Comment 1 for bug 1243251

this looks like a scalability issue. the logs look like the collector is flooded with meters: http://logs.openstack.org/68/52768/1/check/check-tempest-devstack-vm-full/76f83a4/logs/screen-ceilometer-collector.txt.gz

we use oslo code to handle session/flushing... might be a place to look at (maybe there is a config option we need to set to handle such loads.)