On Thu, Aug 2, 2012 at 8:22 AM, Sylvain Munaut <s.munaut@xxxxxxxxxxxxxxxxxxxx> wrote: > Hi, > > >> Will need logs in order to say what happened exactly, if that's of any >> interest. You removed some crucial data for the full operation of that >> bucket. It may also be that some of the data was still cached so that >> you were able to complete some operations but not the others. > > In this particular case, some rados object were lost, in particular > the .dir.NNNN.N file that described the bucket. > I tried recreating an empty file in its place but that doesn't seem to > be enough. No, the bucket index needs to be initialized. > > As I said, they're test scenarios on test data. I expect this case to > happen only on bucket that we mapped on low replication pools (size=1 > or 2) and I don't really care if even the entire bucket is lost > (they're used for temporary data transfer between services), but I > need to be able to return to a consistent state. > Because in this case I'm stuck ( can't call create bucket again, it > thinks it's done already. can't call delete on the bucket because it > thinks it doesn't exist ... ). > > That's why I needed to know where every trace of a bucket can be so I > can go and wipe it manually to recreate the bucket from scratch. > > The loss of rados objects related to S3 objects seems easier to fix: > just rm all rados object for it (seems there can be several with > special prefixes) and then remove it from the entries in the > .dir.NNNN.N object. So it should be fairly easy to write a "fix" > utility that just rescans all objects in a bucket and deletes the > inconsistent ones. > > >> the users bucket index itself is in .users.uid, where it is kept under >> an object named <uid>.buckets. > > Ah yes, I see. Clearing it from there works. One final bit of info that references the bucket is on the user buckets index. You can try running radosgw-admin bucket unlink --bucket=<bucket> --uid=<uid> if you need to remove that ref. Yehuda -- 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
