Re: Avoiding Statement.cancel() races

On Fri, Nov 21, 2003 at 06:18:07AM -0500, Kris Jurka wrote:
>
>
> On Fri, 21 Nov 2003, Oliver Jowett wrote:
>
> > However there's an obvious race condition here between starting query
> > execution and cancelling the statement -- it's possible for
> > stopWorkerThread() to run at just the wrong time and "cancel" the statement
> > before it starts execution, having no effect and leaving the worker thread
> > still running a query (at least as I understand cancel() .. that's how it's
> > implemented in the postgresql driver anyway).
>
> What if calling cancel on a non running statement threw an SQLException?
> Then your stopWorkerThread could retry the cancel.

That might work. You'd need a separate SQLException subclass (or SQL state
code) to distinguish "no statement running" from "cancel failed".

I'm not sure if other applications would expect this behaviour, though.
Also, I don't really like spinning on a retry -- you become quite dependent
on the fairness of the JVMs threading/locking model.

-O