On 10/24/2016 08:54 PM, Theodore Ts'o wrote:
On Mon, Oct 24, 2016 at 10:28:14AM -0600, Jens Axboe wrote:
How about the below? Bump the timeout to 5 min, 1 min is a little on the
short side, we want normal error handling to be out of the way before
that happens. And additionally, break out if we have been marked as
reaped/exited, so we avoid grabbing the stat mutex again.
Yep, that works. I tried a test with just the second change:
+ /*
+ * If we took too long to shut down, the main thread could
+ * already consider us reaped/exited. If that happens, break
+ * out and clean up.
+ */
+ if (td->runstate >= TD_EXITED)
+ break;
+
And that's sufficient to solve the problem.
Yes, it should be, so glad that it is!
Increasing the timeout to 5 minute also would be a good idea, so we
can let the worker threads exit cleanly so the reported stats will be
completely accurate.
I made that separate change as well. If the job is stuck in the kernel
for some sync operation, we could feasibly be uninterruptible for
minutes. So 1 minutes is too short in any case, and I'd rather just make
this check than sending kill signals since it won't fix the
uninterruptible problem.
Thanks for your help in figuring out this long-standing problem!
It was easy based on all your info, since I could not reproduce. So
thanks for your help! Everything should be committed now, and I'll cut a
new release tomorrow so we can hopefully put this behind us.
--
Jens Axboe
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
