Re: Problem interrupting threads in PR_Accept() wiht PR_Interrupt()

On Mon, Aug 11, 2008 at 2:00 AM, Joachim Ziegler <[email protected]> wrote:
>
> Hello Wan-Teh,
>
> I did exactly what you proposed: Compiled my own version of nspr-4.7.1
> with having blindly (since I do not know what I am really doing)
> substituted every occurrence of
>
> if (self->state & PT_THREAD_ABORTED)
>
> by
>
> if (_PT_THREAD_INTERRUPTED(self))
>
> (This happened 4 times, in lines 459, 520, 601 and 659.)

Yes, these are exactly the lines that you need to change.

> Then I used the approach of issuing PR_BlockInterrupt() after
> PR_Accept() and PR_UnblockInterrupt() before PR_Accept().
>
> But my worker threads are still interrupted during a PR_TransmitFile():
>
> [...]
>        [17504] Thread 4 got request: GET /www/hugefile.msi HTTP/1.1
> MASTER[17504]: Sending interrupt signal to worker threads.
> MASTER[17504]: Thread 0 is blocked. Sent PR_Interrupt.
> MASTER[17504]: Thread 1 is blocked. Sent PR_Interrupt.
> MASTER[17504]: Thread 2 is blocked. Sent PR_Interrupt.
> MASTER[17504]: Thread 3 is blocked. Sent PR_Interrupt.
> MASTER[17504]: Thread 4 is blocked. Sent PR_Interrupt.
> MASTER[17504]: Joining worker threads.
> oops: PR_TransmitFile in do_cat: -5993: Operation interrupted by another
> thread
>        [17504] Thread 4 is shutting down.
> oops: PR_TransmitFile in do_cat: -5993: Operation interrupted by another
> thread
>        [17504] Thread 2 is shutting down.
> oops: PR_TransmitFile in do_cat: -5993: Operation interrupted by another
> thread
>        [17504] Thread 0 is shutting down.
> [...]

Hi Joachim,

Could you file a bug report about this problem?  I'm a little
busy this week.  I don't want to lose track of this problem
unintentionally.

On what platform do you observe this problem?  We have
several implementations of PR_TransmitFile, so it's
important to know which implementation you're using.

I'm going to assume you're using Linux.  You can give
the following a try: in pt_linux_sendfile_cont at line 1121,
add a test for EINTR similar to the test in pt_linux_sendfile_cont
at line 1070.  I doubt this is the problem though.

Thanks,
Wan-Teh