Run a queryRun a SQL query and get the rows back, each value already the type its column says it is - whole numbers as numbers, true/false as yes/no, timestamps as dates and times, repeated columns as lists. Waits for the query, pages the result and reports how much data it read and roughly what that cost. Also runs INSERT, UPDATE, DELETE and MERGE, and tells you how many rows changed.
Check a queryAsk BigQuery what a query would do without running it: whether it works, how much data it would read, roughly what that costs, which tables it touches and which columns it would return. Costs nothing, and never fails the step for a broken query - it reports the problem so a workflow can decide what to do about it.
Save query results to a tableRun a query and write its rows straight into a table, creating the table if it isn't there. The way to rebuild a summary table every morning without pulling a single row through the workflow. Choose whether to add to the table, replace what's in it, or refuse if it already has data.
List projectsThe Google Cloud projects the connected account can run BigQuery in, each with the project id every other step asks for.
List datasetsThe datasets in a project, with where each one's data lives. A dataset is BigQuery's folder of tables.
List tablesThe tables and views in a dataset, each with the full project.dataset.table name a query needs, when it was created, and which column it is split by day on.
Get table detailsEverything about one table: its columns and their types, how many rows it holds, how big it is, where it lives and when it last changed. The step to put in front of a query when the columns aren't known yet.
Preview rowsRead rows straight out of a table without running a query. This reads storage rather than scanning data, so it costs nothing - reach for it instead of SELECT * when all you want is a look at what's in there. Pick the columns you want and page through a large table a block at a time.
Add a rowAdd one row to a table. Pick the table and its columns appear as fields to fill in, so there is nothing to type from memory. The row is queryable within seconds.
Add rowsAdd many rows to a table at once - usually the rows an earlier step produced. All of them land or none of them do, so a partly-written batch never has to be cleaned up.
Create a datasetCreate an empty dataset - BigQuery's folder of tables - for tables to live in. Where it lives is fixed from the moment it is made and a query can't read across two locations, so the location field offers the places your own data already is, with how many datasets are in each, rather than a scroll through forty-odd regions.
Create a tableCreate an empty table with the columns you describe, ready for rows to be added to it. Optionally split it by day on a date column, so a query filtered to a few days reads only those days and costs a fraction as much.
Copy a tableCopy a whole table, rows and all, to another name or another dataset. Copying doesn't scan the data, so it costs nothing to run - the usual way to take a snapshot before something rewrites a table.
Delete a tableDelete a table and every row in it. For cleaning up the scratch tables a workflow made for itself.
Delete a datasetDelete a dataset. One that still holds tables is left alone unless you deliberately say to delete those too, so a mistyped name can't take a folder of tables with it.
