snat arp entry missing in qrouter namespace

[Hemanth Nakkina]

[Impact]
Load Balancers deployed on the cloud are unreachable

[Test Case]
1. Deploy openstack with atleast 4 compute nodes with networking features DVR SNAT+L3HA
2. Execute the script test_snat_arp_entry.sh
3. The script loops for 20 times creating network, router and connecting router to external, internal network and checking if ARP entries are populated properly on qrouter namespaces
4. The script stops if arp entries are missing.
5. If the script runs for 20 loops, then there are no issues.

[Regression Potential]
The issue only happens a few times when a router is created, external gateway set and internal subnet attached to router in quick succession. In other cases, the arp entry of snat is already added.
The fix just adds extra logic to add arp entry retrieving snat information from the router. In working cases, this extra logic will execute commands to add arp entry twice which should not cause further issues.

[Original Bug Report]
In one of the cloud environment, the FIP attached to the Octavia Loadbalancer VIP is not reachable. After analysis, we found the ARP entry for SNAT IP is missing in the qrouter namespace where Amphora VM is running. And so the return packets are not forwarded from qrouter to snat on active l3-agent node.

Version:
Ubuntu Ussuri packages (16.3.2 point release)
DVR+SNAT+L3HA enabled

Expectation is to have PERMANENT arp entry for snat ip on qrouter namespace on all compute nodes
192.168.33.238 dev qr-4ee692e0-7a lladdr fa:16:3e:25:6a:73 used 38/38/38 probes 0 PERMANENT

How to reproduce:

Attaching a script to simulate the problem (without octavia) with following steps
1. network/subnet/router is created, network attached to router
2. verify if qrouter on all compute nodes has arp entries related to snat ip
3. if arp entries exists, delete network/subnet/router
4. Repeat steps 1,2,3 until missing arp entry is observed.

I am able to reproduce missing arp entry sometimes in 3rd loop and sometimes in 6th loop.

Observed arp entries for snat ip is updated at the following places [1] [2] but get_snat_interfaces() and get_ports_by_subnet() are not updated with snat ip in non-working cases.

[1] https://opendev.org/openstack/neutron/src/commit/dfd04115b059c2263cdd8ac44ccc2ec47614bcc3/neutron/agent/l3/dvr_local_router.py#L570
[2] https://opendev.org/openstack/neutron/src/commit/dfd04115b059c2263cdd8ac44ccc2ec47614bcc3/neutron/agent/l3/dvr_local_router.py#L317

[Hemanth Nakkina (hemanth-n)]

Test script to verify if arp entries for snat ip exists

[Hemanth Nakkina (hemanth-n)]

Here is the ip neigh show output for non-working scenario (SNAT IP: 192.168.33.238 is missing on nova-compute/2, nova-compute/3)

$ juju run -a nova-compute -- sudo ip netns exec qrouter-db77f537-ca82-4d25-8539-d7891e2d6b8a ip -s neigh show
- Stdout: |
    169.254.103.137 dev rfp-db77f537-c lladdr 76:31:59:17:f2:b5 used 3215/3215/3215 probes 0 PERMANENT
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3216/3216/3216 probes 0 PERMANENT
    192.168.33.238 dev qr-4ee692e0-7a lladdr fa:16:3e:25:6a:73 used 3214/3214/3214 probes 0 PERMANENT
  UnitId: nova-compute/0
- Stdout: |
    192.168.33.238 dev qr-4ee692e0-7a lladdr fa:16:3e:25:6a:73 used 3214/3214/3214 probes 0 PERMANENT
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3218/3218/3218 probes 0 PERMANENT
    169.254.75.95 dev rfp-db77f537-c lladdr 1a:f5:09:ef:2f:b1 used 3215/3215/3215 probes 0 PERMANENT
  UnitId: nova-compute/1
- Stdout: |
    169.254.75.95 dev rfp-db77f537-c lladdr 4a:6e:d0:af:88:10 used 3226/3226/3226 probes 0 PERMANENT
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3214/3214/3214 probes 0 PERMANENT
  UnitId: nova-compute/2
- Stdout: |
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3218/3218/3218 probes 0 PERMANENT
    169.254.75.95 dev rfp-db77f537-c lladdr a2:63:61:ca:14:31 used 3229/3229/3229 probes 0 PERMANENT
  UnitId: nova-compute/3

Launched a test VM on nova-compute/2. VM got IP 192.168.33.171 and corresponding arp entry added. At this point of time SNAT IP is also added but it is in FAILED state.

$ juju run -a nova-compute -- sudo ip netns exec qrouter-db77f537-ca82-4d25-8539-d7891e2d6b8a ip -s neigh show
- Stdout: |
    169.254.103.137 dev rfp-db77f537-c lladdr 76:31:59:17:f2:b5 used 3292/3292/3292 probes 0 PERMANENT
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3293/3293/3293 probes 0 PERMANENT
    192.168.33.238 dev qr-4ee692e0-7a lladdr fa:16:3e:25:6a:73 used 9/3291/3291 probes 0 PERMANENT
    fe80::f816:3eff:fe1a:3c82 dev qr-4ee692e0-7a lladdr fa:16:3e:1a:3c:82 used 12/72/12 probes 0 STALE
  UnitId: nova-compute/0
- Stdout: |
    192.168.33.238 dev qr-4ee692e0-7a lladdr fa:16:3e:25:6a:73 used 3291/3291/3291 probes 0 PERMANENT
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3296/3296/3296 probes 0 PERMANENT
    169.254.75.95 dev rfp-db77f537-c lladdr 1a:f5:09:ef:2f:b1 used 3293/3293/3293 probes 0 PERMANENT
  UnitId: nova-compute/1
- Stdout: |
    192.168.33.238 dev qr-4ee692e0-7a used 6/69/6 probes 6 FAILED
    192.168.33.171 dev qr-4ee692e0-7a lladdr fa:16:3e:1a:3c:82 ref 1 used 6/2/2 probes 1 REACHABLE
    169.254.75.95 dev rfp-db77f537-c lladdr 4a:6e:d0:af:88:10 used 3303/3303/3303 probes 0 PERMANENT
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3291/3291/3291 probes 0 PERMANENT
  UnitId: nova-compute/2
- Stdout: |
    192.168.33.2 dev qr-4ee692e0-7a lladdr fa:16:3e:4a:1e:a4 used 3295/3295/3295 probes 0 PERMANENT
    169.254.75.95 dev rfp-db77f537-c lladdr a2:63:61:ca:14:31 used 3306/3306/3306 probes 0 PERMANENT
  UnitId: nova-compute/3

[Hemanth Nakkina (hemanth-n)]

After looking closer, the problem is in step 1 - network/subnet/router is created, network attached to router

The script which is used for reproducer, while attaching network to router is not waiting until the router is active with 1 active/2 stanadby l3 agents. So the snat arp entry is not populated in cases where network is attached to router before the router completes l3 agent assignment and completion of ha setup.

Adding sleep of 60 seconds (sufficient enough in my environment for router to properly set with ha) before adding network to router, I dont see the missing snat arp entry.

Leaving this bug open to see if it is worth exploring sending post update after router HA is setup to update snat arp entry.

[Slawek Kaplonski (slaweq)]

I have env for debug that so I will try to reproduce and triage it at least.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/neutron/+/799197

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/neutron/+/799197
Committed: https://opendev.org/openstack/neutron/commit/be7d0bb6abc893e53dfc864c52506928b1d38fa3
Submitter: "Zuul (22348)"
Branch: master

commit be7d0bb6abc893e53dfc864c52506928b1d38fa3
Author: Hemanth Nakkina <email address hidden>
Date: Fri Jul 2 17:01:55 2021 +0530

    Update arp entry of snat port on qrouter ns

    In some cases, the arp entry of snat port is not updated
    in qrouter namespace. l3-agent calls get_ports_by_subnet()
    while setting arps for the subnet. And the snat port is
    not returned if it is still unbound. One of the scenario
    this is observed is when router is created, external
    gateway set and internal subnet attached to router in
    quick succession.

    This patch retrieves snat port details from router info
    as well and updates arp entry for snat port.

    Closes-Bug: #1933092
    Change-Id: I7ee797b4b930306cf6360922d855f8b24f1b813d

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/wallaby
Review: https://review.opendev.org/c/openstack/neutron/+/799375

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/victoria
Review: https://review.opendev.org/c/openstack/neutron/+/799476

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/ussuri
Review: https://review.opendev.org/c/openstack/neutron/+/799478

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/neutron/+/799375
Committed: https://opendev.org/openstack/neutron/commit/85a668dd1629d71f52ff20da3d848e2cbb397e43
Submitter: "Zuul (22348)"
Branch: stable/wallaby

commit 85a668dd1629d71f52ff20da3d848e2cbb397e43
Author: Hemanth Nakkina <email address hidden>
Date: Fri Jul 2 17:01:55 2021 +0530

    Update arp entry of snat port on qrouter ns

    In some cases, the arp entry of snat port is not updated
    in qrouter namespace. l3-agent calls get_ports_by_subnet()
    while setting arps for the subnet. And the snat port is
    not returned if it is still unbound. One of the scenario
    this is observed is when router is created, external
    gateway set and internal subnet attached to router in
    quick succession.

    This patch retrieves snat port details from router info
    as well and updates arp entry for snat port.

    Closes-Bug: #1933092
    Change-Id: I7ee797b4b930306cf6360922d855f8b24f1b813d
    (cherry picked from commit be7d0bb6abc893e53dfc864c52506928b1d38fa3)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/neutron/+/799476
Committed: https://opendev.org/openstack/neutron/commit/8a43eb4563e787ea43e15cb49968818059a34c2b
Submitter: "Zuul (22348)"
Branch: stable/victoria

commit 8a43eb4563e787ea43e15cb49968818059a34c2b
Author: Hemanth Nakkina <email address hidden>
Date: Fri Jul 2 17:01:55 2021 +0530

    Update arp entry of snat port on qrouter ns

    In some cases, the arp entry of snat port is not updated
    in qrouter namespace. l3-agent calls get_ports_by_subnet()
    while setting arps for the subnet. And the snat port is
    not returned if it is still unbound. One of the scenario
    this is observed is when router is created, external
    gateway set and internal subnet attached to router in
    quick succession.

    This patch retrieves snat port details from router info
    as well and updates arp entry for snat port.

    Closes-Bug: #1933092
    Change-Id: I7ee797b4b930306cf6360922d855f8b24f1b813d
    (cherry picked from commit be7d0bb6abc893e53dfc864c52506928b1d38fa3)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/neutron/+/799478
Committed: https://opendev.org/openstack/neutron/commit/f1a9f4ed62fd2567cac174f80f87de53148ea7b9
Submitter: "Zuul (22348)"
Branch: stable/ussuri

commit f1a9f4ed62fd2567cac174f80f87de53148ea7b9
Author: Hemanth Nakkina <email address hidden>
Date: Fri Jul 2 17:01:55 2021 +0530

    Update arp entry of snat port on qrouter ns

    In some cases, the arp entry of snat port is not updated
    in qrouter namespace. l3-agent calls get_ports_by_subnet()
    while setting arps for the subnet. And the snat port is
    not returned if it is still unbound. One of the scenario
    this is observed is when router is created, external
    gateway set and internal subnet attached to router in
    quick succession.

    This patch retrieves snat port details from router info
    as well and updates arp entry for snat port.

    Closes-Bug: #1933092
    Change-Id: I7ee797b4b930306cf6360922d855f8b24f1b813d
    (cherry picked from commit be7d0bb6abc893e53dfc864c52506928b1d38fa3)

[Hemanth Nakkina (hemanth-n)]

SRU team,
All debdiffs for Ubuntu I/H/G/F and UCA X/W/V/U are uploaded.

[Ubuntu Foundations Team Bug Bot (crichton)]

The attachment "Debdiff for impish" seems to be a debdiff. The ubuntu-sponsors team has been subscribed to the bug report so that they can review and hopefully sponsor the debdiff. If the attachment isn't a patch, please remove the "patch" flag from the attachment, remove the "patch" tag, and if you are member of the ~ubuntu-sponsors, unsubscribe the team.

[This is an automated message performed by a Launchpad user owned by ~brian-murray, for any issue please contact him.]

[Dan Streetman (ddstreet)]

pushed to U/V/W and master branches:

https://git.launchpad.net/~ubuntu-openstack-dev/ubuntu/+source/neutron/commit/?id=100e69d4593dde0363d8d368c836c2cc49dd2d7b

https://git.launchpad.net/~ubuntu-openstack-dev/ubuntu/+source/neutron/commit/?id=6348afdd66039ca5365169e315872fb38ad64f3f

https://git.launchpad.net/~ubuntu-openstack-dev/ubuntu/+source/neutron/commit/?id=5a27fae295fc475ea3d7105850066655c23c1269

https://git.launchpad.net/~ubuntu-openstack-dev/ubuntu/+source/neutron/commit/?id=22a4f5867c2a44d5ee6636af9fe2f79624a9cd6f

chris plans to start new stable point releases tomorrow so this will get uploaded along with that

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 16.4.0 release.

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 17.2.0 release.

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 18.1.0 release.

[Launchpad Janitor (janitor)]

This bug was fixed in the package neutron - 2:18.1.0+git2021072117.147830620f-0ubuntu1

---------------
neutron (2:18.1.0+git2021072117.147830620f-0ubuntu1) impish; urgency=medium

  [ Hemanth Nakkina ]
  * d/p/0001-Update-arp-entry-of-snat-port-on-qrouter-ns.patch: Fix to
    update arp entry of snat port on qrouter ns (LP: #1933092)

  [ Corey Bryant ]
  * New upstream snapshot for OpenStack Xena.
  * d/control: Align (Build-)Depends with upstream.
  * d/p/*: Rebased.
  * d/p/0001-Update-arp-entry-of-snat-port-on-qrouter-ns.patch: Dropped.
    Fixed in upstream snapshot.

 -- Corey Bryant <email address hidden> Wed, 21 Jul 2021 17:22:28 -0400

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/train
Review: https://review.opendev.org/c/openstack/neutron/+/807244

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 19.0.0.0rc1 release candidate.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/c/openstack/neutron/+/807244
Committed: https://opendev.org/openstack/neutron/commit/c689ac5a661a15ec7ad1098d21dc5affce97fb04
Submitter: "Zuul (22348)"
Branch: stable/train

commit c689ac5a661a15ec7ad1098d21dc5affce97fb04
Author: Hemanth Nakkina <email address hidden>
Date: Fri Jul 2 17:01:55 2021 +0530

    Update arp entry of snat port on qrouter ns

    In some cases, the arp entry of snat port is not updated
    in qrouter namespace. l3-agent calls get_ports_by_subnet()
    while setting arps for the subnet. And the snat port is
    not returned if it is still unbound. One of the scenario
    this is observed is when router is created, external
    gateway set and internal subnet attached to router in
    quick succession.

    This patch retrieves snat port details from router info
    as well and updates arp entry for snat port.

    Conflicts:
        neutron/agent/l3/dvr_local_router.py

    Closes-Bug: #1933092
    Change-Id: I7ee797b4b930306cf6360922d855f8b24f1b813d
    (cherry picked from commit be7d0bb6abc893e53dfc864c52506928b1d38fa3)
    (cherry picked from commit f1a9f4ed62fd2567cac174f80f87de53148ea7b9)

[OpenStack Infra (hudson-openstack)]

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