all-in-one overcloud with Glance + RBD: image upload hangs forever

[Emilien Macchi]

I'm currently adding Ceph to tripleo/scenario001 CI job and configure Nova, Glance and Gnocchi to use RBD backend.
The pingtest fails to upload an image to Glance: it hangs forever and timeout (at the time of the HAproxy value, 1 minute by default). I tried to increase the HAproxy timeout to 1h30 but it doesn't help, I've got same results where it timeouts after 1h30.

I can see the image upload in Glance logs and I see zero error in logs.

This is the client trying to upload an image and timeout:
http://logs.openstack.org/12/366812/3/check/gate-tripleo-ci-centos-7-scenario001-multinode-nv/ed56d11/console.html#_2016-09-08_05_26_44_159354

Glance API logs:
http://logs.openstack.org/12/366812/3/check/gate-tripleo-ci-centos-7-scenario001-multinode-nv/ed56d11/logs/subnode-2/var/log/glance/api.txt.gz#_2016-09-08_03_56_28_747

Glance Registry logs:
http://logs.openstack.org/12/366812/3/check/gate-tripleo-ci-centos-7-scenario001-multinode-nv/ed56d11/logs/subnode-2/var/log/glance/registry.txt.gz#_2016-09-08_03_56_44_506

I compared Glance, Ceph and HAproxy configurations between OVB (that also test Ceph + RBD backend) and scenario001 jobs, and they are the same, so there is something else that prevent the upload to happen.

I thought it could be the CPU load somehow, but 1h30 of timeout for a cirros image would be really too much and I don't see anything in Glance logs.