Performance regression uploading images to glance in juno

Are you using SSL (https)? Judging on the numbers alone I'd be suspicious that SSL compression has been enabled -- but it's just a guess.

The rbd connection on the backend appears to be OK:

ubuntu@juju-t-machine-6:~$ time sudo rbd --id glance --pool glance export 42f31416-261c-46c6-823d-5b38722ed800 localdisk
Exporting image: 100% complete...done.

real 0m30.209s
user 0m4.982s
sys 0m3.226s

ubuntu@juju-t-machine-6:~$ time sudo rbd --id glance --pool glance import localdisk 42f31416-261c-46c6-823d-5b38722ed800-clone
Importing image: 100% complete...done.

real 0m4.122s
user 0m0.164s
sys 0m0.354s

The image is 249MB in size.

I think the problem is that glance is using a tiny calculated chunk size (from python-glance-store):

            chunk = self.conf.glance_store.rbd_store_chunk_size
            self.chunk_size = chunk * (1024 ^ 2)

this should be (from original glance rbd driver):

           1024 ** 2

Resulting in alot of tiny chucked writes instead of the default 8MB writes.

I see this is fixed already in glance_store; hopefully there will be another release soon, in the meantime I'll cherry-pick the commit that fixes this issue.

This was fixed under https://bugs.launchpad.net/glance/+bug/1370247 in glance_store.

This bug was fixed in the package python-glance-store - 0.1.8-1ubuntu1

---------------
python-glance-store (0.1.8-1ubuntu1) utopic; urgency=medium

  * d/p/fix-rbd-chunk-size.patch: Cherry pick fix from upstream VCS to
    correctly calculate RBD chunk size, resolving performance regression
    with rbd backend (LP: #1378388).
 -- James Page <email address hidden> Tue, 07 Oct 2014 16:08:15 +0100