Hi Kefu, I think it was also 65536 (currently it's 4194304). My server has 72 cores. I change it back and this error happens again. When debug into this issue, I saw line[1] returns EAGAIN, that's why I have the above conclusion. [1] https://github.com/scylladb/seastar/blob/master/core/linux-aio.cc#L53 On Wed, Aug 22, 2018 at 4:23 PM kefu chai <tchaikov@xxxxxxxxx> wrote: > > On Mon, Aug 20, 2018 at 10:23 AM Yingxin Cheng <yingxincheng@xxxxxxxxx> wrote: > > > > Hi chunmei & kefu: > > > > Sorry, seems it is another issue I have met before. And I think this > > is not a bug, just try to increase the value in > > /proc/sys/fs/aio-max-nr. > > > > Reference: > > http://man7.org/linux/man-pages/man2/io_setup.2.html > > quote "EAGAIN The specified nr_events exceeds the user's limit of > > available events, as defined in /proc/sys/fs/aio-max-nr." > > sorry Yingxin, i missed your analysis and solution. what's your > fs.aio-max-nr before you increased it? > > all of our arm64 testbed have following output: > > $ sysctl fs.aio-max-nr > fs.aio-max-nr = 65536 > > and max_aio is 128, so i guess it's unlikely we exceed the limit. am i right? > > @Liu, Chunmei what's your /proc/sys/fs/aio-max-nr setting ? > > > > > --- > > Regards, > > Yingxin > > > > On Mon, Aug 20, 2018 at 9:54 AM Yingxin Cheng <yingxincheng@xxxxxxxxx> wrote: > > > > > > hi chunmei, > > > There's a similar report at > > > https://github.com/ceph/ceph/pull/23539#discussion_r209479903 > > > I believe it should already be fixed recently. > > > Does your code include this change? > > > > > > kefu, > > > According to my tests, I see there are no issues with > > > unittest_seastar_messenger. Is it ready to be included in the make > > > check? > > > > > > --- > > > Regards, > > > Yingxin > > > > -- > Regards > Kefu Chai -- Regards, Yingxin
