Model synchronization from dbt to Metabase.
If dbt is your source of truth for database schemas and you use Metabase as your analytics tool, dbt-metabase can propagate table relationships, model and column descriptions and semantic types (e.g. currency, category, URL) to your Metabase data model.
Requires Python 3.6 or above.
The main features provided by dbt-metabase are:
- Parsing your dbt project (either through the
manifest.json
or directly through the YAML files) - Triggering a Metabase schema sync before propagating the metadata
- Propagating table descriptions to Metabase
- Propagating columns description to Metabase
- Propagating columns semantic types and visibility types to Metabase through the use of dbt meta fields
- Propagating table relationships represented as dbt
relationships
column tests - Extracting dbt model exposures from Metabase and generating YAML files to be included and revisioned with your dbt deployment
You can install dbt-metabase from PyPI:
pip install dbt-metabase
Sections below demonstrate basic usage examples, for all CLI options:
dbt-metabase --help
When invoking programmatically, click through to implementation and refer to header comments.
Let's start by defining a short sample schema.yml
as below.
models:
- name: stg_users
description: User records.
columns:
- name: id
description: Primary key.
tests:
- not_null
- unique
- name: email
description: User's email address.
- name: group_id
description: Foreign key to user group.
tests:
- not_null
- relationships:
to: ref('groups')
field: id
- name: stg_groups
description: User groups.
columns:
- name: id
description: Primary key.
tests:
- not_null
- unique
- name: name
description: Group name.
That's already enough to propagate the primary keys, foreign keys and descriptions to Metabase by executing the below command.
dbt-metabase models \
--dbt_path . \
--dbt_database business \
--metabase_host metabase.example.com \
--metabase_user [email protected] \
--metabase_password Password123 \
--metabase_database business \
--dbt_schema public
Check your Metabase instance by going into Settings > Admin > Data Model, you
will notice that ID
in STG_USERS
is now marked as "Entity Key" and
GROUP_ID
is marked as "Foreign Key" pointing to ID
in STG_GROUPS
.
dbt-metabase also allows us to extract exposures from Metabase. The invocation is almost identical to our models function with the addition of output name and location args. dbt exposures let us understand how our dbt models are exposed in BI which closes the loop between ELT, modelling, and consumption.
dbt-metabase exposures \
--dbt_manifest_path ./target/manifest.json \
--dbt_database business \
--metabase_host metabase.example.com \
--metabase_user [email protected] \
--metabase_password Password123 \
--metabase_database business \
--output_path ./models/ \
--output_name metabase_exposures
Once execution completes, a look at the output metabase_exposures.yml
will
reveal all metabase exposures documented with the documentation, descriptions, creator
emails & names, links to exposures, and even native SQL propagated over from Metabase.
exposures:
- name: Number_of_orders_over_time
description: '
### Visualization: Line
A line chart depicting how order volume changes over time
#### Metadata
Metabase Id: __8__
Created On: __2021-07-21T08:01:38.016244Z__'
type: analysis
url: http://your.metabase.com/card/8
maturity: medium
owner:
name: Indiana Jones
email: [email protected]
depends_on:
- ref('orders')
Questions which are native queries will have the SQL propagated to a code block in the documentation's
description for full visibility. This YAML, like the rest of your dbt project can be committed to source
control to understand how exposures change over time. In a production environment, one can trigger
dbt docs generate
after dbt-metabase exposures
(or alternatively run the exposure extraction job
on a cadence every X days) in order to keep a dbt docs site fully synchronized with BI. This makes dbt docs
a
useful utility for introspecting the data model from source -> consumption with zero extra/repeated human input.
There are two approaches provided by this library to read your dbt project:
You can instruct dbt-metabase to read your manifest.json
, a dbt artifact containing
the full representation of your dbt project's resources. If your dbt project uses multiple schemas,
multiple databases or model aliases, you must use this approach.
Note that you you have to run dbt compile --target prod
or any of the other dbt commands
listed in the dbt documentation above to get a fresh copy of your manifest.json
. Remember
to run it against your production target.
When using the dbt-metabase
CLI, you must provide a --dbt_manifest_path
argument
pointing to your manifest.json
file (usually in the target/
folder of your dbt
project).
Alternatively, you can provide the path to your dbt project root folder using the argument
--dbt_path
. dbt-metabase will then look for all .yml files and parse your documentation
and tests directly from there. It does not support dbt projects with custom schemas.
Now that we have primary and foreign keys, let's tell Metabase that email
column contains email addresses.
Change the email
column as follows:
- name: email
description: User's email address.
meta:
metabase.semantic_type: type/Email
Once you run dbt-metabase models
again, you will notice that EMAIL
is
now marked as "Email".
Here are common semantic types (formerly known as special types) accepted by Metabase:
type/PK
type/FK
type/Number
type/Currency
type/Category
type/Title
type/Description
type/City
type/State
type/ZipCode
type/Country
type/Latitude
type/Longitude
type/Email
type/URL
type/ImageURL
type/SerializedJSON
type/CreationTimestamp
See documentation for a more complete list.
Built-in relationship tests are the recommended way of defining foreign keys,
however you can alternatively use fk_target_table
and fk_target_field
meta fields (semantic_type
is optional and will be inferred). If both are
set for a column, meta fields take precedence.
- name: country_id
description: FK to User's country in the dim_countries table.
meta:
metabase.semantic_type: type/FK
metabase.fk_target_table: analytics_dims.dim_countries
metabase.fk_target_field: id
You can provide fk_target_table
in the format schema_name.table_name
or
just table_name
to use the current schema. If your model has an alias, provide
that alias (rather than the original name).
In addition to semantic types, you can optionally specify visibility for each table and field. This affects whether or not they are displayed in the Metabase UI.
Here is how you would hide that same email:
- name: email
description: User's email address.
meta:
metabase.semantic_type: type/Email
metabase.visibility_type: sensitive
Here are the field visibility types supported by Metabase:
normal
(default)details-only
sensitive
Tables only support the following:
- No value for visible (default)
hidden
technical
cruft
If you notice new ones, please submit a PR to update this readme.
In addition to the model description, Metabase accepts two extra information fields. Those optional
fields are called caveats
and points_of_interest
and can be defined under the meta
tag
of the model.
This is how you can specify them in the stg_users
example:
- name: stg_users
description: User records.
meta:
metabase.points_of_interest: Relevant records.
metabase.caveats: Sensitive information about users.
By default, dbt-metabase will tell Metabase to synchronize database fields and wait for the data model to contain all the tables and columns in your dbt project.
You can control this behavior with two arguments:
--metabase_sync_skip
- boolean to optionally disable pre-synchronization--metabase_sync_timeout
- number of seconds to wait and re-check data model before giving up
dbt-metabase config
Using the above command, you can enter an interactive configuration session where you can cache default selections
for arguments. This creates a config.yml
in ~/.dbt-metabase. This is particularly useful for arguments which are repeated on every invocation like metabase_user, metabase_host,
metabase_password, dbt_manifest_path, etc.
In addition, there are a few injected env vars that make deploying dbt-metabase in a CI/CD environment simpler without exposing secrets. Listed below are acceptable env vars which correspond to their CLI flags:
DBT_DATABASE
DBT_PATH
DBT_MANIFEST_PATH
MB_USER
MB_PASSWORD
MB_HOST
MB_DATABASE
If any one of the above is present in the environment, the corresponding CLI flag is not needed unless overriding the environment value. In the absence of a CLI flag, dbt-metabase will first look to the environment for any env vars to inject, then we will look to the config.yml for cached defaults.
A config.yml
can be created or updated manually as well if needed. The only
requirement is that it must be located in ~/.dbt-metabase. The layout is as follows:
config:
dbt_database: reporting
dbt_manifest_path: /home/user/dbt/target/manifest.json
metabase_database: Reporting
metabase_host: reporting.metabase.io
metabase_user: [email protected]
metabase_password: ...
metabase_use_http: false
metabase_sync: true
metabase_sync_timeout: null
dbt_schema_excludes:
- development
- testing
dbt_excludes:
- test_monday_io_site_diff
As you have already seen, you can invoke dbt-metabase from the command line. But if you prefer to call it from your code, here's how to do it:
from dbtmetabase.models.interface import MetabaseInterface, DbtInterface
# Instantiate dbt interface
dbt = DbtInterface(
path=dbt_path,
manifest_path=dbt_manifest_path,
database=dbt_database,
schema=dbt_schema,
schema_excludes=dbt_schema_excludes,
includes=dbt_includes,
excludes=dbt_excludes,
)
# Load models
dbt_models, aliases = dbt.read_models(
include_tags=dbt_include_tags,
docs_url=dbt_docs_url,
)
# Instantiate Metabase interface
metabase = MetabaseInterface(
host=metabase_host,
user=metabase_user,
password=metabase_password,
use_http=metabase_use_http,
verify=metabase_verify,
database=metabase_database,
sync=metabase_sync,
sync_timeout=metabase_sync_timeout,
)
# Propagate models to Metabase
metabase.client.export_models(
database=metabase.database,
models=dbt_models,
aliases=aliases,
)
# Parse exposures from Metabase into dbt schema yml
metabase.client.extract_exposures(
models=dbt_models,
output_path=output_path,
output_name=output_name,
include_personal_collections=include_personal_collections,
collection_excludes=collection_excludes,
)
All contributors are expected to follow the PyPA Code of Conduct.