Troubleshooting Package Execution
Integration Services includes features and tools that you can use to troubleshoot packages when you execute them after they have been completed and deployed.
At design time, Business Intelligence Development Studio provides breakpoints to pause package execution, the Progress window, and data viewers to watch your data as it passes through the data flow. However, these features are not available when you are running packages that have been deployed. The main techniques for troubleshooting deployed packages are the following:
Catch and handle package errors by using event handlers.
Capture bad data by using error outputs.
Track the steps of package execution by using logging.
You can also use the following tips and techniques to avoid problems with running packages
You can respond to the many events that are raised by the package and the objects in the package by using event handlers.
Create an event handler for the OnError event. In the event handler, you can use a Send Mail task to notify an administrator of the failure, use a Script task and custom logic to obtain system information for troubleshooting, or clean up temporary resources or incomplete output. For more information, see Integration Services Event Handlers and Creating Package Event Handlers.
You can use the error output available on many data flow components to direct rows that contain errors to a separate destination for later analysis.
Capture bad data by using error outputs. Send rows that contain errors to a separate destination such as an error table or a text file. The error output automatically adds two numeric columns that contain the number of the error that caused the row to be rejected, and the ID of the column in which the error occurred. For more information, see Handling Errors in Data and How to: Configure an Error Output in a Data Flow Component.
Add friendly information to the error outputs. You can make it easier to analyze the error output by adding descriptive information in addition to the two numeric identifiers that are supplied by the error output.
Add the description of the error. It is easy to look up the error description by using a Script component. For more information, see Enhancing an Error Output with the Script Component.
Add the name of the error column. Looking up the column name that corresponds to the column ID saved by the error output cannot easily be done in the Script component and requires additional steps. Each column ID in a data flow is unique within that Data Flow task, and is persisted in the package at design time. The following approach is one suggestion for adding the column name to the error output:
Create a lookup table of column names. Create a separate application that uses the Integration Services API to iterate over each saved package, each data flow in the package, each object in the data flow, and each input and output in the data flow object. The application should persist the column ID and name of each column to a lookup table, along with the ID of the parent Data Flow task and the ID of the package.
Add the column name to the output. Add a Lookup transformation to the error output that looks up the column name in the lookup table created in the preceding step. The lookup can use the column ID in the error output, the package ID (available in the system variable System::PackageID), and the ID of the Data Flow task (available in the system variable System::TaskID).
You can track much of what occurs in your running packages by enabling logging. Log providers capture information about the specified events for later analysis, and save that information in a database table, a flat file, an XML file, or another supported output format.
Enable logging. You can refine the logging output by selecting only the events and only the items of information that you want to capture. For more information, see Integration Services Log Providers and Implementing Logging in Packages.
Select the package's Diagnostic event to troubleshoot provider issues. There are logging messages that help you troubleshoot a package's interaction with external data sources. For more information, see the section, Troubleshooting Issues with External Data Providers, in this topic.
Enhance the default logging output. Logging typically appends rows to the logging destination each time that a package is run. Although each row of logging output identifies the package by its name and unique identifier, and also identifies the execution of the package by a unique ExecutionID, the large quantity of logging output in a single list can become difficult to analyze.
The following approach is one suggestion for enhancing the default logging output and making it easier to generate reports:
Create a parent table that logs each execution of a package. This parent table has only a single row for each execution of a package, and uses the ExecutionID to link to the child records in the Integration Services logging table. You can use an Execute SQL task at the beginning of each package to create this new row and to record the start time. Then you can use another Execute SQL task at the end of the package to update the row with the end time, duration, and status.
Add auditing information to the data flow. You can use the Audit transformation to add information to rows in the data flow about the package execution that created or modified each row. The Audit transformation makes nine pieces of information available, including the PackageName and ExecutionInstanceGUID. For more information, see Audit Transformation. If you have custom information that you would also like to include in each row for auditing purposes, you can add this information to rows in the data flow by using a Derived Column transformation. For more information, see Derived Column Transformation.
Consider capturing row count data. Consider creating a separate table for row count information, where each instance of package execution is identified by its ExecutionID. Use the Row Count transformation to save the row count into a series of variables at critical points in the data flow. After the data flow ends, use an Execute SQL task to insert the series of values into a row in the table for later analysis and reporting.
For more information about this approach, see the section, "ETL Auditing and Logging," in the Microsoft white paper, Project REAL: Business Intelligence ETL Design Practices.
In Integration Services, you can create debug dump files that provide information about the execution of a package. For more information, see Working with Debug Dump Files.
Sometimes you might not be able to connect to your data sources, or portions of your package cannot be validated, until prior tasks in the package have executed. Integration Services includes the following features to help you avoid the validation errors that would otherwise result from these conditions:
Configure the DelayValidation property on package elements that are not valid when the package is loaded. You can set DelayValidation to True on package elements whose configuration is not valid, to prevent validation errors when the package is loaded. For example, you may have a Data Flow task that uses a destination table that does not exist until an Execute SQL task creates the table at run time. The DelayValidation property can be enabled at the package level, or at the level of the individual tasks and containers that the package includes.
The DelayValidation property can be set on a Data Flow task, but not on individual data flow components. You can achieve a similar effect by setting the ValidateExternalMetadata property of individual data flow components to false. However, when the value of this property is false, the component is not aware of changes to the metadata of external data sources. When set to true, the ValidateExternalMetadata property can help to avoid blocking issues caused by locking in the database, especially when the package is using transactions.
If you encounter errors when trying to run deployed packages by using SQL Server Agent, the accounts used by Agent might not have the required permissions. For information on how to troubleshoot packages that are run from SQL Server Agent jobs, see An SSIS package does not run when you call the SSIS package from a SQL Server Agent job step. For more information on how to run packages from SQL Server Agent jobs, see Scheduling Package Execution in SQL Server Agent and How to: Run a Package.
To connect to Excel or Access data sources, SQL Server Agent requires an account that has permission to read, write, create, and delete temporary files in the folder that is specified by the TEMP and TMP environment variables.
When you encounter errors on a 64-bit server with packages that run successfully in 32-bit mode or on 32-bit servers, consider the following common issues. For more information, see 64-bit Considerations for Integration Services.
Some data providers are not available on the 64-bit platform. In particular, the Microsoft Jet OLE DB Provider that is required to connect to Excel or Access data sources is not available in a 64-bit version.
DTS packages cannot run in 64-bit mode. If your package uses the Execute DTS 2000 Package task to run SQL Server 2000 Data Transformation Services (DTS) packages, you must run the package in 32-bit mode. There is no 64-bit run-time support for DTS packages.
If you encounter an Integration Services error that does not have an accompanying description, you can locate the description in Integration Services Error and Message Reference by looking up the error by its number. The list does not include troubleshooting information at this time.