This bug is still happening and it hurts big Openstack Deployments.
Its easy to reproduce with Parallel on latest devstack with the fixes that were introduced.
for i in {1..300}; do echo port${i};done | parallel -vv -j 50 neutron port-create public --name {}
We can confirm that also it is visible when it generates IP from large IPv6 class, even with latest patch.
Introduced patch makes the whole ip-generation process very slow due a lot of retries and waits.
We consider to rewrite the gen function to make it Deadlock and Duplicate Error resistant.
This bug is still happening and it hurts big Openstack Deployments.
Its easy to reproduce with Parallel on latest devstack with the fixes that were introduced.
for i in {1..300}; do echo port${i};done | parallel -vv -j 50 neutron port-create public --name {}
According to what LIU Yulong say the main issue is the generator of IP addresses: /github. com/openstack/ neutron/ blob/master/ neutron/ ipam/drivers/ neutrondb_ ipam/driver. py#L153- L178
https:/
We can confirm that also it is visible when it generates IP from large IPv6 class, even with latest patch.
Introduced patch makes the whole ip-generation process very slow due a lot of retries and waits.
We consider to rewrite the gen function to make it Deadlock and Duplicate Error resistant.