父子程序包中的事务控制

Question

I have parent package in which i have foreach loop for counting flat files and within it i process each file one by one while calling(Exec package task) child package where i have sequence container. While in child package sequence container i have two flow tasks -one is for deleting data from import schema tables and import all data from flat file into it which is passed by parent package. -second is moving new and updated rows from import schema to the relative staging schema tables.

Now i set "required" transaction property for parent control flow while all other components in it are "supported". and set "supported" transaction property for child control flow and same for all components in it. So when i execute the parent package it stuck at second flow task in the child package.

I want to rollback all transactions if any failure or error occur in any pacakge.

Please any help will be great.

Answer 1

If SSIS transactions are not working for you, consider re-architecting the package design. Is it possible to break up the steps and stage data multiple times before finally committing it to the end result table.

I honestly never use transactions. For an ETL process I break it out into 4 separate areas of concern:

1. Extract - Gets data as-is from source and physically stages it in a table. The staging table will look exactly like the source table. This minimizes interruption with locking in the source system.
2. Cleanse - Cleanses the data in place without consideration to other tables. Cleansing can be performed in a view or with a separate package which stages the data.
3. Conform - Puts data from multiple sources into one table. This table will look like target table (ie fact, dimension)
4. Load - Loads into the end result table, is only concerned with loading to minimize interruption to access of the table to end users.

I build in restartability to the packages so that if they have made it through the first few steps, they do not need to start over. They can pick up where they left off and finish the ETL process.

If I had multiple files to extract from, I would build a package for each. If there were a series of files with the same schema that needed to be looped through. I would build the process with restartability in mind. That is:

1. Truncate the staging table
2. Extract into the staging table one file at a time or in parallel
3. Archive the files

If the process were to fail, in this case, we can start from the top. Because we are dumping the data into a staging table (which can be a heap) this can be a super fast efficient process.

If your main concern is that you do not want a particular end-result table in an inconsistent state, by staging the data, you can easily wrap the final insert/update process into a SQL transaction.

This is pretty general advice, but the gist of it is, don't use transactions, break the process into steps and use SQL transactions where it is actually needed. Hope this helps.