Starburst Snowflake connector#

The Snowflake connector allows querying and creating tables in an external Snowflake database. This can be used to join data between different systems like Snowflake and Hive, or between different Snowflake instances.

Note

The connector requires a valid Starburst Enterprise license.

Configuration#

To configure the Snowflake connector, create a catalog properties file in etc/catalog named, for example, snowflake.properties, to mount the Snowflake connector as the snowflake catalog.

There are two flavors of the Snowflake connector:

snowflake-jdbc
snowflake-distributed

snowflake-jdbc uses JDBC for all reads and writes and is more efficient when the result set returned from Snowflake is small.

When larger result sets are extracted from Snowflake, the snowflake-distributed connector may be a better choice. Instead of requesting query results over a JDBC connection, the connector asks Snowflake to export them to S3 and SEP reads them from there. Since both the write and the read are parallelized, this approach scales better for large data sets, but has a higher latency.

A requirement of using the distributed connector is that your Snowflake deployment runs on AWS. The connector automatically creates temporary stages in temporary S3 buckets using Snowflake.

Create the catalog properties file with the following contents, replacing the connection properties as appropriate for your setup (for example, replace <account_name> with the full name of your account, as provided by Snowflake).

connector.name=<snowflake-jdbc or snowflake-distributed>
connection-url=jdbc:snowflake://<account_name>.snowflakecomputing.com/
connection-user=<user_name>
connection-password=<password>
snowflake.warehouse=<warehouse_name>
snowflake.database=<database_name>

The role used by Snowflake to execute operations can be specified as snowflake.role=<role_name>. This configuration is optional, and can not be used together with User impersonation.

Additionally, there are a number of configuration properties that apply only to the distributed connector.

Distributed connector configuration properties#
Property name	Description	Default
`snowflake.stage-schema`	Name of the schema in which stages are created for exporting data
`snowflake.max-export-retries`	Number of export retries	3
`snowflake.parquet.max-read-block-size`	Maximum block size when reading from the export file	16MB
`snowflake.max-split-size`	Maximum split size for processing the export file	64MB
`snowflake.max-initial-split-size`	Maximum initial split size	Half of `snowflake.max-split-size`
`snowflake.export-file-max-size`	Maximum size of files to create when exporting data	16MB

Performance#

The connector includes a number of performance improvements, detailed in the following sections.

Table statistics#

The Snowflake connector supports only table statistics to improve query processing performance based on the actual data in the data source.

The statistics are collected by Snowflake and retrieved by the connector.

They are based on Snowflake’s INFORMATION_SCHEMA.TABLES table. Table statistics are automatically updated by Snowflake.

Pushdown#

The connector supports pushdown for a number of operations:

Aggregate pushdown for the following functions:

Dynamic filtering#

Dynamic filtering is enabled by default. It causes the connector to wait for dynamic filtering to complete before starting a JDBC query.

You can disable dynamic filtering by setting the property dynamic-filtering.enabled in your catalog properties file to false.

JDBC connection pooling#

You can improve performance by enabling JDBC connection pooling, which is disabled by default.

Caching table projections#

The connectors supports table scan redirection to improve performance and reduce load on the data source.

Security#

The connector includes a number of security-related features, detailed in the following sections.

User impersonation#

The Snowflake connector supports User impersonation. It can be configured to use a number of different impersonation mechanisms, specified by the values configured for the property snowflake.impersonation-type:

snowflake.impersonation-type=ROLE

NONE: Connect as the service user with the credentials from the catalog properties file and assume the role defined by the snowflake.role property, or the service user’s default role, if the property is missing.
ROLE: Connect as the service user and use auth-to-local mapping to map the user to a role.
OKTA_LDAP_PASSTHROUGH: Assume the identity of the SEP user, authenticated with Okta using LDAP credentials, and use the default Snowflake user role.
ROLE_OKTA_LDAP_PASSTHROUGH: As above, but additionally use auth-to-local mapping to map the user to a role.

Authentication with Okta#

The Snowflake connector supports the usage of the Okta Single Sign-On system to authenticate users via SEP to Snowflake.

The setup allows users to authenticate to SEP using their LDAP credentials and use the same credentials to authenticate to Snowflake through Okta. The same credentials are used by SEP when accessing data in Snowflake.

Behind the scenes SEP and the Snowflake connector authenticate to Okta with the LDAP credentials of the user. After the user authenticates with Okta, including MFA potentially, a SAML assertion allows Snowflake to issue an an OAuth 2.0 token pair. The tokens are cached in SEP and used for further authentications until they expire, and another authentication is requested.

If Okta multi-factor authentication (MFA) is configured, users have to confirm authentication with it. One time codes are not supported.

To enable the Okta integration, SEP and Snowflake need to be configured correctly.

Okta and SEP are configured to use LDAP authentication using the same user identifiers and LDAP directory. In addition to the usual LDAP configuration, LDAP authentication, for SEP, you need to enable password forwarding in etc/config.properties:

http.server.authentication.password.forwarding-enabled=true

Snowflake is configured in Okta as a SAML application as detailed in the Snowflake documentation. Note that the Snowflake login_name must match the corresponding SAML Subject NameID attribute value.

SEP is configured as an OAuth client in Snowflake, again detailed in the Snowflake OAuth documentation.

A number of properties are required to configure the Okta integration in the Snowflake catalog properties file:

snowflake.account-name: Name of Snowflake account.
snowflake.account-url: URL of the Snowflake account. The URL usually has the form https://account_name.snowflakecomputing.com, but might include additional segments.
snowflake.client-id: Snowflake OAuth client id. This can be retrieved with the secret name and a query like select system$show_oauth_client_secrets('OAUTH_TEST_INT');.
snowflake.client-secret: Snowflake OAuth client secret.
snowflake.credential.cache-ttl: Duration the OAuth refresh token is cached. This value cannot exceed the oauth_refresh_token_validity value used when the OAuth integration was created. E.g. 24h.
okta.account-url: The Okta URL, typically https://your_okta_account_name.okta.com).

Optional properties allow you to override the default values:

snowflake.credential.cache-size: The size of the OAuth credentials cache. Use a value that accommodates the expected number of users that might connect to Snowflake through SEP during the period defined by the TTL of the token. Defaults to 10000.
snowflake.credential.http-connect-timeout: Connection timeout. Defaults to 30s.
snowflake.credential.http-read-timeout: Connection read timeout. Defaults to 30s.
snowflake.credential.http-write-timeout: Connection write timeout. Defaults to 30s.
snowflake.redirect-uri: The redirect URI for OAUTH. Value must match the redirect URI specified when creating the security integration (oauth_redirect_uri). Defaults to https://localhost.
okta.credential.http-connect-timeout: Connection timeout. Defaults to 30s.
okta.credential.http-read-timeout: Connection read timeout. Defaults to 30s.
okta.credential.http-write-timeout: Connection write timeout. Defaults to 30s.

Limitations#

The following SQL statements are not yet supported: