Corpus structure

ceph.git/ceph-object-corpus is a submodule.:

bin/   # misc scripts
archive/$version/objects/$type/$hash  # a sample of encoded objects from a specific version

You can also mark known or deliberate incompatibilities between versions with:


The presence of a file indicates that new versions of code cannot decode old objects across that $version (this is normally the case).

How to generate an object corpus

We can generate an object corpus for a particular version of ceph like so.

  1. Checkout a clean repo (best not to do this where you normally work):

    git clone ceph.git
    cd ceph
    git submodule update --init --recursive
  2. Build with flag to dump objects to /tmp/foo:

    rm -rf /tmp/foo ; mkdir /tmp/foo
    ./ -e /tmp/foo
  3. Start via vstart:

    cd src
    MON=3 OSD=3 MDS=3 RGW=1 ./ -n -x
  4. Use as much functionality of the cluster as you can, to exercise as many object encoder methods as possible:

    ./rados -p rbd bench 10 write -b 123
    ./ceph osd out 0
    ./init-ceph stop osd.1
    for f in ../qa/workunits/cls/*.sh ; do PATH=".:$PATH" $f ; done
    ./init-ceph restart mds.a

Do some more stuff with rgw if you know how.

  1. Stop:

  2. Import the corpus (this will take a few minutes):

    test/encoding/ /tmp/foo `./ceph-dencoder version` ../ceph-object-corpus/archive
    test/encoding/ ../ceph-object-corpus/archive
  3. Prune it! There will be a bazillion copies of various objects, and we only want a representative sample.:

    pushd ../ceph-object-corpus
  4. Verify the tests pass:

    make check-local
  5. Commit it to the corpus repo and push:

    pushd ../ceph-object-corpus
    git checkout -b wip-new
    git add archive/`../src/ceph-dencoder version`
    git commit -m `../src/ceph-dencoder version`
    git remote add cc
    git push cc wip-new
  6. Go test it out:

    cd my/regular/tree
    cd ceph-object-corpus
    git fetch origin
    git checkout wip-new
    cd ../src
    make check-local
  7. If everything looks good, update the submodule master branch, and commit the submodule in ceph.git.