This documentation is for an older version (1.4.7) of Dagster. You can view the version of this page from our latest release below.
This library provides an integration with the BigQuery database and Pandas data processing library.
Related Guides:
The GCP project to use.
Name of the BigQuery dataset to use. If not provided, the last prefix before the asset name will be used.
The GCP location. Note: When using PySpark DataFrames, the default location of the project will be used. A custom location can be specified in your SparkSession configuration.
GCP authentication credentials. If provided, a temporary file will be created with the credentials and GOOGLE_APPLICATION_CREDENTIALS will be set to the temporary file. To avoid issues with newlines in the keys, you must base64 encode the key. You can retrieve the base64 encoded key with this shell command: cat $GOOGLE_AUTH_CREDENTIALS | base64
When using PySpark DataFrames, optionally specify a temporary GCS bucket to store data. If not provided, data will be directly written to BigQuery.
When using Pandas DataFrames, optionally specify a timeout for the BigQuery queries (loading and reading from tables).
An I/O manager definition that reads inputs from and writes pandas DataFrames to BigQuery.
IOManagerDefinition
Examples
from dagster_gcp_pandas import BigQueryPandasIOManager
from dagster import Definitions, EnvVar
@asset(
key_prefix=["my_dataset"] # will be used as the dataset in BigQuery
)
def my_table() -> pd.DataFrame: # the name of the asset will be the table name
...
defs = Definitions(
assets=[my_table],
resources={
"io_manager": BigQueryPandasIOManager(project=EnvVar("GCP_PROJECT"))
}
)
You can tell Dagster in which dataset to create tables by setting the “dataset” configuration value. If you do not provide a dataset as configuration to the I/O manager, Dagster will determine a dataset based on the assets and ops using the I/O Manager. For assets, the dataset will be determined from the asset key, as shown in the above example. The final prefix before the asset name will be used as the dataset. For example, if the asset “my_table” had the key prefix [“gcp”, “bigquery”, “my_dataset”], the dataset “my_dataset” will be used. For ops, the dataset can be specified by including a “schema” entry in output metadata. If “schema” is not provided via config or on the asset/op, “public” will be used for the dataset.
@op(
out={"my_table": Out(metadata={"schema": "my_dataset"})}
)
def make_my_table() -> pd.DataFrame:
# the returned value will be stored at my_dataset.my_table
...
To only use specific columns of a table as input to a downstream op or asset, add the metadata “columns” to the In or AssetIn.
@asset(
ins={"my_table": AssetIn("my_table", metadata={"columns": ["a"]})}
)
def my_table_a(my_table: pd.DataFrame) -> pd.DataFrame:
# my_table will just contain the data from column "a"
...
If you cannot upload a file to your Dagster deployment, or otherwise cannot authenticate with GCP via a standard method, you can provide a service account key as the “gcp_credentials” configuration. Dagster will store this key in a temporary file and set GOOGLE_APPLICATION_CREDENTIALS to point to the file. After the run completes, the file will be deleted, and GOOGLE_APPLICATION_CREDENTIALS will be unset. The key must be base64 encoded to avoid issues with newlines in the keys. You can retrieve the base64 encoded key with this shell command: cat $GOOGLE_APPLICATION_CREDENTIALS | base64
The GCP project to use.
Name of the BigQuery dataset to use. If not provided, the last prefix before the asset name will be used.
The GCP location. Note: When using PySpark DataFrames, the default location of the project will be used. A custom location can be specified in your SparkSession configuration.
GCP authentication credentials. If provided, a temporary file will be created with the credentials and GOOGLE_APPLICATION_CREDENTIALS will be set to the temporary file. To avoid issues with newlines in the keys, you must base64 encode the key. You can retrieve the base64 encoded key with this shell command: cat $GOOGLE_AUTH_CREDENTIALS | base64
When using PySpark DataFrames, optionally specify a temporary GCS bucket to store data. If not provided, data will be directly written to BigQuery.
When using Pandas DataFrames, optionally specify a timeout for the BigQuery queries (loading and reading from tables).
An I/O manager definition that reads inputs from and writes pandas DataFrames to BigQuery.
IOManagerDefinition
Examples
from dagster_gcp_pandas import bigquery_pandas_io_manager
from dagster import Definitions
@asset(
key_prefix=["my_dataset"] # will be used as the dataset in BigQuery
)
def my_table() -> pd.DataFrame: # the name of the asset will be the table name
...
defs = Definitions(
assets=[my_table],
resources={
"io_manager": bigquery_pandas_io_manager.configured({
"project" : {"env": "GCP_PROJECT"}
})
}
)
You can tell Dagster in which dataset to create tables by setting the “dataset” configuration value. If you do not provide a dataset as configuration to the I/O manager, Dagster will determine a dataset based on the assets and ops using the I/O Manager. For assets, the dataset will be determined from the asset key, as shown in the above example. The final prefix before the asset name will be used as the dataset. For example, if the asset “my_table” had the key prefix [“gcp”, “bigquery”, “my_dataset”], the dataset “my_dataset” will be used. For ops, the dataset can be specified by including a “schema” entry in output metadata. If “schema” is not provided via config or on the asset/op, “public” will be used for the dataset.
@op(
out={"my_table": Out(metadata={"schema": "my_dataset"})}
)
def make_my_table() -> pd.DataFrame:
# the returned value will be stored at my_dataset.my_table
...
To only use specific columns of a table as input to a downstream op or asset, add the metadata “columns” to the In or AssetIn.
@asset(
ins={"my_table": AssetIn("my_table", metadata={"columns": ["a"]})}
)
def my_table_a(my_table: pd.DataFrame) -> pd.DataFrame:
# my_table will just contain the data from column "a"
...
If you cannot upload a file to your Dagster deployment, or otherwise cannot authenticate with GCP via a standard method, you can provide a service account key as the “gcp_credentials” configuration. Dagster will store this key in a temporary file and set GOOGLE_APPLICATION_CREDENTIALS to point to the file. After the run completes, the file will be deleted, and GOOGLE_APPLICATION_CREDENTIALS will be unset. The key must be base64 encoded to avoid issues with newlines in the keys. You can retrieve the base64 encoded key with this shell command: cat $GOOGLE_APPLICATION_CREDENTIALS | base64
Plugin for the BigQuery I/O Manager that can store and load Pandas DataFrames as BigQuery tables.
Examples
from dagster_gcp import BigQueryIOManager
from dagster_bigquery_pandas import BigQueryPandasTypeHandler
from dagster import Definitions, EnvVar
class MyBigQueryIOManager(BigQueryIOManager):
@staticmethod
def type_handlers() -> Sequence[DbTypeHandler]:
return [BigQueryPandasTypeHandler()]
@asset(
key_prefix=["my_dataset"] # my_dataset will be used as the dataset in BigQuery
)
def my_table() -> pd.DataFrame: # the name of the asset will be the table name
...
defs = Definitions(
assets=[my_table],
resources={
"io_manager": MyBigQueryIOManager(project=EnvVar("GCP_PROJECT"))
}
)