dbt-ibis

With dbt-ibis you can write your dbt models using Ibis.

Supported dbt adapters:

• DuckDB
• Snowflake
• Postgres
• Redshift
• Trino
• MySQL
• SQLite
• Oracle

You can install dbt-ibis via pip:

pip install dbt-ibis

In addition, you'll need to install the relevant ibis backend for your database.

Basic example

You can write your Ibis model in files with the extension .ibis. Each .ibis file needs to correspond to one model which is defined as a model function returning an Ibis table expression:

stg_stores.ibis:

from dbt_ibis import depends_on, source


@depends_on(source("sources_db", "stores"))
def model(stores):
    return stores.mutate(store_id=stores["store_id"].cast("int"))

You can now reference the stg_stores model in either a normal SQL model using {{ ref('stg_stores') }} or in another Ibis model:

usa_stores.ibis:

from dbt_ibis import depends_on, ref


@depends_on(ref("stg_stores"))
def model(stores):
    return stores.filter(stores["country"] == "USA")

Whenever your Ibis model references either a source, a seed, a snapshot, or a SQL model, you'll need to define the column data types as described in Model Contracts - getdbt.com (data_type refers to the data types as they are called by your database system) (for sources, snapshots, and SQL models) or in Seed configurations - getdbt.com (for seeds). If you reference another Ibis model, this is not necessary. In the examples above, you would need to provide it for the stores source table:

sources:
  - name: sources_db
    ...
    tables:
      - name: stores
        columns:
          - name: store_id
            data_type: varchar
          - name: store_name
            data_type: varchar
          - name: country
            data_type: varchar

For more examples, including column data type definitions, see the demo project.

You can use all the dbt commands you're used to, you simply need to replace dbt with dbt-ibis. For example:

dbt-ibis run --select stg_stores+

You'll notice that for every .ibis file, dbt-ibis will generate a corresponding .sql file in a __ibis_sql subfolder. This is because dbt-ibis simply compiles all Ibis expressions to SQL and then let's DBT do its thing. You should not edit those files as they are overwritten every time you execute a dbt-ibis command. However, it might be interesting to look at them if you want to debug an expression. dbt-ibis will take care of cleaning up unused .sql files in these subfolders which might happen if you rename or delete an .ibis file.

You can also execute dbt-ibis precompile if you only want to compile the .ibis to .sql files:

# This
dbt-ibis run

# Is the samee as
dbt-ibis precompile
dbt run

Editor configuration

You might want to configure your editor to treat .ibis files as normal Python files. In VS Code, you can do this by putting the following into your settings.json file:

    "files.associations": {
        "*.ibis": "python"
    },

Column name casing

dbt-ibis relies on the default Ibis behavior when it comes to quoting column names and writing them as upper or lowercase which can depend on your database. However, for databases such as Snowflake which store case-insensitive identifiers in all uppercase letters, this might not always be the most convenient for you to write dbt models. This section shows with an example what different case conventions can mean and how you can configure dbt-ibis to deal with it. For databases where identifiers are always case-insensitive, e.g. DuckDB, you can skip this part.

For the columns, for which dbt-ibis loads the data types from the .yml files (see above), it assumes that the column name appears exactly in the database as they are specified in the .yml file. Taking the following example:

models:
  - name: customers
    columns:
      - name: customer_id
        data_type: integer
      - name: customer_name
        data_type: varchar

and a dbt model which references this table:

from dbt_ibis import ref, depends_on

@depends_on(ref("customers"))
def model(customers):
    return customers.select("customer_id")

This will be rendered as the following query if you're using Snowflake:

SELECT
  t0."customer_id"
FROM {{ ref('customers') }} as t0

If the column identifier is stored as case-insensitive, this query will fail as the lowercase column "customer_id" does not exist. To fix this, you'll have to write the column names in the .yml file in uppercase:

models:
  - name: customers
    columns:
      - name: CUSTOMER_ID
        data_type: integer
      - name: CUSTOMER_NAME
        data_type: varchar

and also change it in the model

@depends_on(ref("customers"))
def model(customers):
    return customers.select("CUSTOMER_ID")

If you want to keep using lowercase column names in your model but case-insensitive (i.e. uppercase) identifiers in the database, it would look something like this:

@depends_on(ref("customers"))
def model(customers):
    customers = customers.rename("snake_case")
    customers = customers.select("customer_id")
    customers = customers.rename("ALL_CAPS")
    return customers

This is rather cumbersome to do for every model and many of us are used to work with lowercase column names as a convention. To simplify the process, you can tell dbt-ibis to do these conversions for you. Going back to our original example of using all lowercase names in the .yml file as well as in the model, you can make that work by setting the following variables in your dbt_project.yml file:

vars:
  dbt_ibis_letter_case_in_db_jaffle_shop_prod: upper
  dbt_ibis_letter_case_in_model: lower

This tells dbt-ibis that in the database, uppercase letters should be used and can be expected, and that in your dbt model you want to use lowercase letters. Both variables accept upper and lower as values. In addition, the first variable is specific to a profile (jaffle_shop) and target (prod) following the format dbt_ibis_letter_case_in_db_{profile}_{target}. This allows you to set different conventions for different databases. If in the above example, you would have a dev target which points to a local duckdb file, this dev target would still use the default letter case behavior of Ibis.

If all of this sounds confusing, I'd recommend to play around with the different configurations and run dbt-ibis precompile to inspect the generated SQL. If you have any questions, feel free to open an Issue in this repository.

See this GitHub issue for some further explanations and examples on case handling in Ibis and Snowflake.

Limitations

• There is no database connection available in the Ibis model functions. Hence, you cannot use Ibis functionality which would require this.
• For non-Ibis models, seeds, snapshots, and for sources, you need to specify the data types of the columns. See "Basic example" above.

Advanced

Use dbt command instead of dbt-ibis

If you want to continue to use dbt instead of dbt-ibis on the command line, you can configure an alias in your shell. If you use bash, you can add the following to your ~/.bashrc file:

alias dbt="dbt-ibis"

See here for more detailed instructions if you use Bash and here for Zsh.

CI/CD integration

As dbt-ibis compiles your .ibis files into .sql, it can be useful to check in your CI/CD pipeline if these files are in sync, i.e. a run of dbt-ibis precompile should not change any .sql files anymore. You can achieve this with something like:

#!/bin/bash

dbt-ibis precompile
# This gets the paths of all files which were either deleted, modified
# or are not yet tracked by Git
files=`git ls-files --deleted --modified --others --exclude-standard`
# Depending on the shell it can happen that 'files' contains empty
# lines which are filtered out in the for loop below
files_cleaned=()
for i in "${files[@]}"; do
# Skip empty items
if [ -z "$i" ]; then
    continue
fi
# Add the rest of the elements to a new array
files_cleaned+=("${i}")
done
if [ ${#files_cleaned[@]} -gt 0 ]; then
    echo "The dbt-ibis precompile command modified the following files:"
    echo $files
    exit 1
fi

Potential closer integration with DBT

There are discussions on adding a plugin system to dbt which could be used to provide first-class support for other modeling languages such as Ibis (see this PoC by dbt and the discussion on Ibis as a dataframe API) or PRQL (see dbt-prql).

As this feature didn't make it onto the roadmap of dbt for 2023, I've decided to create dbt-ibis to bridge the time until then. Apart from the limitations mentioned above, I think this approach can scale reasonably well. However, the goal is to migrate to the official plugin system as soon as it's available.

Development

pip install -e '.[dev]'

You can run linters and tests with

hatch run linters
hatch run tests