I tested it on Hammer and I can recreate what you are seeing. The good news is that Infernalis and later releases behave correctly -- they list the range 1M-4M as dirty. Since Hammer is approaching end-of-life, I wouldn't realistically expect this to be fixed -- but I did open a tracker ticket to document it [1].
[1] http://tracker.ceph.com/issues/19570 On Sun, Apr 9, 2017 at 4:40 AM, 许雪寒 <xuxue...@360.cn> wrote: > Oh, sorry again. We didn't resize the image, just "aio_discard"ed the data > from the offset 1048576 to the end of the rbd image. > ________________________________________ > 发件人: 许雪寒 > 发送时间: 2017年4月9日 16:37 > 收件人: dilla...@redhat.com > Cc: ceph-users@lists.ceph.com > 主题: 答复: [ceph-users] 答复: 答复: rbd expord-diff aren't counting AioTruncate op > correctly > > Ah, sorry, I didn't understand you correctyly. We did use the > librbd::Image::aio_discard method to resize the image from 4MB to 1MB. > ________________________________________ > 发件人: ceph-users [ceph-users-boun...@lists.ceph.com] 代表 许雪寒 [xuxue...@360.cn] > 发送时间: 2017年4月3日 23:27 > 收件人: dilla...@redhat.com > Cc: ceph-users@lists.ceph.com > 主题: [ceph-users] 答复: 答复: rbd expord-diff aren't counting AioTruncate op > correctly > > Hi, actually, in our test, we created an image with the size 4MB which means > it has only one object, then we write data into it, create a snap, clone it > and create a snap of the new image, say "snap1", and then we send a > "AioTruncate" to truncate the image to 1MB and create the second snap of the > new image, say "snap2". Then, we use export-diff to calculate the "diff" > between snap1 and snap2. By examing the result of export-diff, we confirmed > that only the data that left in object are recorded into the export-diff's > result, the truncate operation is lost. > > By the way, according to our understanding of the source code, it seems that > the reason for this is that when doing copy-up, no clone is created even if > there has already been some snapshot of the new image. So, if there's only > one "WRITE" op is sent to the object, which leads to a "copy-up", before a > new snapshot is created after which a export-diff is conducted, the > export-diff will copy all the data in the HEAD object, which, in our case, is > not the "diff" that we want. > ________________________________________ > 发件人: Jason Dillaman [jdill...@redhat.com] > 发送时间: 2017年4月3日 23:00 > 收件人: 许雪寒 > Cc: ceph-users@lists.ceph.com > 主题: Re: 答复: [ceph-users] rbd expord-diff aren't counting AioTruncate op > correctly > > Are you referring to the class "AioTruncate"? If so, *how* did you > issue the truncate against a specific object? Did you use the RBD > API's discard method? Did you resize an image? I need a repeatable set > of instructions. > > On Mon, Apr 3, 2017 at 10:58 AM, 许雪寒 <xuxue...@360.cn> wrote: >> Hi, the operation we performed is AioTruncate. >> ________________________________________ >> 发件人: Jason Dillaman [jdill...@redhat.com] >> 发送时间: 2017年4月3日 22:11 >> 收件人: 许雪寒 >> Cc: ceph-users@lists.ceph.com >> 主题: Re: [ceph-users] rbd expord-diff aren't counting AioTruncate op correctly >> >> On Fri, Mar 31, 2017 at 10:36 PM, 许雪寒 <xuxue...@360.cn> wrote: >>> We created an image, write data into it, then protect it and clone a new >>> image. Then we create a snapshot, snap1, of the cloned image, then truncate >>> it and create a snapshot, snap2. In the diff of the two snapshot, we found >>> that only the data that are not truncated is recorded in the diff, which >>> means the truncate operation is not reflected by the diff. >> >> When you say "truncate" above, what specific operation are you >> performing so that I can attempt to recreate it? >> >> >> -- >> Jason > > > > -- > Jason > _______________________________________________ > ceph-users mailing list > ceph-users@lists.ceph.com > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com -- Jason _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
