- Supports dbt version
1.4.*
- Supports Seeds
- Correctly detects views and their columns
- Supports table materialization
- Iceberg tables is supported only with Athena Engine v3 and a unique table location (see table location section below)
- Hive tables is supported by both Athena engines.
- Supports incremental models
- On iceberg tables :
- Support the use of
unique_key
only with themerge
strategy - Support the
append
strategy
- Support the use of
- On Hive tables :
- Support two incremental update strategies:
insert_overwrite
andappend
- Does not support the use of
unique_key
- Support two incremental update strategies:
- On iceberg tables :
- Supports snapshots
- Does not support Python models
pip install dbt-athena-community
- Or
pip install git+https://github.com/dbt-athena/dbt-athena.git
To start, you will need an S3 bucket, for instance my-bucket
and an Athena database:
CREATE DATABASE IF NOT EXISTS analytics_dev
COMMENT 'Analytics models generated by dbt (development)'
LOCATION 's3:https://my-bucket/'
WITH DBPROPERTIES ('creator'='Foo Bar', 'email'='[email protected]');
Notes:
- Take note of your AWS region code (e.g.
us-west-2
oreu-west-2
, etc.). - You can also use AWS Glue to create and manage Athena databases.
This plugin does not accept any credentials directly. Instead, credentials are determined automatically based on aws cli
/boto3
conventions and
stored login info. You can configure the AWS profile name to use via aws_profile_name
. Checkout DBT profile configuration below for details.
A dbt profile can be configured to run against AWS Athena using the following configuration:
Option | Description | Required? | Example |
---|---|---|---|
s3_staging_dir | S3 location to store Athena query results and metadata | Required | s3:https://bucket/dbt/ |
s3_data_dir | Prefix for storing tables, if different from the connection's s3_staging_dir |
Optional | s3:https://bucket2/dbt/ |
s3_data_naming | How to generate table paths in s3_data_dir |
Optional | schema_table_unique |
region_name | AWS region of your Athena instance | Required | eu-west-1 |
schema | Specify the schema (Athena database) to build models into (lowercase only) | Required | dbt |
database | Specify the database (Data catalog) to build models into (lowercase only) | Required | awsdatacatalog |
poll_interval | Interval in seconds to use for polling the status of query results in Athena | Optional | 5 |
aws_profile_name | Profile to use from your AWS shared credentials file. | Optional | my-profile |
work_group | Identifier of Athena workgroup | Optional | my-custom-workgroup |
num_retries | Number of times to retry a failing query | Optional | 3 |
Example profiles.yml entry:
athena:
target: dev
outputs:
dev:
type: athena
s3_staging_dir: s3:https://athena-query-results/dbt/
s3_data_dir: s3:https://your_s3_bucket/dbt/
s3_data_naming: schema_table
region_name: eu-west-1
schema: dbt
database: awsdatacatalog
aws_profile_name: my-profile
work_group: my-workgroup
Additional information
threads
is supporteddatabase
andcatalog
can be used interchangeably
external_location
(default=none
)- If set, the full S3 path in which the table will be saved. (Does not work with Iceberg table).
partitioned_by
(default=none
)- An array list of columns by which the table will be partitioned
- Limited to creation of 100 partitions (currently)
bucketed_by
(default=none
)- An array list of columns to bucket data, ignored if using Iceberg
bucket_count
(default=none
)- The number of buckets for bucketing your data, ignored if using Iceberg
table_type
(default='hive'
)- The type of table
- Supports
hive
oriceberg
format
(default='parquet'
)- The data format for the table
- Supports
ORC
,PARQUET
,AVRO
,JSON
,TEXTFILE
write_compression
(default=none
)- The compression type to use for any storage format that allows compression to be specified. To see which options are available, check out [CREATE TABLE AS][create-table-as]
field_delimiter
(default=none
)- Custom field delimiter, for when format is set to
TEXTFILE
- Custom field delimiter, for when format is set to
table_properties
: table properties to add to the table, valid for Iceberg only
The location in which a table is saved is determined by:
- If
external_location
is defined, that value is used. - If
s3_data_dir
is defined, the path is determined by that ands3_data_naming
- If
s3_data_dir
is not defined data is stored unders3_staging_dir/tables/
Here all the options available for s3_data_naming
:
uuid
:{s3_data_dir}/{uuid4()}/
table_table
:{s3_data_dir}/{table}/
table_unique
:{s3_data_dir}/{table}/{uuid4()}/
schema_table
:{s3_data_dir}/{schema}/{table}/
s3_data_naming=schema_table_unique
:{s3_data_dir}/{schema}/{table}/{uuid4()}/
It's possible to set the s3_data_naming
globally in the target profile, or overwrite the value in the table config,
or setting up the value for groups of model in dbt_project.yml
Support for incremental models.
These strategies are supported:
insert_overwrite
(default): The insert overwrite strategy deletes the overlapping partitions from the destination table, and then inserts the new records from the source. This strategy depends on thepartitioned_by
keyword! If no partitions are defined, dbt will fall back to theappend
strategy.append
: Insert new records without updating, deleting or overwriting any existing data. There might be duplicate data (e.g. great for log or historical data).merge
: Conditionally updates, deletes, or inserts rows into an Iceberg table. Used in combination withunique_key
. Only available when using Iceberg.
on_schema_change
is an option to reflect changes of schema in incremental models.
The following options are supported:
ignore
(default)fail
append_new_columns
sync_all_columns
In detail, please refer to dbt docs.
The adapter supports table materialization for Iceberg.
To get started just add this as your model:
{{ config(
materialized='table',
table_type='iceberg',
format='parquet',
partitioned_by=['bucket(user_id, 5)'],
table_properties={
'optimize_rewrite_delete_file_threshold': '2'
}
) }}
SELECT
'A' AS user_id,
'pi' AS name,
'active' AS status,
17.89 AS cost,
1 AS quantity,
100000000 AS quantity_big,
current_date AS my_date
Iceberg supports bucketing as hidden partitions, therefore use the partitioned_by
config to add specific bucketing conditions.
Iceberg supports several table formats for data : PARQUET
, AVRO
and ORC
.
It is possible to use iceberg in an incremental fashion, specifically 2 strategies are supported:
append
: new records are appended to the table, this can lead to duplicatesmerge
: must be used in combination withunique_key
and it's only available with Engine version 3. It performs an upsert, new record are added, and record already existing are updated
The current implementation of the table materialization can lead to downtime, as target table is dropped and re-created.
To have the less destructive behavior it's possible to use table='table_hive_ha'
materialization.
table_hive_ha leverage the table versions feature of glue catalog, creating a tmp table and swapping
the target table to the location of the tmp table.
This materialization is only available for table_type=hive
and requires using unique locations.
{{ config(
materialized='table_hive_ha',
format='parquet',
partition_by=['status'],
s3_data_naming='table_unique'
) }}
select
'a' as user_id,
'pi' as user_name,
'active' as status
union all
select
'b' as user_id,
'sh' as user_name,
'disabled' as status
By default, the materialization keeps the last 4 table versions, you can change it that setting versions_to_keep
.
- When swapping from a table with partitions to a table without (and the other way around), there could be a little downtime. In case high performances are needed consider bucketing instead of partitions
- By default, Glue "duplicate" the versions internally, so the last 2 versions of a table point to the same location
- It's recommended to have versions_to_keep>= 4, as this will avoid to have the older location removed
The adapter supports snapshot materialization. It supports both timestamp and check strategy. To create a snapshot create a snapshot file in the snapshots directory. If directory does not exist create one.
To use the timestamp strategy refer to the dbt docs
To use the check strategy refer to the dbt docs
The materialization also supports invalidating hard deletes. Check the docs to understand usage.
-
Incremental Iceberg models - Sync all columns on schema change can't remove columns used as partitioning. The only way, from a dbt perspective, is to do a full-refresh of the incremental model.
-
Tables, schemas and database should only be lowercase
-
In order to avoid potential conflicts, make sure
dbt-athena-adapter
is not installed in the target environment. See dbt-labs#103 for more details. -
Snapshot does not support dropping columns from the source table. If you drop a column make sure to drop the column from the snapshot as well. Another workaround is to NULL the column in the snapshot definition to preserve history
This connector works with Python from 3.7 to 3.11.
In order to start developing on this adapter clone the repo and run this make command (see Makefile) :
make setup
It will :
- Install all dependencies.
- Install pre-commit hooks.
- Generate your
.env
file
Next, adjust .env
file by configuring the environment variables to match your Athena development environment.
We have 2 different types of testing:
- unit testing: you can run this type of tests running
make unit_test
- functional testing: you must have an AWS account with Athena setup in order to launch this type of tests and have a
.env
file in place with the right values. You can run this type of tests runningmake functional_test
All type of tests can be run using make
:
make test
- Create a commit with your changes and push them to a fork.
- Create a pull request on Github.
- Pull request title and message (and PR title and description) must adhere to conventionalcommits.
- Pull request body should describe motivation.