Coming from a different ETL tool, I’m trying to figure out how to get (production) statistics on each component as it runs in SSIS.
For example, if the flat file is reading from an external source that has a high deviation (the rows/sec changes drastically at different times), I would like to know that information.
If an SSIS has a significant ‘slow point’ (buffer filling up / data stream impacted), I would also like to know that information.
And using sprocs for example from the DMV’s, the CPU time and readIO/writeIO would also be ideal (and useful for people showing improvement by moving from sproc to SSIS in a consistent/measurable approach).
The reason I’m asking this question is I see the rows going through BIDS during debugging, but it may not reflect the actual rows/sec on each component in production.
How would one either enable/introspect/obtain these kinds of statistics for production environments (even if it takes a small hit, the numbers are a big deal).
Thanks!
-Darren
This is difficult to do in SSIS 2005. I have seen the runtime engine “just stop” when trying to perform task-level logging from event handlers in complex SSIS packages. One thought: to instrument the Data Flows only by adding Row Count Transformations just after Source Adapters and on each Data Flow Path that outputs rows. Then add an Execute SQL Task to each Data Flow Task’s OnPreExecute event handler to log the start of execution, and add another Execute SQL Task to the corresponding OnPostExecute event handler. In the onPostExecute logic, store the row counts and the end time of the data flow task execution. I believe that will provide enough metrics to calculate throughput for the data flow pipeline.
Hope this helps,
Andy