Wow, thanks for elevating this issue to my attention. This is definitely a bug that was introduced accidentally during all the internal refactoring work in support the new journaling feature. I will get a fix up for review today against the master branch and we can hopefully include it in the next Jewel point release as well. Thanks, Jason On Wed, Aug 17, 2016 at 11:03 AM, Mark Korondi <korondi.mark@xxxxxxxxx> wrote: > Hi all, > > my team struggles with enabling rbd caching with Jewel clusters. In > Hammer, we successfully use rbd caching with the `writethrough until > flush` directive set to true and we get expected IOPS numbers. > However, with the same configuration using Jewel, IOPS drops; after > analyzing logs it seems to us that the writeback caching mechanism > never gets enabled regardless of flush request is being sent. > > We found a bug entry describing an issue similar to ours: > http://tracker.ceph.com/issues/16654 > > As our note #3 says, we concluded that it seems that the imagecontext > user_flushed() signal is missing from the later version, therefore the > caching mechanism will never be switched to writeback. > > Our questions are these: > 1) is this change intentional and one needs to configure their > environment differently? If yes, how? > 2) documentation states that this workaround is only needed with older > virtio kernel driver too old. With newer kernels, what would be the > possible implications of setting `rbd cache writethrough until flush` > to false? Is my understanding correct that this virtio driver version > what matters is that is *in the VM*? > > Thanks, > Mark -- Jason -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
