Error Configuration for Cube, Partition, and Dimension Processing

Applies to: SQL Server Analysis Services Azure Analysis Services Fabric/Power BI Premium

Error configuration properties on cube, partition, or dimension objects determine how the server responds when data integrity errors occur during processing. Duplicate keys, missing keys, and null values in a key column typically trigger such errors, and while the record causing the error will not be added to the database, you can set properties that determine what happens next. By default, processing stops. However, during cube development, you might want processing to continue when errors occur so that you can test cube behaviors with imported data, even if it is incomplete.

This topic includes the following sections:

Execution order

The server always executes NullProcessing rules before ErrorConfiguration rules for each record. This is important to understand because null processing properties that convert nulls to zeroes can subsequently introduce duplicate key errors when two or more error records have zero in a key column.

Default behaviors

By default, processing stops at the first error implicating a key column. This behavior is controlled by an error limit that specifies zero as the number of allowed errors and the Stop Processing directive that tells the server to stop processing when the error limit is reached.

Records triggering an error, due to null or missing or duplicate values, are either converted to the unknown member or discarded. SQL Server Analysis Services will not import data that violates data integrity constraints.

  • Conversion to unknown member occurs by default, due to the ConvertToUnknown setting for KeyErrorAction. Records allocated to unknown member are quarantined in the database as evidence of a problem that you might want to investigate after processing is finished.

    Unknown members are excluded from query workloads, but they will be visible in some client applications if the UnknownMember is set to Visible.

    If you want to track how many nulls were converted to the unknown member, you can modify the NullKeyConvertedToUnknown property to report these errors to the log or in the Processing window.

  • Discard occurs when you manually set the KeyErrorAction property to DiscardRecord.

Through error configuration properties, you can determine how the server responds when an error occurs. Options include stop processing immediately, continue processing but stop logging, or continue both processing and logging of errors. Defaults vary based on the severity of the error.

The error count keeps track of how many errors occur. When you set an upper limit, server response changes when that limit is reached. By default, the server stops processing after the limit is reached. The default limit is 0, causing processing to stop on the first error that is counted.

High impact errors, such as a missing key or null value in a key field, should be addressed quickly. By default, these errors adhere to ReportAndContinue server behaviors, where the server catches the error, adds it to the error count, and then continues processing (except the error limit is zero, so processing stops immediately).

Other errors are generated but not counted or logged by default (this is the IgnoreError setting) because the error does not necessarily pose a data integrity problem.

Error counts are affected by null processing settings. For dimension attributes, null processing options determine how the server responds when null values are encountered. By default, nulls in a numeric column are converted to zeroes, while nulls in a string column are processed as blank strings. You can override NullProcessing properties to catch null values before they turn into KeyNotFound or KeyDuplicate errors. See Null keys in a dimension for details.

Errors are logged in the Process dialog box but not saved. You can specify a key error log file name to collect errors in a text file.

Error Configuration Properties

There are nine error configuration properties. Five are used to determine server response when a specific error occurs. The other four are scoped to error configuration workloads, such as how many errors to allow, what to do when that limit is reached, whether to collect errors in a log file.

Server response to specific errors

Property Default Other Values
CalculationError

Occurs when initializing error configuration.
IgnoreError neither logs nor counts the error; processing continues as long as the error count is under the maximum limit. ReportAndContinue logs and counts the error.

ReportAndStop reports the error and stops processing immediately, regardless of the error limit.
KeyNotFound

Occurs when a foreign key in a fact table does not have a matching primary key in a related dimension table (for example, a Sales fact table has a record with a product ID that doesn't exist in the Product dimension table). This error can occur during partition processing, or dimension processing of snowflaked dimensions.
ReportAndContinue logs and counts the error. ReportAndStop reports the error and stops processing immediately, regardless of the error limit.

IgnoreError neither logs nor counts the error; processing continues as long as the error count is under the maximum limit. Records that trigger this error are converted to the unknown member by default, but you can change the KeyErrorAction property to discard them instead.
KeyDuplicate

Occurs when duplicate attribute keys are found in a dimension. In most cases, it is acceptable to have duplicate attribute keys, but this error informs you of the duplicates so that you can check the dimension for design flaws that might lead to inconsistent relationships between attributes.
IgnoreError neither logs nor counts the error; processing continues as long as the error count is under the maximum limit. ReportAndContinue logs and counts the error.

ReportAndStop reports the error and stops processing immediately, regardless of the error limit.
NullKeyNotAllowed

Occurs when NullProcessing = Error is set on a dimension attribute or when null values exists in an attribute key column used to uniquely identify a member.
ReportAndContinue logs and counts the error. ReportAndStop reports the error and stops processing immediately, regardless of the error limit.

IgnoreError neither logs nor counts the error; processing continues as long as the error count is under the maximum limit. Records that trigger this error are converted to the unknown member by default, but you can set the KeyErrorAction property to discard them instead.
NullKeyConvertedToUnknown

Occurs when null values are subsequently converted to the unknown member. Setting NullProcessing = ConvertToUnknown on a dimension attribute will trigger this error.
IgnoreError neither logs nor counts the error; processing continues as long as the error count is under the maximum limit. If you consider this error to be informational, keep the default. Otherwise, you can choose ReportAndContinue to report the error to the Processing window and count the error towards the error limit.

ReportAndStop reports the error and stops processing immediately, regardless of the error limit.

General Properties

Property Values
KeyErrorAction This is the action taken by the server when a KeyNotFound error occurs. Valid responses to this error include ConvertToUnknown or DiscardRecord.
KeyErrorLogFile This is a user-defined filename that must have a .log file extension, located in a folder on which the service account has read-write permissions. This log file will only contain errors generated during processing. Use the Flight Recorder if you require more detailed information.
KeyErrorLimit This is the maximum number of data integrity errors that the server will allow before failing the processing. A value of -1 indicates that there is no limit. The default is 0, which means processing stops after the first error occurs. You can also set it to a whole number.
KeyErrorLimitAction This is the action taken by the server when the number of key errors has reached the upper limit. With Stop Processing, processing terminates immediately. With Stop Logging, processing continues but errors are no longer reported or counted.

Where to set Error Configuration properties

Use the property pages in either SQL Server Management Studio after the database is deployed, or in the model project in SQL Server Data Tools. The same properties are found in both tools. You can also set error configuration properties in the msmdrsrv.ini file to change server defaults for error configuration, and in Batch and Process commands if processing runs as a scripted operation.

You can set error configuration on any object that can be processed as a standalone operation.

SQL Server Management Studio

  1. In Object Explorer, rick-click Properties one of these objects: dimension, cube, or partition.

  2. In Properties, click Error Configuration.

SQL Server Data Tools

  1. In Solution Explorer, double-click a dimension or cube. ErrorConfiguration appears in Properties in the pane below.

  2. Alternatively, for a single dimension only, right-click the dimension in Solution Explorer, choose Process, and then choose Change Settings in the Process Dimension dialog box. Error configuration options appear on the Dimension Key Errors tab.

Missing keys (KeyNotFound)

Records with a missing key value cannot be added to the database, not even when errors are ignored or the error limit is unlimited.

The server produces the KeyNotFound error during partition processing, when a table in fact record contains a foreign key value, but the foreign key has no corresponding record in a related dimension table. This error also occurs when processing related or snowflaked dimension tables, where a record in one dimension specifies a foreign key that doesn't exist in the related dimension.

When a KeyNotFound error occurs, the offending record is allocated to the unknown member. This behavior is controlled through the Key Action, set to ConvertToUnknown, so that you can view the allocated records for further investigation.

Null foreign keys in a fact table (KeyNotFound)

By default, a null value in a foreign key column of a fact table is converted to zero. Assuming zero is not a valid foreign key value, the KeyNotFound error will be logged and counted towards the error limit that is zero by default.

To allow processing to continue, you can handle the null before it is converted and checked for errors. To do this, set NullProcessing to Error.

Set NullProcessing property on a measure

  1. In SQL Server Data Tools, in Solution Explorer, double-click the cube to open it in Cube Designer.

  2. Right-click a measure in the Measures pane and choose Properties.

  3. In Properties, expand Source to view NullProcessing property. It is set to Automatic by default, which for OLAP items, converts nulls to zeroes for fields containing numeric data.

  4. Change the value to Error to exclude any records having a null value, avoiding the null-to-numeric (zero) conversion. This modification lets you avoid duplicate key errors related to multiple records having zero in the key column, and also avoid KeyNotFound errors when a zero-valued foreign key has no primary key equivalent in a related dimension table.

Null keys in a dimension

To continue processing when null values are found in foreign keys in a snowflaked dimension, handle the null values first by setting NullProcessing on the KeyColumn of the dimension attribute. This discards or converts the record, before the KeyNotFound error can occur.

You have two options for handling nulls on dimension attribute:

  • Set NullProcessing=UnknownMember to allocate records with null values to the unknown member. This produces the NullKeyConvertedToUnknown error, which is ignored by default.

  • Set NullProcessing=Error to exclude records with null values. This produces the NullKeyNotAllowed error, which is logged and counts toward the key error limit. You can set error configuration property on Null Key Not Allowed to IgnoreError to allow processing to continue.

Nulls can be problem for non-key fields, in that MDX queries return different results depending on whether null is interpreted as zero or empty. For this reason, SQL Server Analysis Services provides null processing options that let you predefine the conversion behavior you want.

Set NullProcessing property on a dimension attribute

  1. In SQL Server Data Tools, in Solution Explorer, double-click the dimension to open it in Dimension Designer.

  2. Right-click an attribute in the Attributes pane and choose Properties.

  3. In Properties, expand KeyColumns to view NullProcessing property. It is set to Automatic by default, which converts nulls to zeroes for fields containing numeric data. Change the value to either Error or UnknownMember.

    This modification removes the underlying conditions that trigger KeyNotFound by either discarding or converting the record before it is checked for errors.

    Depending on error configuration, either of these actions can result in an error that is reported and counted. You might need to adjust additional properties, such as setting KeyNotFound to ReportAndContinue or KeyErrorLimit to a non-zero value, to allow processing to continue when these errors are reported and counted.

Duplicate keys resulting inconsistent relationships (KeyDuplicate)

By default, the presence of a duplicate key does not stop processing, but the error is ignored and the duplicate record is excluded from the database.

To change this behavior, set KeyDuplicate to ReportAndContinue or ReportAndStop to report the error. You can then examine the error to determine potential flaws in dimension design.

Change the error limit or error limit action

You can raise the error limit to allow more errors through during processing. There is no guidance for raising the error limit; the appropriate value will vary depending on your scenario. Error limits are specified as KeyErrorLimit in ErrorConfiguration properties in SQL Server Data Tools, or as Number of Errors in the Error Configuration tab for properties of dimensions, cubes, or measure groups in SQL Server Management Studio.

Once the error limit is reached, you can specify that processing stops or that logging stops. For example, suppose you set the action to StopLogging on an error limit of 100. On the 101st error, processing continues, but errors are no longer logged or counted. Error limit actions are specified as KeyErrorLimitAction in ErrorConfiguration properties in SQL Server Data Tools, or as On error action in the Error Configuration tab for properties of dimensions, cubes, or measure groups in SQL Server Management Studio.

Set the error log path

You can specify a file to store key-related error messages that are reported during processing. By default, errors are visible during interactive processing in the Process window and then discarded when you close the window or session. The log will only contain error information related to keys, identical to the errors you see reported in the processing dialog boxes.

Errors will be logged to a text file and must have .log file extension. The file will be empty unless errors occur. By default, a file will be created in the DATA folder. You can specify another folder as long as the Analysis Services service account can write to that location.

Next step

Decide whether errors will stop processing or be ignored. Remember that only the error is ignored. The record that caused the error is not ignored; it is either discarded or converted to unknown member. Records that violate data integrity rules are never added to the database. By default, processing stops when the first error occurs, but you can change this by raising the error limit. In cube development, it can be useful to relax error configuration rules, allowing processing to continue, so that there is data to test with.

Decide whether to change default null processing behaviors. By default, nulls in a string column are processed as empty values, while nulls in a numeric column are processed as zero.

See Also

Log Properties
Defining the Unknown Member and Null Processing Properties