Default gateway in HA router namespace not set if using Keepalived 1.x

[Alexander Gräb]

With Rocky commit f2d234e introduces a change making Neutron more compatible with Keepalived 2.x. A new option `no_track` is passed to Keepalived by the L3 agent which Keepalived 1.x doesn't recognize, thus the configuration is partially not applied by Keepalived 1.x. resulting in a missing default gateway. Thus instances using HA routers to communicate with the outside world are unable to reach the internet and cannot be reached via their floating IP addresses.

There are some workarounds to trigger the creating of the default gateway like disable and enable the router again or disable/re-start the L3 agent which hosts the master namespace of the HA router.

Steps to reproduce:
1. Create an HA router
2. Add the gateway network (now the default gateway should be set within the routers master network namespace but there is none)
3. Connect a tenant subnet to the router
4. Create an instance connected to the tenant network created in step 3
5. Try to reach the internet from within the instances created in step 4

I was able to get some log output out of Keepalived:
Tue Aug 4 11:59:58 2020: Starting Keepalived v1.3.9 (10/21,2017)
Tue Aug 4 11:59:58 2020: Opening file '/var/lib/neutron/ha_confs/019f8036-8730-4584-9e07-c4a6504447ab/keepalived.conf'.
Tue Aug 4 11:59:58 2020: Starting VRRP child process, pid=2864
Tue Aug 4 11:59:58 2020: Registering Kernel netlink reflector
Tue Aug 4 11:59:58 2020: Registering Kernel netlink command channel
Tue Aug 4 11:59:58 2020: Registering gratuitous ARP shared channel
Tue Aug 4 11:59:58 2020: Opening file '/var/lib/neutron/ha_confs/019f8036-8730-4584-9e07-c4a6504447ab/keepalived.conf'.
Tue Aug 4 11:59:58 2020: Unknown configuration entry 'no_track' for ip address - ignoring
Tue Aug 4 11:59:58 2020: Unknown configuration entry 'no_track' for ip address - ignoring
Tue Aug 4 11:59:58 2020: Cannot specify scope for IPv6 addresses (fe80::f816:3eff:fe2c:6622/64) - ignoring scope
Tue Aug 4 11:59:58 2020: VRRP parsed invalid IP no_track. skipping IP...
Tue Aug 4 11:59:58 2020: unknown route keyword no_track
Tue Aug 4 11:59:58 2020: VRRP_Instance(VR_23) removing protocol VIPs.
Tue Aug 4 11:59:58 2020: VRRP_Instance(VR_23) removing protocol E-VIPs.
Tue Aug 4 11:59:58 2020: Using LinkWatch kernel netlink reflector...
Tue Aug 4 11:59:58 2020: VRRP_Instance(VR_23) Entering BACKUP STATE
Tue Aug 4 11:59:58 2020: VRRP sockpool: [ifindex(1033), proto(112), unicast(0), fd(9,10)]

You see it complaining about the 'no_track' option.

We use Kolla containers with Ubuntu base. Even though Keepalived 2 was release quite a while ago, it still only provides Keepalived 1.x via the package repositories. Even Kolla in the latest version still uses Ubuntu 18.04 as base with Keepalived 1.x. Theoretically all users using Kolla containers with Ubuntu base (other base images not tested) are affected. There seems to be no apt sources for Keepalived 2.x for Ubuntu 18.04. You need to compile it from source in order to get a newer version.

Maybe it should depend on the Keepalived version whether to pass the 'no_track' option or make it configurable.

[Rodolfo Alonso (rodolfo-alonso-hernandez)]

Hello:

Although this is not recommended, we'll need to make a runtime check. We can't enforce the keepalived version and we should support both.

We need something similar to https://review.opendev.org/#/c/726079/1/neutron/agent/linux/dhcp.py.

Regards.

[Alexander Gräb (alexander.graeb)]

Sounds reasonable. The Keepalived version could be determined with `keepalived -v`.

[Slawek Kaplonski (slaweq)]

Hi,

I just wanted to propose something similar what Rodolfo proposed.
If You want to propose patch, please assign this bug to You, otherwise I can propose patch today.

[Alexander Gräb (alexander.graeb)]

If you propose a patch I would be glad to try it out.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/745641

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/745641
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=7abe0ee34c367b4abf84820048b4aed643fc1162
Submitter: Zuul
Branch: master

commit 7abe0ee34c367b4abf84820048b4aed643fc1162
Author: Slawek Kaplonski <email address hidden>
Date: Tue Aug 11 10:47:24 2020 +0200

    Add 'keepalived_use_no_track' config option

    Patch [1] added option "no_track" to the keepalived's config file which
    is generated by L3 agent in HA mode.
    This was added to handle properly keepalived 2.x and interfaces which
    are in DOWN state in the backup nodes.
    But this "no_track" option is not compatible with keepalived 1.x series
    which is available e.g. on Ubuntu 18.04.

    As there is no easy way to check automatically if keepalived supports or
    not this config flag, this patch introduces new config option
    "keepalived_use_no_track".
    If this config option will be set to False, neutron L3 agent will not
    add "no_track" to the keepalived's config.

    As master branch is moving to gate on Ubuntu 20.04 where keepalived 2.x
    is already available, this new config option default value is set to
    True.

    [1] https://review.opendev.org/#/c/721799/

    Change-Id: I2dfdb9f56de28d56ca0f240ff34fa7c3a12e339b
    Closes-Bug: #1890400

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/ussuri
Review: https://review.opendev.org/747855

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/train
Review: https://review.opendev.org/747856

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/stein
Review: https://review.opendev.org/747857

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/rocky
Review: https://review.opendev.org/747862

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/747855
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=43079e3fc919abd15f6b352ccc38cc8b6218e9e9
Submitter: Zuul
Branch: stable/ussuri

commit 43079e3fc919abd15f6b352ccc38cc8b6218e9e9
Author: Slawek Kaplonski <email address hidden>
Date: Tue Aug 11 10:47:24 2020 +0200

    Add 'keepalived_use_no_track' config option

    Patch [1] added option "no_track" to the keepalived's config file which
    is generated by L3 agent in HA mode.
    This was added to handle properly keepalived 2.x and interfaces which
    are in DOWN state in the backup nodes.
    But this "no_track" option is not compatible with keepalived 1.x series
    which is available e.g. on Ubuntu 18.04.

    As there is no easy way to check automatically if keepalived supports or
    not this config flag, this patch introduces new config option
    "keepalived_use_no_track".
    If this config option will be set to False, neutron L3 agent will not
    add "no_track" to the keepalived's config.

    As master branch is moving to gate on Ubuntu 20.04 where keepalived 2.x
    is already available, this new config option default value is set to
    True.

    [1] https://review.opendev.org/#/c/721799/

    Change-Id: I2dfdb9f56de28d56ca0f240ff34fa7c3a12e339b
    Closes-Bug: #1890400
    (cherry picked from commit 7abe0ee34c367b4abf84820048b4aed643fc1162)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/747862
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=8ffbcb5a854ac12977e275f96f5f68f6cf41e24f
Submitter: Zuul
Branch: stable/rocky

commit 8ffbcb5a854ac12977e275f96f5f68f6cf41e24f
Author: Slawek Kaplonski <email address hidden>
Date: Tue Aug 11 10:47:24 2020 +0200

    Add 'keepalived_use_no_track' config option

    Patch [1] added option "no_track" to the keepalived's config file which
    is generated by L3 agent in HA mode.
    This was added to handle properly keepalived 2.x and interfaces which
    are in DOWN state in the backup nodes.
    But this "no_track" option is not compatible with keepalived 1.x series
    which is available e.g. on Ubuntu 18.04.

    As there is no easy way to check automatically if keepalived supports or
    not this config flag, this patch introduces new config option
    "keepalived_use_no_track".
    If this config option will be set to False, neutron L3 agent will not
    add "no_track" to the keepalived's config.

    As master branch is moving to gate on Ubuntu 20.04 where keepalived 2.x
    is already available, this new config option default value is set to
    True.

    [1] https://review.opendev.org/#/c/721799/

    Conflicts:
        neutron/conf/agent/l3/config.py
        neutron/tests/functional/agent/linux/test_keepalived.py

    Change-Id: I2dfdb9f56de28d56ca0f240ff34fa7c3a12e339b
    Closes-Bug: #1890400
    (cherry picked from commit 7abe0ee34c367b4abf84820048b4aed643fc1162)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/747856
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=b2b9fd66b9e673a4cbc43258ab0ac306db2f38d1
Submitter: Zuul
Branch: stable/train

commit b2b9fd66b9e673a4cbc43258ab0ac306db2f38d1
Author: Slawek Kaplonski <email address hidden>
Date: Tue Aug 11 10:47:24 2020 +0200

    Add 'keepalived_use_no_track' config option

    Patch [1] added option "no_track" to the keepalived's config file which
    is generated by L3 agent in HA mode.
    This was added to handle properly keepalived 2.x and interfaces which
    are in DOWN state in the backup nodes.
    But this "no_track" option is not compatible with keepalived 1.x series
    which is available e.g. on Ubuntu 18.04.

    As there is no easy way to check automatically if keepalived supports or
    not this config flag, this patch introduces new config option
    "keepalived_use_no_track".
    If this config option will be set to False, neutron L3 agent will not
    add "no_track" to the keepalived's config.

    As master branch is moving to gate on Ubuntu 20.04 where keepalived 2.x
    is already available, this new config option default value is set to
    True.

    [1] https://review.opendev.org/#/c/721799/

    Conflicts:
        neutron/conf/agent/l3/config.py

    Change-Id: I2dfdb9f56de28d56ca0f240ff34fa7c3a12e339b
    Closes-Bug: #1890400
    (cherry picked from commit 7abe0ee34c367b4abf84820048b4aed643fc1162)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/747857
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=abd4e849257b401a5e7c2d61b753cc6ab52bdc6d
Submitter: Zuul
Branch: stable/stein

commit abd4e849257b401a5e7c2d61b753cc6ab52bdc6d
Author: Slawek Kaplonski <email address hidden>
Date: Tue Aug 11 10:47:24 2020 +0200

    Add 'keepalived_use_no_track' config option

    Patch [1] added option "no_track" to the keepalived's config file which
    is generated by L3 agent in HA mode.
    This was added to handle properly keepalived 2.x and interfaces which
    are in DOWN state in the backup nodes.
    But this "no_track" option is not compatible with keepalived 1.x series
    which is available e.g. on Ubuntu 18.04.

    As there is no easy way to check automatically if keepalived supports or
    not this config flag, this patch introduces new config option
    "keepalived_use_no_track".
    If this config option will be set to False, neutron L3 agent will not
    add "no_track" to the keepalived's config.

    As master branch is moving to gate on Ubuntu 20.04 where keepalived 2.x
    is already available, this new config option default value is set to
    True.

    [1] https://review.opendev.org/#/c/721799/

    Conflicts:
        neutron/conf/agent/l3/config.py

    Change-Id: I2dfdb9f56de28d56ca0f240ff34fa7c3a12e339b
    Closes-Bug: #1890400
    (cherry picked from commit 7abe0ee34c367b4abf84820048b4aed643fc1162)

[Corey Bryant (corey.bryant)]

I opened the following bug against this fix because there's no run-time check of keepalived version making this fix not desirable if running with keepalived < 2.x:
https://bugs.launchpad.net/ubuntu/+source/neutron/+bug/1896506

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron rocky-eol release.