Skip to content

Latest commit

 

History

History
145 lines (78 loc) · 6.3 KB

FAQ.md

File metadata and controls

145 lines (78 loc) · 6.3 KB
Raw

U-SQL Frequently Asked Questions

This document lives at http://aka.ms/usqlfaq.

Table of Content

##Extractor Questions

Q: Using the built-in U-SQL Extractors is there any way to record the input lines that are invalid so that they can be analyzed later?

You cannot capture errors with the built-in extractors.

As a workaround, you can build your own extractor that uses a try catch and other checks and then add a column that stores the error and the input.*

NOTE: because you may have a large amount in information in the debug column, you can use byte[] or string. Note that strings have a maximum size of 128KB.

Q: I am trying to load data I exported from SQL Server, but when I try to extract from the generated CSV file, I get encoding or invalid character errors. What do I need to do when exporting data from SQL Server to load into Azure Data Lake to process with U-SQL?

First, you probably should look into using Azure Data Factory that has a lot of data movement options, including moving data from SQL Server into Azure Data Lake Storage and vice versa.

If you want to export the data from SQL Server yourself, you should make sure that you get the right encoding. U-SQL's built-in extractors only support the following encodings: ASCII (7-bit), UTF-7, UTF-8, UTF-16 (aka Unicode) and UTF-32. So make sure you specify -w when using bcp out to generate the output in UTF-16 encoding. Then use the parameter encoding:System.Encoding.Unicode when extracting from the file. Or alternatively, transform the UTF-16 file into UTF-8 before loading it into the Azure Data Lake Storage account (see https://azure.microsoft.com/en-us/documentation/articles/sql-data-warehouse-load-polybase-guide/ for some examples on how to transform the data).

User-defined C# code: Inline C#, User-defined Functions, User-defined Operator Questions

Q: Can I read/write files stored in ADLS or WASB using code that is running within a U-SQL User-defined Operator (UDO)?

No. This is disallowed.

Q: Can my U-SQL script make a network call to some other machine?

No. This is explicitly disallowed and unsupported.

For a specific scenario, please file a request at http://aka.ms/adlfeedback.

General U-SQL Questions

Q: Can I get a single value back from a rowset?

You may be trying to do something like

@MaxSize = SELECT MAX(Size) AS MaxSize FROM input;

or

SELECT @MaxSize=MAX(Size) AS MaxSize FROM input;

and expect @MaxDate to be a single scalar value.

U-SQL does not support retrieving a single scalar value from a rowset. You can only retrieve another rowset.

The second query above will result in a syntax error, while @MaxSize in the first example is not a value of type long, but instead is a rowset with a single row and a single column of type long called @MaxSize.

Q: How can I do something like Excel's AVERAGEIF( )?

Use C# expression syntax, for example if the column you want (called foo) is of type int then use

@rows = SELECT AVG(foo > 0 ? (int?)foo : (int?)null) AS AvgFoo

Q: How can a create an "empty" rowset - a rowset with zero rows?

You have several options:

Option 1: Create an empty table that you refer to when you need it in subsequent scripts. Note that since you are not inserting any data, you will not need to specify an index:

CREATE TABLE Empty(a int, b int, c int); // Note: you still need to provide a schema

Then you can say:

@empty = SELECT * FROM Empty;

Option 2: Create an empty rowset by reading from an empty file set:

@empty = EXTRACT a int, b int, c int 
         FROM "{*}.doesnotexist"
         USING Extractors.Csv();

Option 3: Create an empty rowset by an always false predicate:

@empty = SELECT * FROM (VALUES(1,2,3)) AS T(a,b,c) WHERE a != 1;

Q: How can I test if a stream exists in a U-SQL script?

There is no way to do this from within a U-SQL Script.

Q: How can I do a correlated subquery like in SQL?

U-SQL doesn't have correlated subqueries.

Most scenarios for correlated subquery can be expressed via a JOIN. Use that technique instead.

Q: How can I Assign Unique IDs to rows?

Use Window Functions.

@rs1 = SELECT *, ROW_NUMBER() OVER ( ) AS unique_id 
     FROM @unique_id;

Do not use Guid.NewGuid() to create unique ids for rows since the Guid creation is not deterministic and can lead to job aborts if a node is getting retried.

Q: How can get a random sampling of rows from a rowset?

U-SQL doesn't have a real random sampling method.

However if you only care about getting a subset of rows and don't really care about truwe random sampling then you can use the ROW_NUMBER Window Function as shown below where only ever one out of thousand rows are returned.

@rs1 = SELECT *, ROW_NUMBER() OVER ( ) AS rownum 
     FROM @rs0;
     
@rs1 = SELECT * FROM @rs1
     WHERE ( (rownum % 1000) == 0 );

Q: Can I have a different schema for each row in a a U-SQL rowset?

No. A U-SQL RowSet must contains rows and each row must have the same schema: same column names, with the same types, in the same order. Consider using a SQL.MAP typed column for semistructured data.

If you want to union two rowsets with different schemas, null padd the missing columns in the SELECT clauses.

Q: How can I Extract all the files in a folder?

You can do this by using the FileSet syntax.

@rows =
    EXTRACT Name string, ID int
    FROM "/dimensiondata/carrier/{*}"
    USING Extractors.Csv();

U-SQL Catalog Questions

Q: Is there an API to read or write a U-SQL table locally?

There is an API to read meta data object description that is being exposed in the REST APIs, the SDKs, the Portal and the ADL Tools in VisualStudio.

The only supported way of creating a catalog object such as a table is to run a U-SQL script. The ADL Tools provide a UI to help you create the script.

There is currently no API to preview a table. Please submit a script that extracts the requested data, write it to a file and preview the file.