On Mon, Apr 25, 2016 at 08:09:54PM +0200, Ilya Dryomov wrote: > On Mon, Apr 25, 2016 at 7:47 PM, Stefan Lissmats <stefan@xxxxxxxxxx> wrote: > > Hello again! > > > > I understand that it's not recommended running osd and rbd-nbd on the same host and i actually moved my rbd-nbd to a completely clean host (same kernel and OS though), but with same result. > > > > I hope someone can resolve this and you seem to indicate it is some kind of known error but i didn't really understand the github commit that you linked. > > Yes, it is a bug. rbd-nbd code expects writes to have rval (return > code) equal to the size of the write. I'm pretty sure that's wrong, > because rval for writes should be 0 or a negative error. > > I think what happens is your writes complete successfully, but rbd-nbd > then throws an -EIO to the kernel because 0 != write size. I could be > wrong, so let's wait for Mykola to chime in - he added that check to > fix discards. Sorry for delay (I missed this thread due to a wrong filter). I don't recall details but I think I had an impression that on success aio_write completion returned the number of bytes written. I might be confused by this test that checks for r >= 0: https://github.com/ceph/ceph/blob/master/src/test/librbd/test_librbd.cc#L1254 Now, looking at it again, it is certainly not true and my patch is wrong. I see the fix is already requested: https://github.com/ceph/ceph/pull/8775/ Thanks. -- Mykola Golub _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |