Summary
- Google Cloud added preview support for BigQuery federated queries against MySQL, Oracle, and PostgreSQL databases on 13 November.
- The `EXTERNAL_QUERY` function executes read-only SQL against a connected database and returns results to BigQuery for analysis.
- Federation avoids some advance data replication but creates performance, security, networking, and cost considerations because source databases still process queries and transfer results.
Google Cloud has expanded BigQuery’s ability to query external operational databases, introducing connections for MySQL, Oracle, and PostgreSQL that allow analysts to access existing information without first copying complete datasets into the analytics platform.
The three connections became available in preview on 13 November, extending BigQuery’s federated querying functionality to additional relational database environments commonly used by businesses. Users can run queries against supported databases and combine the returned results with information already held in BigQuery.
The update provides another integration option for organisations managing operational systems alongside analytical platforms. Although it can eliminate the need to replicate an entire database before performing a particular analysis, the relevant query results must still be transferred into BigQuery for processing.
All three new capabilities remain in preview rather than general availability. Businesses evaluating them will need to consider the support and service limitations associated with Google’s pre-release offerings before relying on the connections for important production workloads.
External queries run against source databases
BigQuery federation uses the platform’s Connection API to establish access to an external database. Analysts can then use the EXTERNAL_QUERY function to execute a SQL statement through the configured connection.
The statement is processed by the source database using its own SQL dialect. Its results are returned to BigQuery as a temporary table, with values converted into compatible GoogleSQL data types so they can be used alongside other information.
An organisation might, for example, hold current transaction records in PostgreSQL while maintaining historical or aggregated customer information in BigQuery. Federation would allow an analyst to retrieve selected operational records for a particular query without constructing a separate process to copy the complete source database.
The arrangement differs from a conventional data transfer pipeline, which moves information into an analytical environment before users query it. Federation can reduce the preparation required for certain questions, particularly where the necessary data changes frequently or is accessed only occasionally.
It does not remove the need to understand the source database. Analysts must still identify relevant tables, write suitable queries, and interpret the results correctly, while administrators establish the credentials and network connections required for access.
Data conversion introduces another consideration. MySQL, PostgreSQL, Oracle, and BigQuery support different data types and SQL behaviour, meaning some values require conversion before they can be combined with information from other sources.
Those differences may affect financial calculations, timestamps, or other fields where precision and consistent interpretation are important.
Established database environments gain another route
MySQL and PostgreSQL are widely used in business applications, while Oracle databases remain an important component of many established enterprise systems. Organisations may maintain substantial operational information within those environments even when they use a separate platform for wider reporting.
The new connections support specified database versions and deployment arrangements rather than every system associated with the three products.
For Oracle, Google’s documentation identifies supported versions including 19c, 21c, and 23ai. Connections can operate with environments such as Oracle Database@Google Cloud, Oracle Exadata Database Service, and on-premises installations connected through supported networking arrangements.
MySQL federation supports versions 8.0 and 8.4 LTS, including supported cloud-hosted and on-premises deployments. Google maintains a separate federation connector for Cloud SQL for MySQL, which has its own technical characteristics.
These distinctions matter because two databases using the same underlying technology may have different connection requirements depending on how they are hosted and administered.
Administrators must configure the connection resource, establish suitable credentials, and provide the necessary identity and access management permissions. Supported network arrangements include public connectivity and private connections through appropriate Google Cloud services.
For Oracle and the other new connectors, the location in which a federated query runs must match the region assigned to its connection resource. Private networking arrangements introduce further geographical and configuration requirements.
Consequently, businesses need to assess where the database is located, how the query reaches it, and which region processes the returned information before deployment.
Direct access introduces performance trade-offs
Executing a query against an operational database places work on that system. If the source is already processing business transactions, additional analytical activity may affect the resources available to its existing applications.
Google warns that federated queries may be slower than queries performed entirely against data stored in BigQuery. The analytical platform must wait for the external database to execute the statement and transfer the results before subsequent processing can finish.
Connection behaviour can also introduce delays. Google’s documentation notes that inactive connections may be placed into a dormant state, meaning the first query after a period of inactivity can require additional time while the connector initialises.
The quantity of information returned has a direct effect on performance and network costs. A query retrieving selected columns and a small number of relevant rows may be considerably more efficient than one transferring a complete table for further filtering in BigQuery.
Google recommends limiting the information requested from the external database. Certain query optimisations can push filters and column selections towards the source, although the supported behaviour varies by connector and query structure.
Federation therefore does not automatically provide the same performance as a carefully designed data replication process. Where organisations repeatedly analyse large volumes of the same information, maintaining a separate analytical copy may remain more appropriate.
Conversely, an occasional query involving a relatively small amount of current operational information may not justify the expense and maintenance associated with a full transfer pipeline.
Permissions and costs remain part of the design
The new connections support read-only queries through EXTERNAL_QUERY. Users cannot employ the function to modify source database records or execute data definition statements.
That restriction limits the actions available through federation, but read-only access can still expose sensitive commercial or personal information if the connection has broader permissions than the intended analysis requires.
Google’s Oracle setup documentation recommends a dedicated database account with permission to read only the necessary tables or views. Comparable principles apply when organisations configure access to other operational databases.
The service also requires administrators to manage the permissions allowing BigQuery users to access individual connection resources. These controls determine who can issue queries against the connected source.
Cost calculations span several systems. The external database may incur computing charges while processing a query, transferring results can generate network egress costs, and BigQuery charges for its own processing.
Google does not impose an additional federation-specific charge, but that does not mean an end-to-end federated query is free. Networking components such as private connectivity or VPN services can create further expenditure.
Governance is particularly relevant where businesses operate across the UK and European Union. Organisations must establish whether the movement and processing of query results complies with their existing requirements concerning data location, confidentiality, and supplier access.
The use of an external connection does not guarantee that information remains exclusively within the source database. Results are transferred into BigQuery, making the processing location and permissions part of the compliance assessment.
There are also technical limits. Google’s documentation specifies a maximum of ten unique connections within a federated query, which could constrain analytical operations involving numerous separate database environments.
Preview functionality requires operational testing
BigQuery already supports federation with several other database services, including Spanner, Cloud SQL, and AlloyDB. The November update broadens that approach rather than introducing the underlying concept for the first time.
Google has not published independently verified performance benchmarks for the newly introduced MySQL, Oracle, and PostgreSQL connections. Their suitability for a particular business will depend on database configuration, network performance, query characteristics, and the resources available to the source system.
Organisations assessing the preview may therefore need to compare federation with established alternatives such as scheduled replication, existing integration services, or analytical copies maintained within BigQuery.
The new connections are available for evaluation, with administrators responsible for configuring access and the supporting network arrangements. Wider adoption will depend on whether direct querying provides sufficient flexibility without creating unacceptable performance, cost, or governance demands on operational systems.












