I looked back t your comment as I too found weird that a tempest network was created so early.
That network is indeed the 'private' devstack network, which is used throughout the non-isolated jobs.
It seems the network (and the subnet) are created before the q-agt service is started, and this could lead to the situation described above. This failure mode would affect only non-isolated jobs. But if we look at [1], non-isolated jobs currently account for 70% of the total failures for bug 1253896. I do not have a way to say how much of this failures currently occur for this issue, but I guess that it is a consistent percentage of them.
Perhaps this could be easily solved by rearranging how devstack creates network and starts neutron services.
Eugene,
I looked back t your comment as I too found weird that a tempest network was created so early.
That network is indeed the 'private' devstack network, which is used throughout the non-isolated jobs.
It seems the network (and the subnet) are created before the q-agt service is started, and this could lead to the situation described above. This failure mode would affect only non-isolated jobs. But if we look at [1], non-isolated jobs currently account for 70% of the total failures for bug 1253896. I do not have a way to say how much of this failures currently occur for this issue, but I guess that it is a consistent percentage of them.
Perhaps this could be easily solved by rearranging how devstack creates network and starts neutron services.
[1] http:// logstash. openstack. org/#eyJmaWVsZH MiOltdLCJzZWFyY 2giOiJtZXNzYWdl OlwiU1NIVGltZW9 1dDogQ29ubmVjdG lvbiB0byB0aGVcI iBBTkQgbWVzc2Fn ZTpcInZpYSBTU0g gdGltZWQgb3V0Ll wiIEFORCBmaWxlb mFtZTpcImNvbnNv bGUuaHRtbFwiXG4 iLCJ0aW1lZnJhbW UiOiI2MDQ4MDAiL CJncmFwaG1vZGUi OiJjb3VudCIsIm9 mZnNldCI6MCwidG ltZSI6eyJ1c2VyX 2ludGVydmFsIjow fSwibW9kZSI6InN jb3JlIiwiYW5hbH l6ZV9maWVsZCI6I mJ1aWxkX25hbWUi fQ==