有什么优雅的方法可以取消/停止长时间运行的线程？

Question

I know about CancellationToken support in TPL, but (IMO) that could work only if I have appropriate token.ThrowifCancellationRequested(), which is not possible in my scenario.

Is there any other elegant way (other than aborting the thread altogether) to achieve this?

I do not care about the transactional state of the thread. I do not care at what state the thread is stopped/cancelled.

Thanks for your interest!

More Details: I need to spawn a new thread (Task/Threadpool/Thread) to call an API which calls multiple stored procedures to fetch near 1million records, and then it does some decoration (business calculation) on each of the records. In the end, this API returns a data table which is a business significant report. I want to give the users the option to cancel this processing mid-way, if needed. For instance, if wrong of set of filters are set. The API only does the read-only operations, so no transaction management is required.

Answer 1

To do this properly, you should use built-up methods in a Task class via a cancellation token, or use cancel a background worker. Or, you may try to use a volatile bool to check if a user request has been cancelled. However, volatile bools aren't behaving consistently across different architectures (single/multi processor). Alternatively you may use *ResetEvent s to indication that cancel was called. This all needs to be done in the API itself, otherwise how would it know when to check for a cancellation condition? You won't be able to cancel it gracefully.

Of course, do not ever call Abort on a thread - that's evil and may leave the whole application in an undefined state, unless you plan to restart whole application, don't call it.

Bonus suggestion

Also consider doing all the processing work on the database side. Do not bring more than a very small amount of records to the client side, say 1k is a good practical threshold. Consider a situation where many users request different reports that all load a server that much - this doesn't scale well. Moreover servers are very efficient with data processing and they have data at theirs side - lots of optimisation is done (CPU line caching, fast memory access etc.).