Power Query is an ETL tool created by Microsoft for data extraction, loading and transformation, and is used to retrieve data from sources, process it, and load them into one or more target systems. Power Query is available in several variations within the Microsoft Power Platform, and is used for business intelligence on fully or partially self-service platforms. It is found in software such as Excel, Power BI, Analysis Services, Dataverse,[1] Power Apps, Azure Data Factory, SSIS, Dynamics 365, and in cloud services such as Microsoft Dataflows,[2] including Power BI Dataflow used with the online Power BI Service or the somewhat more generic version of Microsoft Dataflow used with Power Automate.
ETL is closely related to data modeling,[3] and for transformation, Power Query can be used to develop a logical data model in those cases where the data does not already have one, or where there is a need to further develop the data model.
Power Query was firstly included as an optional additional import feature in Excel besides PowerPivot (used for providing a data model to the PivotTables pivot tables, and more) in Excel 2010 and 2013. In Excel 2016, the function was included in standard Excel and renamed Get & Transform for a short time, but has since been named Power Query again.
Power Query is built on what was then a new query language called M. It is a mashup language (hence the letter M) designed to create queries that mix together data. It is similar to the F Sharp programming language, and according to Microsoft it is a "mostly pure, higher-order, dynamically typed, partially lazy, functional language." The M language is case-sensitive.
Much of the user interaction with Power Query can be done via graphical user interfaces with wizards, and this can be used for many common or basic tasks. It is also possible to use the advanced editing mode where the developer can write in the M formula language; this gives greater expressive power, more possibilities, and can also be used to change the code generated by the graphical wizards.
User queries are typically written with a top level let expression. The let expression contains a list of comma-separated named reference (variables) bindings and an in expression which is what the let expression evaluates too. The in expression can reference the variables and the variables can reference each other. Backwards and forward referencing is allowed, and self-referencing is allowed by prefixing the @ on the variable. Variables are recursively evaluated as needed to evaluate the in expression. No variable is evaluated more than once.
Variables are not typed in Power Query. Instead, an expression can have a type assertion which will evaluate to an error when the expression does not evaluate to a value compatible with the assertion. Assertions can be preceded by nullable to include null in the allowed values.
| number | Assertion for integer and floating-point numbers | |||
| int | Signed 32-bit integer | |||
| long | Signed 64-bit integer | |||
| double | IEEE 754 float | |||
| decimal | 128-bit float. Same as C#'s decimal | |||
| time | Time of day | |||
| date | A Calander date ranging from 1 CE to 9999 CE in the Georgian Calendar | |||
| datetime | A composite of the date and time datatypes | |||
| duration | A measurement of elapsed time (can be negative) | |||
| logical | Represents a Boolean true or false value | |||
| text | A Unicode string | |||
| guid | A Globally Unique Identifier (Converts to a text as needed automatically) | |||
| list | An ordered list of values | |||
| record | An ordered map from text to any value | |||
| table | A 2D matrix where each column has a unique name and type (type not checked on table contents) | |||
| function | A power query function | |||
| type | Represents a datatype and may contain assertion information | |||
| action | An internally used datatype | |||
| null | The null singleton | |||
| any | Represents all values | |||
| anynonnull | Represents all values except null | |||
| none | Represents no values and always fails as an assertion | |||
| error | A pseudo value representing an error |
Power Query supports the C block (/* ... */) and C line (// ...) comments.
In Power BI, use of M-code is somewhat limited in DirectQuery, as opposed to Import which has all capabilities. This is due to the requirement that M-code in DirectQuery has to be translated into SQL at runtime.
Query Folding is the ability for the Power Query steps to be transpiled into a single query at the data source (for example in Transact SQL). As such, Query Folding works like a traditional ETL process, and enables working on the data before loading. Query Folding is not always supported. Steps like filtering, selecting columns and simple SQL arithmetic are supported. Steps like creating index and appending or merging non foldable sources with foldable sources are not. Folding indicators (such as folding, not folding, might fold, opaque, unknown) might indicate up to which step a query might fold. Non-folding queries will have to be performed on the client-side. The order of queries can determine how many of the steps which get folded.