Looks like the neutron network allocation is taking so long for the 150 servers that the test times out. Grep for 'Instance network_info:' in the screen-n-cpu log and compare to when the test times out. Its only slow by a few seconds which is why the test only fails sometimes. Perhaps we should drop TEMPEST_LARGE_OPS_NUMBER a little bit so it fits within the timeout?
Looks like the neutron network allocation is taking so long for the 150 servers that the test times out. Grep for 'Instance network_info:' in the screen-n-cpu log and compare to when the test times out. Its only slow by a few seconds which is why the test only fails sometimes. Perhaps we should drop TEMPEST_ LARGE_OPS_ NUMBER a little bit so it fits within the timeout?