On Wed, Apr 19, 2017 at 04:04:46PM +0200, Carlos Maiolino wrote: > If I am not blind, the task currently owning the cpu isn't listed there, which I > suppose to be already running. > > Next time you hit it, could you please paste the stack trace of the task owning > the cpu? Which I suppose to be xfsaild according to your report. > > A simple `cat /proc/<pid>/stack` is probably enough > > thanks > Btw, I'm not very good at xfs journaling code, but I wonder by the stack traces you provided, if you don't have a too small journal, which, with your workload, it might be getting to flush the logs too many times, making the disk too busy and slowing down the rest of the IO. Another folks with more knowledge of the journaling code might be able to confirm or not this, but having too many tasks waiting in: xlog_grant_head_wait() actually made me wonder if your journal isn't too small for the FS size and the current workload you have. Again, this is just a guess, but maybe somebody else can put one more pair of eyes on it. cheers -- Carlos -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
