We have a web service that acccepts an XML file for any faults that occur on a vehicle. The web service then uses EF 3.5 to load these files to a hyper normalized database. Typically an XML file is processed in 10-20 seconds. There are two concurrency scenarios that I need to handle:
Different vehicles sending XML files at the same time: This isn't a problem. EF's default optimistic concurrency ensures that I am able to store all these files in the same tables as their data is mutually exclusive.
Same vehicle sending multiple files at the same time: This creates a problem as my system tries to write same or similar data to the database simultaneously. And this isn't rare.
We needed a solution for point 2.
To solve this I introduced a lock table. Basically, I insert a concatenated vehicle id and fault timestamp (which is same for the multiple files sent by a vehicle for the same fault) into this table when I start writing to the DB and I delete the record once I am done. However, there are a lot of times when both the files try to insert this row into the database simultaneously. In such cases, one file succeeds, while the other throws a duplicate key exception that goes to the caller of the webservice.
What's the best way to handle such scenarios? I wouldn't like to rollback anything from the db as there are many tables involved for a single file.
And what solution do you expect? Your current approach with lock table is exactly what you need. If the exception is fired because of duplicate you can either wait and try it again later or fire typed fault back to client and let him upload the file later. Both solutions are ugly but that is what your application currently offer.
The better solution would be replacing current web service with another solution where web service call would only add job to the queue and some background process would process these jobs and ensure that two files for the same car would not be processed concurrently. This would also offer much better throughput control for peek situations. The disadvantage is that you must implement some notification that file has been processed because it will not be online.