Skip to content

CLI

In this quick start guide, you'll use the SQLMesh command line interface (CLI) to get up and running with SQLMesh's scaffold generator. This example project will run locally on your computer using DuckDB as an embedded SQL engine.

Before beginning, ensure that you meet all the prerequisites for using SQLMesh.

Learn more about the quickstart project structure

This project demonstrates key SQLMesh features by walking through the SQLMesh workflow on a simple data pipeline. This section describes the project structure and the SQLMesh concepts you will encounter as you work through it.

The project contains three models with a CSV file as the only data source:

┌─────────────┐
│seed_data.csv│
└────────────┬┘
             │
            ┌▼─────────────┐
            │seed_model.sql│
            └─────────────┬┘
                          │
                         ┌▼────────────────────┐
                         │incremental_model.sql│
                         └────────────────────┬┘
                                              │
                                             ┌▼─────────────┐
                                             │full_model.sql│
                                             └──────────────┘

Although the project is simple, it touches on all the primary concepts needed to use SQLMesh productively.

1. Create the SQLMesh project

First, create a project directory and navigate to it:

mkdir sqlmesh-example

cd sqlmesh-example

If using a python virtual environment, ensure it's activated first by running the source .env/bin/activate command from the folder used during installation.

Create a SQLMesh scaffold with the following command, specifying a default SQL dialect for your models. The dialect should correspond to the dialect most of your models are written in; it can be overridden for specific models in the model's MODEL specification. All SQL dialects supported by the SQLGlot library are allowed.

In this example, we specify the duckdb dialect:

sqlmesh init duckdb

The scaffold will include a SQLMesh configuration file for the example project.

Learn more about the project's configuration

SQLMesh project-level configuration parameters are specified in the config.yaml file in the project directory.

This example project uses the embedded DuckDB SQL engine, so its configuration specifies duckdb as the local gateway's connection and the local gateway as the default.

The command to run the scaffold generator requires a default SQL dialect for your models, which it places in the config model_defaults dialect key. In this example, we specified the duckdb SQL dialect as the default:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
gateways:
  local:
    connection:
      type: duckdb
      database: ./db.db

default_gateway: local

model_defaults:
  dialect: duckdb

Learn more about SQLMesh project configuration here.

The scaffold will also include multiple directories where SQLMesh project files are stored and multiple files that constitute the example project (e.g., SQL models).

Learn more about the project directories and files

SQLMesh uses a scaffold generator to initiate a new project. The generator will create multiple sub-directories and files for organizing your SQLMesh project code.

The scaffold generator will create the following configuration file and directories:

  • config.yaml
    • The file for project configuration. More info about configuration here.
  • ./models
    • SQL and Python models. More info about models here.
  • ./seeds
    • Seed files. More info about seeds here.
  • ./audits
    • Shared audit files. More info about audits here.
  • ./tests
    • Unit test files. More info about tests here.
  • ./macros
    • Macro files. More info about macros here.

It will also create the files needed for this quickstart example:

  • ./models
    • full_model.sql
    • incremental_model.sql
    • seed_model.sql
  • ./seeds
    • seed_data.csv
  • ./audits
    • assert_positive_order_ids.sql
  • ./tests
    • test_full_model.yaml

Finally, the scaffold will include data for the example project to use.

Learn more about the project's data

The data used in this example project is contained in the seed_data.csv file in the /seeds project directory. The data reflects sales of 3 items over 7 days in January 2020.

The file contains three columns, id, item_id, and event_date, which correspond to each row's unique ID, the sold item's ID number, and the date the item was sold, respectively.

This is the complete dataset:

id item_id event_date
1 2 2020-01-01
2 1 2020-01-01
3 3 2020-01-03
4 1 2020-01-04
5 1 2020-01-05
6 1 2020-01-06
7 1 2020-01-07

2. Create a prod environment

SQLMesh's key actions are creating and applying plans to environments. At this point, the only environment is the empty prod environment.

Learn more about SQLMesh plans and environments

SQLMesh's key actions are creating and applying plans to environments.

A SQLMesh environment is an isolated namespace containing models and the data they generated. The most important environment is prod ("production"), which consists of the databases behind the applications your business uses to operate each day. Environments other than prod provide a place where you can test and preview changes to model code before they go live and affect business operations.

A SQLMesh plan contains a comparison of one environment to another and the set of changes needed to bring them into alignment. For example, if a new SQL model was added, tested, and run in the dev environment, it would need to be added and run in the prod environment to bring them into alignment. SQLMesh identifies all such changes and classifies them as either breaking or non-breaking.

Breaking changes are those that invalidate data already existing in an environment. For example, if a WHERE clause was added to a model in the dev environment, existing data created by that model in the prod environment are now invalid because they may contain rows that would be filtered out by the new WHERE clause. Other changes, like adding a new column to a model in dev, are non-breaking because all the existing data in prod are still valid to use - only new data must be added to align the environments.

After SQLMesh creates a plan, it summarizes the breaking and non-breaking changes so you can understand what will happen if you apply the plan. It will prompt you to "backfill" data to apply the plan - in this context, backfill is a generic term for updating or adding to a table's data (including an initial load or full refresh).

The first SQLMesh plan must execute every model to populate the production environment. Running sqlmesh plan will generate the plan and the following output:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
$ sqlmesh plan
======================================================================
Successfully Ran 1 tests against duckdb
----------------------------------------------------------------------
New environment `prod` will be created from `prod`
Summary of differences against `prod`:
└── Added Models:
    ├── sqlmesh_example.seed_model
    ├── sqlmesh_example.incremental_model
    └── sqlmesh_example.full_model
Models needing backfill (missing dates):
├── sqlmesh_example.full_model: 2020-01-01 - 2023-05-31
├── sqlmesh_example.incremental_model: 2020-01-01 - 2023-05-31
└── sqlmesh_example.seed_model: 2023-05-31 - 2023-05-31
Apply - Backfill Tables [y/n]:

Line 3 of the output notes that sqlmesh plan successfully executed the project's test tests/test_full_model.yaml with duckdb.

Line 5 describes what environments the plan will affect when applied - a new prod environment in this case.

Lines 7-10 of the output show that SQLMesh detected three new models relative to the current empty environment.

Lines 11-14 list each model that will be executed by the plan, along with the date intervals that will be run. Note that full_model and incremental_model both show 2020-01-01 as their start date because:

  1. The incremental model specifies that date in the start property of its MODEL statement and
  2. The full model depends on the incremental model.

The seed_model date range begins on the same day the plan was made because SEED models have no temporality associated with them other than whether they have been modified since the previous SQLMesh plan.

Learn more about the project's models

A plan's actions are determined by the kinds of models the project uses. This example project uses three model kinds:

  1. SEED models read data from CSV files stored in the SQLMesh project directory.
  2. FULL models fully refresh (rewrite) the data associated with the model every time the model is run.
  3. INCREMENTAL_BY_TIME_RANGE models use a date/time data column to track which time intervals are affected by a plan and process only the affected intervals when a model is run.

We now briefly review each model in the project.

The first model is a SEED model that imports seed_data.csv. This model consists of only a MODEL statement because SEED models do not query a database.

In addition to specifying the model name and CSV path relative to the model file, it includes the column names and data types of the columns in the CSV. It also sets the grain of the model to the columns that collectively form the model's unique identifier, id and event_date.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
MODEL (
  name sqlmesh_example.seed_model,
  kind SEED (
    path '../seeds/seed_data.csv'
  ),
  columns (
    id INTEGER,
    item_id INTEGER,
    event_date DATE
  ),
  grain (id, event_date)
);

The second model is an INCREMENTAL_BY_TIME_RANGE model that includes both a MODEL statement and a SQL query selecting from the first seed model.

The MODEL statement's kind property includes the required specification of the data column containing each record's timestamp. It also includes the optional start property specifying the earliest date/time for which the model should process data and the cron property specifying that the model should run daily. It sets the model's grain to columns id and event_date.

The SQL query includes a WHERE clause that SQLMesh uses to filter the data to a specific date/time interval when loading data incrementally:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
MODEL (
  name sqlmesh_example.incremental_model,
  kind INCREMENTAL_BY_TIME_RANGE (
    time_column event_date
  ),
  start '2020-01-01',
  cron '@daily',
  grain (id, event_date)
);

SELECT
  id,
  item_id,
  event_date,
FROM
  sqlmesh_example.seed_model
WHERE
  event_date between @start_date and @end_date

The final model in the project is a FULL model. In addition to properties used in the other models, its MODEL statement includes the audits property. The project includes a custom assert_positive_order_ids audit in the project audits directory; it verifies that all item_id values are positive numbers. It will be run every time the model is executed.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
MODEL (
  name sqlmesh_example.full_model,
  kind FULL,
  cron '@daily',
  grain item_id,
  audits (assert_positive_order_ids),
);

SELECT
  item_id,
  count(distinct id) AS num_orders,
FROM
  sqlmesh_example.incremental_model
GROUP BY item_id

Line 15 asks you whether to proceed with executing the model backfills described in lines 11-14. Enter y and press Enter, and SQLMesh will execute the models and return this output:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
Apply - Backfill Tables [y/n]: y
Creating physical tables ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 3/3 • 0:00:00

All model versions have been created successfully

[1/1] sqlmesh_example.seed_model evaluated in 0.01s
[1/1] sqlmesh_example.incremental_model evaluated in 0.01s
[1/1] sqlmesh_example.full_model evaluated in 0.02s
Evaluating models ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 3/3 • 0:00:00

All model batches have been executed successfully

Virtually Updating 'prod' ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 0:00:00

The target environment has been updated successfully

SQLMesh performs three actions when applying the plan:

  • Creating and storing new versions of the models
  • Evaluating/running the models
  • Virtually updating the plan's target environment

Line 2 provides a progress bar and elapsed time for the first step of creating new model versions (very fast in this simple project). Line 4 reports that the first step has completed.

Lines 6-8 show the run time for each model in the project. Line 9 provides a progress bar and total elapsed time for the second step of evaluating the models. Line 11 reports that the second step has completed.

Line 13 provides a progress bar and total elapsed time for the third step of virtually updating the plan's target environment. Line 15 reports that the third step has completed and the prod environment now points to the tables created during model execution.

You've now created a new production environment with all of history backfilled.

3. Update a model

Now that we have populated the prod environment, let's modify one of the SQL models.

We modify the incremental SQL model by adding a new column to the query. Open the models/incremental_model.sql file and add 'z' AS new_column below item_id as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
MODEL (
  name sqlmesh_example.incremental_model,
  kind INCREMENTAL_BY_TIME_RANGE (
    time_column event_date
  ),
  start '2020-01-01',
  cron '@daily',
  grain (id, event_date)
);

SELECT
  id,
  item_id,
  'z' AS new_column, -- Added column
  event_date,
FROM
  sqlmesh_example.seed_model
WHERE
  event_date between @start_date and @end_date

4. Work with a development environment

4.1 Create a dev environment

Now that you've modified a model, it's time to create a development environment so that you can validate the model change without affecting production.

Run sqlmesh plan dev to create a development environment called dev:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
$ sqlmesh plan dev
======================================================================
Successfully Ran 1 tests against duckdb
----------------------------------------------------------------------
New environment `dev` will be created from `prod`
Summary of differences against `dev`:
Models:
├── Directly Modified:
│   └── sqlmesh_example__dev.incremental_model
└── Indirectly Modified:
    └── sqlmesh_example__dev.full_model
---

+++

@@ -10,6 +10,7 @@

 SELECT
   id,
   item_id,
+  'z' AS new_column,
   event_date
 FROM sqlmesh_example.seed_model
 WHERE
Directly Modified: sqlmesh_example__dev.incremental_model (Non-breaking)
└── Indirectly Modified Children:
    └── sqlmesh_example__dev.full_model (Indirect Non-breaking)
Models needing backfill (missing dates):
└── sqlmesh_example__dev.incremental_model: 2020-01-01 - 2023-05-31
Enter the backfill start date (eg. '1 year', '2020-01-01') or blank to backfill from the beginning of history:

Line 5 of the output states that a new environment dev will be created from the existing prod environment.

Lines 6-11 summarize the differences between the modified model and the prod environment, detecting that we directly modified incremental_model and that full_model was indirectly modified because it selects from the incremental model. Note that the model schemas are sqlmesh_example__dev, indicating that they are being created in the dev environment.

On line 25, we see that SQLMesh automatically classified the change as Non-breaking because it understood that the change was additive (added a column not used by full_model) and did not invalidate any data already in prod.

Hit Enter at the prompt to backfill data from our start date 2020-01-01. Another prompt will appear asking for a backfill end date; hit Enter to backfill until now. Finally, enter y and press Enter to apply the plan and execute the backfill:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
Enter the backfill start date (eg. '1 year', '2020-01-01') or blank to backfill from the beginning of history:
Enter the backfill end date (eg. '1 month ago', '2020-01-01') or blank to backfill up until now:
Apply - Backfill Tables [y/n]: y
Creating physical tables ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 2/2 • 0:00:00

All model versions have been created successfully

[1/1] sqlmesh_example__dev.incremental_model evaluated in 0.01s
Evaluating models ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 1/1 • 0:00:00


All model batches have been executed successfully

Virtually Updating 'dev' ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 0:00:00

The target environment has been updated successfully

Line 8 of the output shows that SQLMesh applied the change and evaluated sqlmesh_example__dev.incremental_model.

SQLMesh did not need to backfill anything for the full_model since the change was Non-breaking.

4.2 Validate updates in dev

You can now view this change by querying data from incremental_model with sqlmesh fetchdf "select * from sqlmesh_example__dev.incremental_model".

Note that the environment name __dev is appended to the schema namespace sqlmesh_example in the query:

$ sqlmesh fetchdf "select * from sqlmesh_example__dev.incremental_model"

   id  item_id new_column  event_date
0   1        2          z  2020-01-01
1   2        1          z  2020-01-01
2   3        3          z  2020-01-03
3   4        1          z  2020-01-04
4   5        1          z  2020-01-05
5   6        1          z  2020-01-06
6   7        1          z  2020-01-07

You can see that new_column was added to the dataset. The production table was not modified; you can validate this by querying the production table using sqlmesh fetchdf "select * from sqlmesh_example.incremental_model".

Note that nothing has been appended to the schema namespace sqlmesh_example in this query because prod is the default environment.

$ sqlmesh fetchdf "select * from sqlmesh_example.incremental_model"

   id  item_id   event_date
0   1        2   2020-01-01
1   2        1   2020-01-01
2   3        3   2020-01-03
3   4        1   2020-01-04
4   5        1   2020-01-05
5   6        1   2020-01-06
6   7        1   2020-01-07

The production table does not have new_column because the changes to dev have not yet been applied to prod.

5. Update the prod environment

5.1 Apply updates to prod

Now that we've tested the changes in dev, it's time to move them to production. Run sqlmesh plan to plan and apply your changes to the prod environment.

Enter y and press Enter at the Apply - Virtual Update [y/n]: prompt to apply the plan and execute the backfill:

$ sqlmesh plan
======================================================================
Successfully Ran 1 tests against duckdb
----------------------------------------------------------------------
Summary of differences against `prod`:
Models:
├── Directly Modified:
│   └── sqlmesh_example.incremental_model
└── Indirectly Modified:
    └── sqlmesh_example.full_model
---

+++

@@ -10,6 +10,7 @@

 SELECT
   id,
   item_id,
+  'z' AS new_column,
   event_date
 FROM sqlmesh_example.seed_model
 WHERE
Directly Modified: sqlmesh_example.incremental_model (Non-breaking)
└── Indirectly Modified Children:
    └── sqlmesh_example.full_model (Indirect Non-breaking)
Apply - Virtual Update [y/n]: y
Virtually Updating 'prod' ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 0:00:00

The target environment has been updated successfully

Virtual Update executed successfully

Note that a backfill was not necessary and only a Virtual Update occurred.

5.2 Validate updates in prod

Double-check that the data updated in prod by running sqlmesh fetchdf "select * from sqlmesh_example.incremental_model":

$ sqlmesh fetchdf "select * from sqlmesh_example.incremental_model"

   id  item_id new_column  event_date
0   1        2          z  2020-01-01
1   2        1          z  2020-01-01
2   3        3          z  2020-01-03
3   4        1          z  2020-01-04
4   5        1          z  2020-01-05
5   6        1          z  2020-01-06
6   7        1          z  2020-01-07

6. Next steps

Congratulations, you've now conquered the basics of using SQLMesh!

From here, you can: