- Create a service account in your Google Cloud project. Grant BigQuery Job User on the project that runs and pays for jobs, plus Data Viewer on read datasets or Data Editor on write datasets.
- In Connections → BigQuery, upload or paste the service-account JSON. Enter the data project, job billing project, dataset location and maximum bytes billed per query. Test and save verifies access and encrypts the credential.
- Add a BigQuery step and choose your connection. Select List datasets, List tables, Retrieve table schema, Execute parameterized SQL, or Insert rows.
- For queries, keep Dry run enabled first. Inspect the estimated bytes, billing project and location, then disable Dry run when you want workflow execution to retrieve rows.
SELECT @name AS name[{"name":"name","type":"STRING","value":"{{input.name}}"}]SQL text stays static. Bind values with named parameters: STRING, INT64, FLOAT64, BOOL, DATE, TIMESTAMP or NUMERIC. Query actions accept a single SELECT; scripts and SQL writes are rejected. The connection limit and node limit both apply, with the smaller limit used.
Choose the existing dataset and table for inserts. Provide a JSON array matching its schema: at most 500 rows and 1 MB per batch. Loads are atomic: failure reports zero inserted rows and safe error codes. Retry the same workflow event to recover its original job; a new event is a new write.
Responses are limited to 1,000 rows per page and 2 MB. Use a returned page token for the next page. A running job can be polled through the authenticated connection job endpoint; timeouts do not prove that a write failed. Do not retry an uncertain write with a new event ID.
To rotate a customer key, save and test a new connection, update the affected workflows, then disconnect the old connection and revoke its key in Google Cloud. Saved private keys are never returned by the Connections API or included in workflow configuration.