Create and alter Azure Storage external tables

Applies to: ✅ Microsoft Fabric ✅ Azure Data Explorer

The commands in this article can be used to create or alter an Azure Storage external table in the database from which the command is executed. An Azure Storage external table references data located in Azure Blob Storage, Azure Data Lake Store Gen1, or Azure Data Lake Store Gen2.

Permissions

To .create requires at least Database User permissions, and to .alter requires at least Table Admin permissions.

To .create-or-alter an external table using managed identity authentication requires AllDatabasesAdmin permissions.

Syntax

(.create | .alter | .create-or-alter) external table TableName (Schema) kind = storage [partition by (Partitions) [pathformat = (PathFormat)]] dataformat = DataFormat (StorageConnectionString [, ...] ) [with (Property [, ...])]

Learn more about syntax conventions.

Parameters

Authentication and authorization

The authentication method to access an external table is based on the connection string provided during its creation, and the permissions required to access the table vary depending on the authentication method.

The following table lists the supported authentication methods for Azure Storage external tables and the permissions needed to read or write to the table.

Partitions formatting

The partitions list is any combination of partition columns, specified using one of the forms shown in the following table.

Path format

The PathFormat parameter allows you to specify the format for the external data folder URI path in addition to partitions. It consists of a sequence of partition elements and text separators. A partition element refers to a partition that is declared in the partition by clause, and the text separator is any text enclosed in quotes. Consecutive partition elements must be set apart using the text separator.

[ StringSeparator ] Partition [ StringSeparator ] [Partition [ StringSeparator ] ...]

To construct the original file path prefix, partition elements are rendered as strings and separated with corresponding text separators. You can use the datetime_pattern macro (datetime_pattern(DateTimeFormat, PartitionName)) to specify the format used for rendering a datetime partition value. The macro adheres to the .NET format specification, and allows format specifiers to be enclosed in curly brackets. For example, the following two formats are equivalent:

• 'year='yyyy'/month='MM
• year={yyyy}/month={MM}

By default, datetime values are rendered using the following formats:

Virtual columns

When data is exported from Spark, partition columns (that are provided to the dataframe writer's partitionBy method) aren't written to data files. This process avoids data duplication because the data is already present in the folder names (for example, column1=<value>/column2=<value>/), and Spark can recognize it upon read.

External tables support reading this data in the form of virtual colums. Virtual columns can be of either type string or datetime, and are specified using the following syntax:

.create external table ExternalTable (EventName:string, Revenue:double)  
kind=storage  
partition by (CustomerName:string, Date:datetime)  
pathformat=("customer=" CustomerName "/date=" datetime_pattern("yyyyMMdd", Date))  
dataformat=parquet
( 
   h@'https://storageaccount.blob.core.windows.net/container1;secretKey'
)

To filter by virtual columns in a query, specify partition names in query predicate:

external_table("ExternalTable")
 | where Date between (datetime(2020-01-01) .. datetime(2020-02-01))
 | where CustomerName in ("John.Doe", "Ivan.Ivanov")

Optional properties

File filtering logic

When querying an external table, performance is improved by filtering out irrelevant external storage files. The process of iterating files and deciding whether a file should be processed is as follows:

1. Build a URI pattern that represents a place where files are found. Initially, the URI pattern equals a connection string provided as part of the external table definition. If there are any partitions defined, they're rendered using PathFormat, then appended to the URI pattern.

2. For all files found under the URI pattern(s) created, check that:

   • Partition values match predicates used in a query.
   • Blob name starts with NamePrefix, if such a property is defined.
   • Blob name ends with FileExtension, if such a property is defined.

Once all the conditions are met, the file is fetched and processed.

Examples

Non-partitioned external table

In the following non-partitioned external table, the files are expected to be placed directly under the container(s) defined:

.create external table ExternalTable (x:long, s:string)  
kind=storage 
dataformat=csv 
( 
   h@'https://storageaccount.blob.core.windows.net/container1;secretKey' 
) 

Partitioned by date

In the following external table partitioned by date, the files are expected to be placed under directories of the default datetime format yyyy/MM/dd:

.create external table ExternalTable (Timestamp:datetime, x:long, s:string) 
kind=storage
partition by (Date:datetime = bin(Timestamp, 1d)) 
dataformat=csv 
( 
   h@'abfss://[email protected]/path;secretKey'
)

Partitioned by month

In the following external table partitioned by month, the directory format is year=yyyy/month=MM:

.create external table ExternalTable (Timestamp:datetime, x:long, s:string) 
kind=storage 
partition by (Month:datetime = startofmonth(Timestamp)) 
pathformat=(datetime_pattern("'year='yyyy'/month='MM", Month)) 
dataformat=csv 
( 
   h@'https://storageaccount.blob.core.windows.net/container1;secretKey' 
) 

Partitioned by name and date

In the following external table, the data is partitioned first by customer name and then by date, meaning that the expected directory structure is, for example, customer_name=Softworks/2019/02/01:

.create external table ExternalTable (Timestamp:datetime, CustomerName:string) 
kind=storage 
partition by (CustomerNamePart:string = CustomerName, Date:datetime = startofday(Timestamp)) 
pathformat=("customer_name=" CustomerNamePart "/" Date)
dataformat=csv 
(  
   h@'https://storageaccount.blob.core.windows.net/container1;secretKey' 
)

Partitioned by hash and date

The following external table is partitioned first by customer name hash (modulo ten), then by date. The expected directory structure is, for example, customer_id=5/dt=20190201, and data file names end with the .txt extension:

.create external table ExternalTable (Timestamp:datetime, CustomerName:string) 
kind=storage 
partition by (CustomerId:long = hash(CustomerName, 10), Date:datetime = startofday(Timestamp)) 
pathformat=("customer_id=" CustomerId "/dt=" datetime_pattern("yyyyMMdd", Date)) 
dataformat=csv 
( 
   h@'https://storageaccount.blob.core.windows.net/container1;secretKey'
)
with (fileExtension = ".txt")

Filter by partition columns in a query

To filter by partition columns in a query, specify original column name in query predicate:

external_table("ExternalTable")
 | where Timestamp between (datetime(2020-01-01) .. datetime(2020-02-01))
 | where CustomerName in ("John.Doe", "Ivan.Ivanov")

Sample Output

Related content

• Export data to an external table.

• Continuous data export to an external table.