This is a sandbox project for exploring the basic functionality and latest features of dbt. It's based on a fictional restaurant called the Jaffle Shop that serves jaffles.
This README will guide you through setting up the project on dbt Cloud. Working through this example should give you a good sense of how dbt Cloud works and what's involved with setting up your own project. We'll also optionally cover some intermediate topics like setting up Environments and Jobs in dbt Cloud, working with a larger dataset, and setting up pre-commit hooks if you'd like.
Note
This project is geared towards folks learning dbt Cloud with a cloud warehouse. If you're brand new to dbt, we recommend starting with the dbt Learn platform. It's a free, interactive way to learn dbt, and it's a great way to get started if you're new to the tool. If you just want to try dbt locally as quickly as possible without setting up a data warehouse check out jaffle_shop_duckdb
.
Ready to go? Grab some water and a nice snack, and let's dig in!
- A dbt Cloud account
- A data warehouse (BigQuery, Snowflake, Redshift, Databricks, or Postgres) with adequate permissions to create a fresh database for this project and run dbt in it
- Optional Python 3.9 or higher (for generating synthetic data with
jafgen
)
-
Follow the steps to create a new repository. You can choose to only copy the
main
branch for simplicity, or take advantage of the Write-Audit-Publish (WAP) flow we use to maintain the project and copy all branches (which will includemain
andstaging
along with any active feature branches). Either option is fine!
Tip
In a setup that follows a WAP flow, you have a main
branch that serves production data (like downstream dashboards) and is tied to a Production Environment in dbt Cloud, and a staging
branch that serves a clone of that data and is tied to a Staging Environment in dbt Cloud. You then branch off of staging
to add new features or fix bugs, and merge back into staging
when you're done. When you're ready to deploy to production, you merge staging
into main
. Staging is meant to be more-or-less a mirror of production, but safe to test breaking changes, so you can verify changes in a production-like environment before deploying them fully. You write to staging
, audit in staging
, and publish to main
.
-
Create a logical database in your data warehouse for the Jaffle Shop project. We recommend using the name
jaffle_shop
for consistency with the project. This looks different on different platforms (for instance on BigQuery this constitutes creating a new project, on Snowflake this is achieved viacreate database jaffle_shop;
, and if you're running Postgres locally you can probably skip this). If you're not sure how to do this, we recommend checking out the Quickstart Guide for your data platform in the dbt Docs. -
Set up a dbt Cloud account (if you don't have one already, if you do, just create a new project) and follow Step 4 in the Quickstart Guide for your data platform, to connect your platform to dbt Cloud. Make sure the user you configure for your connections has adequate database permissions to run dbt in the
jaffle_shop
database. -
Choose the repo you created in Step 1 of the Create new repo from template section as the repository for your dbt Project's codebase.
The following should now be done:
- dbt Cloud connected to your warehouse
- Your copy of this repo set up as the codebase
- dbt Cloud and the codebase pointed at a fresh database or project in your warehouse to work in
You're now ready to start developing with dbt Cloud! Choose a path below (either the dbt Cloud IDE or the Cloud CLI to get started.
- Click
Develop
in the dbt Cloud nav bar. You should be prompted to run adbt deps
, which you should do. This will install the dbt packages configured in thepackages.yml
file.
-
Run
git clone [new repo git link]
(orgh repo clone [repo owner]/[new repo name]
if you prefer GitHub's excellent CLI) to clone your new repo from the first step of the Create new repo from template section to your local machine. -
Follow the steps on this page to install and set up a dbt Cloud connection with the dbt Cloud CLI.
There are a few ways to load the data for the project:
- Using the sample data in the repo. Add
"jaffle-data"
to theseed-paths
config in yourdbt_project.yml
as below. This means that when dbt is scanning folders forseeds
to load it will look in both theseeds
folder as is default, but also thejaffle-data
folder which contains a sample of the project data. Seeds are static data files in CSV format that dbt will upload, usually for reference models, like US zip codes mapped to country regions for example, but in this case the feature is hacked to do some data ingestion. This is not what seeds are meant to be used for (dbt is not a data loading tool), but it's useful for this project to give you some data to get going with quickly. Run adbt seed
and when it's done either delete thejaffle-data
folder, removejaffle-data
from theseed-paths
list, or ideally, both.
seed-paths: ["seeds", "jaffle-data"]
dbt seed
-
Load the data via S3. If you'd prefer a larger dataset (6 years instead of 1), and are working via the dbt Cloud IDE and your platform's web interface, you can also copy the data from a public S3 bucket to your warehouse into a schema called
raw
in yourjaffle_shop
database. This is discussed here. -
Generate a larger dataset on the command line. If you're working with the dbt Cloud CLI and comfortable with command line basics, you can generate as many years of data as you'd like (up to 10) to load into your warehouse. This is discussed here.
Once your development platform of choice and dependencies are set up, use the following steps to get the project ready for whatever you'd like to do with it.
-
Ensure that you've deleted the
jaffle-data
folder or removed it from theseed-paths
list in yourdbt_project.yml
(or, ideally, both) if you used the seed method to load the data. This is important, if you don't do this,dbt build
will re-run the seeds unnecessarily and things will get messy. -
Run a
dbt build
to build the project.
The following should now be done:
- Synthetic data loaded into your warehouse
- Development environment set up and ready to go
- The project built and tested
You're free to explore the Jaffle Shop from here, or if you want to learn more about setting up Environment and Jobs, generating a larger dataset, or setting up pre-commit hooks to standardize formatting and linting workflows, carry on!
Note
π Here be dragons! The following sections are for folks who are comfortable with the basics and want to explore more advanced topics. If you're just getting started, it's okay to skip these for now and come back later.
dbt Cloud has a powerful abstraction called an Environment. An Environment in dbt Cloud is a set of configurations that dbt uses when it runs your code. It includes things like what version of dbt to use, what schema to build into, credentials to use, and more. You can set up multiple environments in dbt Cloud, and each environment can have its own set of configurations. This is very useful for running Jobs. A Job is a set of dbt commands which run in an Environment. Understanding these two concepts is key for getting those most out of dbt Cloud, especially building a robust deployment workflow. Now that we're able to develop in our project, this section will walk you through setting up an Environment and a Job to deploy our project to production.
-
Go to the Deploy tab in the dbt Cloud nav bar and click
Environments
. -
On the Environment page, click
+ Create Environment
. -
Name your Environment
Prod
and set it as aProduction
Environment. -
Fill out the credentials with your warehouse connection details, in real production you'll want to make a Service Account or similar and only give access to the production schema to that user, so that only dbt Cloud Jobs can build into production. For this demo project, it's okay to just use your account credentials.
-
Set the
branch
that this Environment runs on tomain
, then the schema that this Environment builds into toprod
. This ensures that Jobs configured in this Environment always build into theprod
schema and run on themain
branch which we've protected as our production branch. -
Click
Save
.
Now we'll create a Job to deploy our project to production. This Job will run the dbt build
command in the prod
Environment we just created.
-
Go to the
Prod
Environment you just created. -
Click
+ Create Job
and chooseDeploy Job
as the Job type. -
Name your Job
Production Build
. -
You can otherwise leave the defaults in place and just click
Save
. -
Click into your newly created Job and click
Run Now
in the top right corner. -
This will kick off a Job to build your project in the
Prod
Environment, which will build into theprod
schema in your warehouse. -
Go check out the
prod
schema in yourjaffle_shop
database on your warehouse, you should see the project's models built there!
Tip
If you're working in the dbt Cloud IDE, make sure to turn on the 'Defer to staging/production' toggle once you've done this. This will ensure that only modified code is run when you run commands in the IDE, compared against the Production environment you just set up. This will save you significant time and resources!
Tip
The dbt Cloud CLI will automatically defer unmodified models to the previously built models in your staging or production environment, so you can run dbt build
, dbt test
, etc without worrying about running unnecessary code.
From here, you should be able to use dbt Explorer (in the Explore
tab of the dbt Cloud nav bar) to explore your DAG! Explorer is populated with metadata from your designated Production and Staging Environments, so you can see the lineage of your project visually, and much more.
There are two ways to work with a larger dataset than the default one year of data that comes with the project:
-
Load the data from S3 which will let you access the canonical 6 year dataset the project is tested against.
-
Generate via
jafgen
and seed the data with dbt Core which will allow you to generate up to 10 years of data.
To load the data from S3, consult the dbt Documentation's Quickstart Guides for your data platform to see how to copy data from an S3 bucket to your warehouse. The S3 bucket URIs of the tables you want to copy into your raw
schema are:
raw_customers
:s3://jaffle-shop-raw/raw_customers.csv
raw_orders
:s3://jaffle-shop-raw/raw_orders.csv
raw_order_items
:s3://jaffle-shop-raw/raw_order_items.csv
raw_products
:s3://jaffle-shop-raw/raw_products.csv
raw_supplies
:s3://jaffle-shop-raw/raw_supplies.csv
raw_stores
:s3://jaffle-shop-raw/raw_stores.csv
You'll need to be working on the command line for this option. If you're more comfortable working via web apps, the above method is the path you'll need. jafgen
is a simple tool for generating synthetic Jaffle Shop data that is maintained on a volunteer-basis by dbt Labs employees. This project is more interesting with a larger dataset generated and uploaded to your warehouse. 6 years is a nice amount to fully observe trends like growth, seasonality, and buyer personas that exist in the data. Uploading this amount of data requires a few extra steps, but we'll walk you through them. If you have a preferred way of loading CSVs into your warehouse or an S3 bucket, that will also work just fine, the generated data is just CSV files.
Tip
If you'd like to explore further on the command line, but are a little intimidated by the terminal, we've included configuration for a task runner called, fittingly, task
. It's a simple way to run the commands you need to get started with dbt. You can install it by following the instructions here. We'll call out the task
based alternative to each command below that provides an 'easy button'. It's a useful tool to have installed regardless.
-
Create a
profiles.yml
file in the root of your project. This file is already.gitignore
d so you can keep your credentials safe. If you'd prefer you can instead set up aprofiles.yml
file at the~/.dbt/profiles.yml
path instead to be extra sure you don't accidentally commit the file. -
Add a profile for your warehouse connection in this file and add this configuration to your
dbt_project.yml
file as a top-level key calledprofile
e.g.profile: my-profile-name
.
Important
If you do decide to use task
there is a super-task (task load
) that will do all of the below steps for you. Just run task load YEARS=[integer of years to generate] DB=[name of warehouse]
e.g. task YEARS=4 DB=bigquery
or task YEARS=7 DB=redshift
etc to perform all the commands necessary to generate and seed the data once your profiles.yml
file is set up.
- Create a new virtual environment in your project (I like to call mine
.venv
) and activate it, then install the project's dependencies in it. This will install thejafgen
tool which you can use to generate the larger datasets. Then installdbt-core
and your warehouse's adapter. We install dbt Core temporarily because by connecting directly to your warehouse, it can upload larger file sizes than the dbt Cloud server1. You can do this manually or withtask
:
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt
python3 -m pip install dbt-core dbt-[your warehouse adapter] # e.g. dbt-bigquery
OR
task venv
task install DB=[name of warehouse] # e.g. task install DB=bigquery
Note
Because you have an active virtual environment, this new install of dbt
should take precedence in your $PATH
. If you're not familiar with the PATH
environment variable, just think of this as the order in which your computer looks for commands to run. What's important is that it will look in your active virtual environment first, so when you run dbt
, it will use the dbt
you just installed in your virtual environment.
- Add
jaffle-data
to yourseed-paths
config in yourdbt-project.yml
as detailed here, then runjafgen
andseed
the data it generates.
jafgen [number of years to generate] # e.g. jafgen 6
dbt seed
OR
task gen YEARS=[integer of years to generate] # e.g. task gen YEARS=6
task seed
- Remove the
jaffle-data
folder, then uninstall the temporary dbt Core installation. Again, this was to allow you to seed the large data files, you don't need it for the rest of the project which will use the dbt Cloud CLI. You can then delete yourprofiles.yml
file and the configuration in yourdbt_project.yml
file. You should also delete thejaffle-data
path from theseed-paths
list in yourdbt_project.yml
.
rm -rf jaffle-data
python3 -m pip uninstall dbt-core dbt-[your warehouse adapter] # e.g. dbt-bigquery
OR
task clean
You now have a much more interesting and expansive dataset in your raw
schema to build with! You should now run a dbt build
to build the project with the new data into your dev schema or trigger your Production Build
Job in dbt Cloud to build the project in your prod
schema.
There's an optional tool included with the project called pre-commit
.
pre-commit automatically runs a suite of of processes on your code, like linters and formatters, when you commit. If it finds an issue and updates a file, you'll need to stage the changes and commit them again (the first commit will not have gone through because pre-commit found and fixed an issue). The outcome of this is that your code will be more consistent automatically, and everybody's changes will be running through the same set of processes. We recommend it for any project.
You can see the configuration for pre-commit in the .pre-commit-config.yaml
file. It's installed as part of the project's requirements.txt
, but you'll need to opt-in to using it by running pre-commit install
. This will install git hooks which run when you commit. You can also run the checks manually with pre-commit run --all-files
to see what it does without making a commit.
At present the following checks are run:
ruff
- an incredibly fast linter and formatter for Python, in case you add any Python modelscheck-yaml
- which validates YAML filesend-of-file-fixer
- which ensures all files end with a newlinetrailing-whitespace
- which trims trailing whitespace from files
At present, the popular SQL linter and formatter SQLFluff doesn't play nicely with the dbt Cloud CLI, so we've omitted it from this project for now. We've already built the backend for linting via the Cloud CLI, so this will change very soon! At present if you'd like auto-formatting and linting for SQL, check out the dbt Cloud IDE!
We have kept a .sqlfluff
config file to show what that looks like, and to future proof the repo for when the Cloud CLI support linting and formatting.
Footnotes
-
Again, I can't emphasize enough that you should not use dbt and seeds for data loading in a production project. This is just for convenience within this learning project. β©