Description of problem:

We have instances failing to boot because users are putting very large
subnets (e.g., /15) into allowed_address_pairs in Neutron.  It appears
that we're hitting some sort of 64K limit in the ipset implementation.

The users putting subnets into allowed_address_pairs, so it seems that
they should be stored in ipset as a subnet as well, not individual
addresses.

This results in errors like from an instance booting:
2015-09-25 17:17:21.418 51840 INFO heat.engine.stack [-] Stack CREATE
FAILED (VMG-NEC2-OAMA-24ilgto3rllb): Resource CREATE failed: Error:
Server VMG-NEC2-OAM-A delete failed: (500) Build of instance
6031be29-f194-43fb-b05b-92e0c2d7cd18 aborted: Failed to allocate the
network(s), not rescheduling.


On the compute host, the neutron logs contain items like:
2015-09-25 12:04:13.129 58378 ERROR neutron.agent.linux.utils
[req-c8c2e2e2-1cc5-4fab-b1b6-b84f78043459 None]
Command: ['sudo', 'neutron-rootwrap', '/etc/neutron/rootwrap.conf',
'ipset', 'add', '-exist', 'IPv4c3742bfb-7f15-4c92-b', '192.168.10.0/15']
Exit code: 1
Stdout: ''
Stderr: 'ipset v6.19: Hash is full, cannot add more elements\n'


We can see that the ipset hash is indeed near a 64K limit, if that is
problematic:
[root@host nova]# ipset list | wc -l
65647


We can see that the sets have a limit of 65536:
Name: IPv6f60f0540-fa6c-4a8d-9
Type: hash:ip
Revision: 1
Header: family inet6 hashsize 1024 maxelem 65536



Version-Release number of selected component (if applicable): OSP6


How reproducible:
100% in larger deployments

Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info:

Upstream bug: https://bugs.launchpad.net/neutron/+bug/1439817

Upstream code commit for OSP6: https://git.openstack.org/cgit/openstack/neutron/commit/?id=0325b98c54af276064e367db603bfeb525bbb790