[fullstack] Error assigning IPv4 (network address) in "test_gateway_ip_changed"

[Rodolfo Alonso]

Error assigning IPv4 (network address) in "test_gateway_ip_changed".

LOG: https://8e3d76ba7bcafd7367d8-a42dfacf856f2ce428049edff149969f.ssl.cf1.rackcdn.com/730888/1/check/neutron-fullstack/31482ea/testr_results.html

ERROR MESSAGE: http://paste.openstack.org/show/794029/
"""
neutronclient.common.exceptions.InvalidIpForNetworkClient: IP address 240.135.228.0 is not a valid IP for any of the subnets on the specified network.
"""

[Rodolfo Alonso (rodolfo-alonso-hernandez)]

According to the logs, the subnet allocation pools are: http://paste.openstack.org/show/794043/
{"start":"240.135.228.2","end":"240.135.228.254"}

This is the method calculating the available IPs for this network: https://github.com/openstack/neutron/blob/058dd639a9817fcdb638809b8228f20a1417774a/neutron/tests/fullstack/base.py#L130-L148

This method should never return an IP address outside the first allocation pool, in this case 240.135.228.2-240.135.228.254.

IPs used in this subnet: 240.135.228.2, 240.135.228.56 and 240.135.228.84

Port update request: http://paste.openstack.org/show/794044/

Sorry but I don't understand how "_find_available_ips" returned **240.135.228.0** and 240.135.228.3.

Help is welcomed.

Regards.

[Lajos Katona (lajos-katona)]

I check it

[Lajos Katona (lajos-katona)]

Hmmm, I was not able to create such situation even by playing with numbers.
As this is only a test method we can add debug logs to have more info for next times failure

[Brian Haley (brian-haley)]

I think I figured out the issue here, so will close this.

Here's my thinking.

The referenced log above was from a change on stable/train:

  https://review.opendev.org/c/openstack/neutron/+/730888

Lajos fixed a bug in _find_available_ips that seems related:

  https://review.opendev.org/c/openstack/neutron/+/692135

commit 3c9b0a5fac2e3a1321eadc272c8ed46aa61efd3e
Author: elajkat <email address hidden>
Date: Wed Oct 30 13:38:30 2019 +0100

    [fullstack] find ip based on allocation_pool

    _find_available_ips tried to find available ips based on the given
    subnet's cidr field, which can be misleading if random selection goes
    out-of allocation-pool. This patch changes this behaviour to use
    cidr's allocation_pool field.

    Closes-Bug: #1850292
    Change-Id: Ied2ffb5ed58007789b0f5157731687dc2e0b9bb1

That change is only included in these versions:

  master stable/2023.1 stable/2023.2 stable/victoria stable/wallaby stable/xena stable/zed
  unmaintained/victoria unmaintained/wallaby unmaintained/xena unmaintained/yoga

So I'm guessing merged in Victoria.

Since the change was from stable/train we could have had an issue with the subnet['cidr'] being used, which would have included IP addresses outside the start/end allocation pool.