Passthrough functions for Mode
Early Access

If the resulting query includes an aggregation from a table different to the one defined in the passthrough function, it generates an error. The query should be valid if a) the underlying query is defined upon a single table or b) the underlying query involves a single fact table and multiple dimension tables.

Aggregate formulas are not directly available in the formula editor. It is possible to wrap an aggregate formula in a group_aggregate function. This will allow for the passthrough function to be saved. However, this will likely generate an error.

The passthrough function name is based upon the expected return data type. For example, if the expected result is a STRING value, then the correct passthrough function is sql_string_*. However, a user could write the ThoughtSpot formula referencing a function with a different return data type, for example, sql_number_*. This would likely result in an error.
For each data type, the user can select an aggregate or non-aggregate passthrough function. The selection is dependent upon the underlying function the user is referencing, for example, rank() is an aggregate function while concat() is a non-aggregate function.

The user must ensure that the strings are correctly encapsulated. In the scenario where the statement includes double quotes, then the passthrough function should be wrapped in single quotes.
ThoughtSpot generates GROUP BY statements for all queries. The user must ensure that aggregate columns include the correct aggregation.
For example, the first definition below will return a result because the parameter is wrapped in a sum(). However, the second definition below will result in an error.

If a filter is applied to a passthrough function defined with a window aggregate, then the query will result in a HAVING clause, which generates an error. Filtering window aggregates require that the query is written with either a QUALIFY clause (if supported by the underlying data platform), or the entire query is rewritten in the form of a common table expression (CTE). Note that this is the format generated by ThoughtSpot’s native window functions.

If a column is defined as a PARTITION, then this column must be included in search data. However, this requirement is not explicitly stated.
If a column is defined as a PARTITION and the user drills down on the partition column, an error results. This is because during drill down, the column is removed from the search and replaced with a filter.

Partitioning requires that the partition date matches that in the search. In the formula example below, the definition assumes that WEEKLY is defined in the search. If the search defines the date as MONTHLY, an error results.
sql_int_aggregate_op ( "sum(count(distinct {0})) over (order by date_trunc('week',{1} ) rows between unbounded preceding and current row)", Stations ID, Dim My Date )

In ThoughtSpot, it is possible to change the behavior of a column that is defined as a measure in search. [Region][Sales] returns an aggregate of sales for each region. The passthrough function for a rank statement for this search would be defined as: sql_double_aggregate_op ("rank() over (order by sum({0}) desc)",Amount ).
However, [Region][by Sales] modifies the behavior of the Sales column to return a non-aggregated row. Therefore in this scenario, you could define the passthrough function for a rank statement as: sql_double_aggregate_op ("rank() over (order by {0} desc)",Amount ).

If the column defined in the window aggregation is a measure, it must be modified to act as an attribute. This is because ThoughtSpot includes GROUP BY statements in the underlying queries. If the search is defined as [Region][State][Sales], the formula results in an error. If the search is defined as [Region][State][by Sales], the formula returns a result.
sql_double_aggregate_op ("sum({0}) over (partition by {1} order by {2} rows between unbounded preceding and current row)",sales,region,state)

If the resulting query is invalid, ThoughtSpot returns a generic message, "Error Loading Data, Query execution failed on Mode database." If the same SQL is executed in the underlying database, then a more meaningful error is often displayed.

Assume that a column stores car sales information as a variant. The following formula will return the string value for the dealership.
fxDealership = sql_string_op ("src:dealership::text")
Assuming that a user wants to count the number of records for each dealership, they could ask a question such as: [count src] by [fxDealership]. The resulting query will fail as fxDealership is not included in the GROUP BY clause. This is because no ThoughtSpot attribute column is being referenced directly in the search. To resolve this and force the attribute column’s alias into the GROUP BY clause, you must wrap the formula in a concat statement. Note that the LEFT function does not return any characters.
fxDealership = concat (sql_string_op ("src:dealership::text"),left(src ,0)).

This may cause issues if the underlying data platform expects the function to respect the string case.