Originally, the client <-> mds messages contained the full path of the files written, similar to cifs. This was a bit racy and problematic, but generally with a little modifications to the client code and to the protocol you'd be able to build and send that full path through with the osd I/O operations (again, I'm not sure what can of worms would be opened with the full-path, or whether it's actually feasible nowadays). The next thing you'd need is to modify the osd operations handling, and mirror each write operation to your ext3 partition. I think that'd be the easiest path to go, though somewhat hacky. Yehuda On Tue, Nov 2, 2010 at 7:53 PM, Jeff Wu <cpwu@xxxxxxxxxxxxx> wrote: > Hi Yehuda, > > Thank you for your reply so quickly,your idea is very useful for us. > our design is that it only use ceph as our file system, > don't use NFS or CIFS . i will continue to study ceph codes ,thanks. > > Jeff.wu > > > > > 在 2010-11-03三的 10:39 +0800，Yehuda Sadeh Weinraub写道： >> Hi, >> >> there are probably a few directions that you can go, but I'm not >> sure which would be the easiest or whether it'd actually make sense. >> Just to have an idea about what you're really looking for, why >> wouldn't doing nfs mount over secret/ work? Is there a real need for >> that metadata to go through the mds? >> >> Yehuda >> >> On Tue, Nov 2, 2010 at 7:18 PM, Jeff Wu <cpwu@xxxxxxxxxxxxx> wrote: >> > hi, >> > >> > Any ideas or suggestions about this? or , To solve this issue,i need >> > deeply look into what parts ceph codes ,OSD ,CRUSH ,.OSD cluster >> > expansion .. etc ? Thanks. >> > >> > >> > Jeff.Wu >> > >> > >> > ---------- Forwarded message ---------- >> > From: Jeff Wu <cpwu@xxxxxxxxxxxxx> >> > To: "sage@xxxxxxxxxxxx" <sage@xxxxxxxxxxxx>, "ceph-devel@xxxxxxxxxxxxxxx" <ceph-devel@xxxxxxxxxxxxxxx> >> > Date: Tue, 2 Nov 2010 14:49:58 +0800 >> > Subject: Questions about ceph OSD >> > Hi , >> > >> > I have recently been working the preliminary research whether ceph >> > can use at our cloud computing storage system in the future , >> > ceph is a very very excellent file system.But ,Now ,i hit a very >> > important problem for us: >> > >> > Normally, as ceph described , all of the data are saved at OSDs with >> > CRUSHMAP at ceph system. >> > >> > but my goal is to : >> > 1. add a ext3 disk to ceph system ,but doesn't set it as a OSD; >> > 2. save the normal data at the OSDs of a ceph system; >> > 3. save the special data to a local ext3 disk at a ceph system, >> > the ext3 disk is not set as a ceph OSD,but its infos are added to >> > MDS and MON,when Write the special data at a ceph client ,these >> > special data are not saved to ceph server OSDs ,but saved at the local >> > ext3 disk. >> > >> > like this: >> > At ceph client: >> > $ mount.ceph 1.2.3.4:6789:/ /ceph >> > $ cd /ceph >> > $ ls >> > cloud user secret >> > >> > when do writing,"cloud and user" folder data are saved to ceph OSDs , >> > but "secret" folder data are saved to a local ext3 disk . >> > >> > My question is: >> > Q1- If i add a new OSD to the ceph system ,but i do a special CRUSH >> > placement map . Could this ways solved this problem ?,If can ,How ? >> > >> > Q2 - If to implement this function as a patch myself, Could you give me >> > any ideas ? or what parts of ceph codes(server/client) that i need look >> > into ? >> > >> > Thanks for all help and reply . >> > >> > >> > Jeff.Wu >> > Transoft.inc ,Sr.software engineer. >> > >> > >> > >> > >> > >> > -- >> > 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 >> > >> > > > > -- > 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 > -- 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
