode.OFF`.

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    AutoCategorizationMode,
    CategorizerConfig,
    PlanConfig,
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    plan=PlanConfig(
        auto_categorize_changes=CategorizerConfig(
            external=AutoCategorizationMode.FULL,
            python=AutoCategorizationMode.OFF,
            sql=AutoCategorizationMode.FULL,
            seed=AutoCategorizationMode.FULL,
        )
    ),
)
```

Always comparing against production

By default, SQLMesh compares the current state of project files to the target <env> environment when sqlmesh plan <env> is run. However, a common expectation is that local changes should always be compared to the production environment.

The always_recreate_environment boolean plan option can alter this behavior. When enabled, SQLMesh will always attempt to compare against the production environment by recreating the target environment; If prod does not exist, SQLMesh will fall back to comparing against the target environment.

NOTE:: Upon succesfull plan application, changes are still promoted to the target <env> environment.

=== "YAML"

```yaml linenums="1"
plan:
    always_recreate_environment: True
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    PlanConfig,
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    plan=PlanConfig(
        always_recreate_environment=True,
    ),
)
```

Change Categorization Example

Consider this scenario with always_recreate_environment enabled:

  1. Initial state in prod:
MODEL (name sqlmesh_example.test_model, kind FULL);
SELECT 1 AS col
  1. First (breaking) change in dev:
MODEL (name sqlmesh_example__dev.test_model, kind FULL);
SELECT 2 AS col

??? "Output plan example #1"

```bash
New environment `dev` will be created from `prod`

Differences from the `prod` environment:

Models:
└── Directly Modified:
    └── sqlmesh_example__dev.test_model

---
+++


kind FULL
)
SELECT
-  1 AS col
+  2 AS col
```
  1. Second (metadata) change in dev:
MODEL (name sqlmesh_example__dev.test_model, kind FULL, owner 'John Doe');
SELECT 5 AS col

??? "Output plan example #2"

```bash
New environment `dev` will be created from `prod`

Differences from the `prod` environment:

Models:
└── Directly Modified:
    └── sqlmesh_example__dev.test_model

---

+++

@@ -1,8 +1,9 @@

MODEL (
name sqlmesh_example.test_model,
+  owner "John Doe",
kind FULL
)
SELECT
-  1 AS col
+  2 AS col

Directly Modified: sqlmesh_example__dev.test_model (Breaking)
Models needing backfill:
└── sqlmesh_example__dev.test_model: [full refresh]
```

Even though the second change should have been a metadata change (thus not requiring a backfill), it will still be classified as a breaking change because the comparison is against production instead of the previous development state. This is intentional and may cause additional backfills as more changes are accumulated.

Gateways

The gateways configuration defines how SQLMesh should connect to the data warehouse, state backend, and scheduler. These options are in the gateway section of the configuration reference page.

Each gateway key represents a unique gateway name and configures its connections. Gateway names are case-insensitive - SQLMesh automatically normalizes gateway names to lowercase during configuration validation. This means you can use any case in your configuration files (e.g., MyGateway, mygateway, MYGATEWAY) and they will all work correctly.

For example, this configures the my_gateway gateway:

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    connection:
      ...
    state_connection:
      ...
    test_connection:
      ...
    scheduler:
      ...
```

=== "Python"

The Python `gateways` argument takes a dictionary of gateway names and `GatewayConfig` objects. A `GatewayConfig`'s connection-related arguments take an [engine-specific connection config](#engine-connection-configuration) object, and the `scheduler` argument takes a [scheduler config](#scheduler) object.


```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    ...
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            connection=...,
            state_connection=...,
            test_connection=...,
            scheduler=...,
        ),
    }
)
```

Gateways do not need to specify all four components in the example above. The gateway defaults options control what happens if they are not all specified - find more information on gateway defaults below.

Connections

The connection configuration controls the data warehouse connection. These options are in the connection section of the configuration reference page.

The allowed keys include:

  • The optional concurrent_tasks key specifies the maximum number of concurrent tasks SQLMesh will run. Default value is 4 for engines that support concurrent tasks.
  • Most keys are specific to the connection engine type - see below. The default data warehouse connection type is an in-memory DuckDB database.

Example snowflake connection configuration:

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    connection:
      type: snowflake
      user: <username>
      password: <password>
      account: <account>
```

=== "Python"

A Snowflake connection is specified with a `SnowflakeConnectionConfig` object.

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    SnowflakeConnectionConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            connection=SnowflakeConnectionConfig(
                user=<username>,
                password=<password>,
                account=<account>,
            ),
        ),
    }
)
```

Engine connection configuration

These pages describe the connection configuration options for each execution engine.

State connection

Configuration for the state backend connection if different from the data warehouse connection.

The data warehouse connection is used to store SQLMesh state if the state_connection key is not specified.

Unlike data transformations, storing state information requires database transactions. Data warehouses aren’t optimized for executing transactions, and storing state information in them can slow down your project or produce corrupted data due to simultaneous writes to the same table. Therefore, production SQLMesh deployments should use a dedicated state connection.

!!! note Using the same connection for data warehouse and state is not recommended for production deployments of SQLMesh.

The easiest and most reliable way to manage your state connection is for Tobiko Cloud to do it for you. If you'd rather handle it yourself, we list recommended and unsupported state engines below.

Recommended state engines for production deployments:

Other state engines with fast and reliable database transactions (less tested than the recommended engines):

Unsupported state engines, even for development:

This example gateway configuration uses Snowflake for the data warehouse connection and Postgres for the state backend connection:

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    connection:
      # snowflake credentials here
      type: snowflake
      user: <username>
      password: <password>
      account: <account>
    state_connection:
      # postgres credentials here
      type: postgres
      host: <host>
      port: <port>
      user: <username>
      password: <password>
      database: <database>
```

=== "Python"

A Postgres connection is specified with a `PostgresConnectionConfig` object.

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    PostgresConnectionConfig,
    SnowflakeConnectionConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            # snowflake credentials here
            connection=SnowflakeConnectionConfig(
                user=<username>,
                password=<password>,
                account=<account>,
            ),
            # postgres credentials here
            state_connection=PostgresConnectionConfig(
                host=<host>,
                port=<port>,
                user=<username>,
                password=<password>,
                database=<database>,
            ),
        ),
    }
)
```

State schema name

By default, the schema name used to store state tables is sqlmesh. This can be changed by providing the state_schema config key in the gateway configuration.

Example configuration to store state information in a postgres database's custom_name schema:

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    state_connection:
      type: postgres
      host: <host>
      port: <port>
      user: <username>
      password: <password>
      database: <database>
    state_schema: custom_name
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    PostgresConnectionConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            state_connection=PostgresConnectionConfig(
                host=<host>,
                port=<port>,
                user=<username>,
                password=<password>,
                database=<database>,
            ),
            state_schema="custom_name",
        ),
    }
)
```

This would create all state tables in the schema custom_name.

Test connection

Configuration for a connection used to run unit tests. An in-memory DuckDB database is used if the test_connection key is not specified.

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    test_connection:
      type: duckdb
```

=== "Python"

A DuckDB connection is specified with a `DuckDBConnectionConfig` object. A `DuckDBConnectionConfig` with no arguments specified uses an in-memory DuckDB database.

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    DuckDBConnectionConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            test_connection=DuckDBConnectionConfig(),
        ),
    }
)
```

Scheduler

Identifies which scheduler backend to use. The scheduler backend is used both for storing metadata and for executing plans. By default, the scheduler type is set to builtin, which uses the existing SQL engine to store metadata.

These options are in the scheduler section of the configuration reference page.

Builtin

Example configuration:

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    scheduler:
      type: builtin
```

=== "Python"

A built-in scheduler is specified with a `BuiltInSchedulerConfig` object.

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
    BuiltInSchedulerConfig,
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            scheduler=BuiltInSchedulerConfig(),
        ),
    }
)
```

No additional configuration options are supported by this scheduler type.

Gateway/connection defaults

The default gateway and connection keys specify what should happen when gateways or connections are not explicitly specified. These options are in the gateway/connection defaults section of the configuration reference page.

The gateway specified in default_gateway is used when a sqlmesh command does not explicitly specify a gateway. All SQLMesh CLI commands accept a gateway option after sqlmesh and before the command name; for example, sqlmesh --gateway my_gateway plan. If the option is not specified in a command call, the default_gateway is used.

The three default connection types are used when some gateways in the gateways configuration dictionaries do not specify every connection type.

Default gateway

If a configuration contains multiple gateways, SQLMesh will use the first one in the gateways dictionary by default. The default_gateway key is used to specify a different gateway name as the SQLMesh default.

Example configuration:

=== "YAML"

```yaml linenums="1"
gateways:
  my_gateway:
    <gateway specification>
default_gateway: my_gateway
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    GatewayConfig,
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    gateways={
        "my_gateway": GatewayConfig(
            <gateway specification>
        ),
    },
    default_gateway="my_gateway",
)
```

Default connections/scheduler

The default_connection, default_test_connection, and default_scheduler keys are used to specify shared defaults across multiple gateways.

For example, you might have a specific connection where your tests should run regardless of which gateway is being used. Instead of duplicating the test connection information in each gateway specification, specify it once in the default_test_connection key.

Example configuration specifying a Postgres default connection, in-memory DuckDB default test connection, and builtin default scheduler:

=== "YAML"

```yaml linenums="1"
default_connection:
  type: postgres
  host: <host>
  port: <port>
  user: <username>
  password: <password>
  database: <database>
default_test_connection:
  type: duckdb
default_scheduler:
  type: builtin
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import (
    Config,
    ModelDefaultsConfig,
    PostgresConnectionConfig,
    DuckDBConnectionConfig,
    BuiltInSchedulerConfig
)

config = Config(
    model_defaults=ModelDefaultsConfig(dialect=<dialect>),
    default_connection=PostgresConnectionConfig(
        host=<host>,
        port=<port>,
        user=<username>,
        password=<password>,
        database=<database>,
    ),
    default_test_connection=DuckDBConnectionConfig(),
    default_scheduler=BuiltInSchedulerConfig(),
)
```

Models

Model defaults

The model_defaults key is required and must contain a value for the dialect key. All SQL dialects supported by the SQLGlot library are allowed. Other values are set automatically unless explicitly overridden in the model definition.

All supported model_defaults keys are listed in the models configuration reference page.

Example configuration:

=== "YAML"

```yaml linenums="1"
model_defaults:
  dialect: snowflake
  owner: jen
  start: 2022-01-01
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, ModelDefaultsConfig

config = Config(
    model_defaults=ModelDefaultsConfig(
        dialect="snowflake",
        owner="jen",
        start="2022-01-01",
    ),
)
```

The default model kind is VIEW unless overridden with the kind key. For more information on model kinds, refer to model concepts page.

Identifier resolution

When a SQL engine receives a query such as SELECT id FROM "some_table", it eventually needs to understand what database objects the identifiers id and "some_table" correspond to. This process is usually referred to as identifier (or name) resolution.

Different SQL dialects implement different rules when resolving identifiers in queries. For example, certain identifiers may be treated as case-sensitive (e.g. if they're quoted), and a case-insensitive identifier is usually either lowercased or uppercased, before the engine actually looks up what object it corresponds to.

SQLMesh analyzes model queries so that it can extract useful information from them, such as computing Column-Level Lineage. To facilitate this analysis, it normalizes and quotes all identifiers in those queries, respecting each dialect's resolution rules.

The "normalization strategy", i.e. whether case-insensitive identifiers are lowercased or uppercased, is configurable per dialect. For example, to treat all identifiers as case-sensitive in a BigQuery project, one can do:

=== "YAML"

```yaml linenums="1"
model_defaults:
  dialect: "bigquery,normalization_strategy=case_sensitive"
```

This may be useful in cases where the name casing needs to be preserved, since then SQLMesh won't be able to normalize them.

See here to learn more about the supported normalization strategies.

Gateway-specific model defaults

You can also define gateway specific model_defaults in the gateways section, which override the global defaults for that gateway.

gateways:
  redshift:
    connection:
      type: redshift
    model_defaults:
      dialect: "snowflake,normalization_strategy=case_insensitive"
  snowflake:
    connection:
      type: snowflake

default_gateway: snowflake

model_defaults:
  dialect: snowflake
  start: 2025-02-05

This allows you to tailor the behavior of models for each gateway without affecting the global model_defaults.

For example, in some SQL engines identifiers like table and column names are case-sensitive, but they are case-insensitive in other engines. By default, a project that uses both types of engines would need to ensure the models for each engine aligned with the engine's normalization behavior, which makes project maintenance and debugging more challenging.

Gateway-specific model_defaults allow you to change how SQLMesh performs identifier normalization by engine to align the different engines' behavior.

In the example above, the project's default dialect is snowflake (line 14). The redshift gateway configuration overrides that global default dialect with "snowflake,normalization_strategy=case_insensitive" (line 6).

That value tells SQLMesh that the redshift gateway's models will be written in the Snowflake SQL dialect (so need to be transpiled from Snowflake to Redshift), but that the resulting Redshift SQL should treat identifiers as case-insensitive to match Snowflake's behavior.

Model Kinds

Model kinds are required in each model file's MODEL DDL statement. They may optionally be used to specify a default kind in the model defaults configuration key.

All model kind specification keys are listed in the models configuration reference page.

The VIEW, FULL, and EMBEDDED model kinds are specified by name only, while other models kinds require additional parameters and are provided with an array of parameters:

=== "YAML"

`FULL` model only requires a name:

```sql linenums="1"
MODEL(
  name docs_example.full_model,
  kind FULL
);
```

`INCREMENTAL_BY_TIME_RANGE` requires an array specifying the model's `time_column` (which should be in the UTC time zone):

```sql linenums="1"
MODEL(
  name docs_example.incremental_model,
  kind INCREMENTAL_BY_TIME_RANGE (
    time_column model_time_column
  )
);
```

Python model kinds are specified with model kind objects. Python model kind objects have the same arguments as their SQL counterparts, listed in the models configuration reference page.

This example demonstrates how to specify an incremental by time range model kind in Python:

=== "Python"

```python linenums="1"
from sqlmesh import ExecutionContext, model
from sqlmesh.core.model.kind import ModelKindName

@model(
    "docs_example.incremental_model",
    kind=dict(
        name=ModelKindName.INCREMENTAL_BY_TIME_RANGE,
        time_column="ds"
    )
)
```

Learn more about specifying Python models at the Python models concepts page.

Model Naming

The model_naming configuration controls if model names are inferred based on the project's directory structure. If model_naming is not defined or infer_names is set to false, the model names must be provided explicitly.

With infer_names set to true, model names are inferred based on their path. For example, a model located at models/catalog/schema/model.sql would be named catalog.schema.model. However, if a name is provided in the model definition, it will take precedence over the inferred name.

Example enabling name inference:

=== "YAML"

```yaml linenums="1"
model_naming:
  infer_names: true
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config, NameInferenceConfig

config = Config(
    model_naming=NameInferenceConfig(
        infer_names=True
    )
)
```

Before_all and after_all Statements

The before_all and after_all statements are executed at the start and end, respectively, of the sqlmesh plan and sqlmesh run commands.

These statements can be defined in the configuration file under the before_all and after_all keys, either as a list of SQL statements or by using SQLMesh macros:

=== "YAML"

```yaml linenums="1"
before_all:
  - CREATE TABLE IF NOT EXISTS analytics (table VARCHAR, eval_time VARCHAR)
after_all:
  - "@grant_select_privileges()"
  - "@IF(@this_env = 'prod', @grant_schema_usage())"
```

=== "Python"

```python linenums="1"
from sqlmesh.core.config import Config

config = Config(
    before_all = [
        "CREATE TABLE IF NOT EXISTS analytics (table VARCHAR, eval_time VARCHAR)"
    ],
    after_all = [
        "@grant_select_privileges()",
        "@IF(@this_env = 'prod', @grant_schema_usage())"
    ],
)
```

Examples

These statements allow for actions to be executed before all individual model statements or after all have run, respectively. They can also simplify tasks such as granting privileges.

Example: Granting Select Privileges

For example, rather than using an on_virtual_update statement in each model to grant privileges on the views of the virtual layer, a single macro can be defined and used at the end of the plan:

from sqlmesh.core.macros import macro

@macro()
def grant_select_privileges(evaluator):
    if evaluator.views:
        return [
            f"GRANT SELECT ON VIEW {view_name} /* sqlglot.meta replace=false */ TO ROLE admin_role;"
            for view_name in evaluator.views
        ]

By including the comment /* sqlglot.meta replace=false */, you further ensure that the evaluator does not replace the view name with the physical table name during rendering.

Example: Granting Schema Privileges

Similarly, you can define a macro to grant schema usage privileges and, as demonstrated in the configuration above, using this_env macro conditionally execute it only in the production environment.

from sqlmesh import macro

@macro()
def grant_schema_usage(evaluator):
    if evaluator.this_env == "prod" and evaluator.schemas:
        return [
            f"GRANT USAGE ON SCHEMA {schema} TO admin_role;"
            for schema in evaluator.schemas
        ]

As demonstrated in these examples, the schemas and views are available within the macro evaluator for macros invoked within the before_all and after_all statements. Additionally, the macro this_env provides access to the current environment name, which can be helpful for more advanced use cases that require fine-grained control over their behaviour.

Linting

SQLMesh provides a linter that checks for potential issues in your models' code. Enable it and specify which linting rules to apply in the configuration file's linter key.

Learn more about linting configuration in the linting guide.

Debug mode

To enable debug mode set the SQLMESH_DEBUG environment variable to one of the following values: "1", "true", "t", "yes" or "y".

Enabling this mode ensures that full backtraces are printed when using CLI. The default log level is set to DEBUG when this mode is enabled.

Example enabling debug mode for the CLI command sqlmesh plan:

=== "Bash"

```bash
$ SQLMESH_DEBUG=1 sqlmesh plan
```

=== "MS Powershell"

```powershell
PS> $env:SQLMESH_DEBUG=1
PS> sqlmesh plan
```

=== "MS CMD"

```cmd
C:\> set SQLMESH_DEBUG=1
C:\> sqlmesh plan
```

Python library dependencies

SQLMesh enables you to write Python models and macros which depend on third-party libraries. To ensure each run / evaluation uses the same version, you can specify versions in a sqlmesh-requirements.lock file in the root of your project.

The sqlmesh.lock must be of the format dep==version. Only == is supported.

For example:

numpy==2.1.2
pandas==2.2.3

This feature is only available in Tobiko Cloud.

Excluding dependencies

You can exclude dependencies by prefixing the dependency with a ^. For example:

^numpy
pandas==2.2.3