ovn: Add support for OVN SB Relay

Add support for deploying OVN SB relays [1]. This is enabled by default to improve large scale scenarios. Default mechanism deploys one relay group per 50 ovn-controller hosts and uses random but idempotent method to choose a relay for a given host. neutron-server will connect directly to SB cluster, not via the relays. [1]: https://docs.ovn.org/en/latest/tutorials/ovn-ovsdb-relay.html Co-Authored-By: Krzysztof Tomaszewski <[email protected]> Depends-On: https://review.opendev.org/c/openstack/kolla/+/928898 Change-Id: Id7d6973668d8b2b194913b533f8cb756c6708536 (cherry picked from commit 7b95939)
stackhpc · Feb 6, 2025 · de72559 · de72559
1 parent f890617
commit de72559
Show file tree

Hide file tree

Showing 20 changed files with 272 additions and 4 deletions.
diff --git a/ansible/group_vars/all.yml b/ansible/group_vars/all.yml
@@ -616,8 +616,16 @@ opensearch_dashboards_listen_port: "{{ opensearch_dashboards_port }}"
 
 ovn_nb_db_port: "6641"
 ovn_sb_db_port: "6642"
+# OVN SB Relay related variables
+ovn_sb_db_relay_count: "{{ ((groups['ovn-controller'] | length) / ovn_sb_db_relay_compute_per_relay | int) | round(0, 'ceil') | int }}"
+ovn_sb_db_relay_compute_per_relay: "50"
+ovn_sb_db_relay_port_prefix: "1664"
+ovn_sb_db_relay_port: "{{ ovn_sb_db_relay_port_prefix ~ ovn_sb_db_relay_client_group_id }}"
+ovn_sb_db_relay_client_group_id: "{{ range(1, ovn_sb_db_relay_count | int + 1) | random(seed=inventory_hostname) }}"
 ovn_nb_connection: "{% for host in groups['ovn-nb-db'] %}tcp:{{ 'api' | kolla_address(host) | put_address_in_context('url') }}:{{ ovn_nb_db_port }}{% if not loop.last %},{% endif %}{% endfor %}"
-ovn_sb_connection: "{% for host in groups['ovn-sb-db'] %}tcp:{{ 'api' | kolla_address(host) | put_address_in_context('url') }}:{{ ovn_sb_db_port }}{% if not loop.last %},{% endif %}{% endfor %}"
+ovn_sb_connection: "{{ ovn_sb_connection_relay if enable_ovn_sb_db_relay | bool else ovn_sb_connection_no_relay }}"
+ovn_sb_connection_no_relay: "{% for host in groups['ovn-sb-db'] %}tcp:{{ 'api' | kolla_address(host) | put_address_in_context('url') }}:{{ ovn_sb_db_port }}{% if not loop.last %},{% endif %}{% endfor %}"
+ovn_sb_connection_relay: "{% for host in groups['ovn-sb-db-relay'] %}tcp:{{ 'api' | kolla_address(host) | put_address_in_context('url') }}:{{ ovn_sb_db_relay_port }}{% if not loop.last %},{% endif %}{% endfor %}"
 
 ovsdb_port: "6640"
 
@@ -924,6 +932,7 @@ enable_octavia_driver_agent: "{{ enable_octavia | bool and neutron_plugin_agent
 enable_octavia_jobboard: "{{ enable_octavia | bool and 'amphora' in octavia_provider_drivers }}"
 enable_openvswitch: "{{ enable_neutron | bool and neutron_plugin_agent != 'linuxbridge' }}"
 enable_ovn: "{{ enable_neutron | bool and neutron_plugin_agent == 'ovn' }}"
+enable_ovn_sb_db_relay: "no"
 enable_ovs_dpdk: "no"
 enable_osprofiler: "no"
 enable_placement: "{{ enable_nova | bool or enable_zun | bool }}"

diff --git a/ansible/inventory/all-in-one b/ansible/inventory/all-in-one
@@ -592,6 +592,9 @@ ovn-database
 [ovn-sb-db:children]
 ovn-database
 
+[ovn-sb-db-relay:children]
+ovn-database
+
 [venus-api:children]
 venus
 

diff --git a/ansible/inventory/multinode b/ansible/inventory/multinode
@@ -611,6 +611,9 @@ ovn-database
 [ovn-sb-db:children]
 ovn-database
 
+[ovn-sb-db-relay:children]
+ovn-database
+
 [venus-api:children]
 venus
 

diff --git a/ansible/roles/neutron/templates/ml2_conf.ini.j2 b/ansible/roles/neutron/templates/ml2_conf.ini.j2
@@ -37,7 +37,7 @@ max_header_size = 38
 
 [ovn]
 ovn_nb_connection = {{ ovn_nb_connection }}
-ovn_sb_connection = {{ ovn_sb_connection }}
+ovn_sb_connection = {{ ovn_sb_connection_no_relay }}
 ovn_metadata_enabled = True
 enable_distributed_floating_ip = {{ neutron_ovn_distributed_fip | bool }}
 ovn_emit_need_to_frag = True

diff --git a/ansible/roles/octavia/templates/octavia.conf.j2 b/ansible/roles/octavia/templates/octavia.conf.j2
@@ -29,7 +29,7 @@ enabled_provider_agents = {{ octavia_provider_agents }}
 {% if neutron_plugin_agent == 'ovn' %}
 [ovn]
 ovn_nb_connection = {{ ovn_nb_connection }}
-ovn_sb_connection = {{ ovn_sb_connection }}
+ovn_sb_connection = {{ ovn_sb_connection_no_relay }}
 {% endif %}
 
 [haproxy_amphora]

diff --git a/ansible/roles/ovn-db/defaults/main.yml b/ansible/roles/ovn-db/defaults/main.yml
@@ -21,6 +21,15 @@ ovn_db_services:
     image: "{{ ovn_sb_db_image_full }}"
     volumes: "{{ ovn_sb_db_default_volumes + ovn_sb_db_extra_volumes }}"
     dimensions: "{{ ovn_sb_db_dimensions }}"
+  ovn-sb-db-relay:
+    container_name: ovn_sb_db_relay
+    group: ovn-sb-db-relay
+    enabled: "{{ enable_ovn_sb_db_relay | bool }}"
+    image: "{{ ovn_sb_db_relay_image_full }}"
+    iterate: true
+    iterate_var: "{{ ovn_sb_db_relay_count | int }}"
+    volumes: "{{ ovn_sb_db_relay_default_volumes + ovn_sb_db_relay_extra_volumes }}"
+    dimensions: "{{ ovn_sb_db_relay_dimensions }}"
 
 
 ####################
@@ -40,9 +49,14 @@ ovn_sb_db_image: "{{ docker_registry ~ '/' if docker_registry else '' }}{{ docke
 ovn_sb_db_tag: "{{ ovn_tag }}"
 ovn_sb_db_image_full: "{{ ovn_sb_db_image }}:{{ ovn_sb_db_tag }}"
 
+ovn_sb_db_relay_image: "{{ docker_registry ~ '/' if docker_registry else '' }}{{ docker_namespace }}/ovn-sb-db-relay"
+ovn_sb_db_relay_tag: "{{ ovn_tag }}"
+ovn_sb_db_relay_image_full: "{{ ovn_sb_db_relay_image }}:{{ ovn_sb_db_relay_tag }}"
+
 ovn_northd_dimensions: "{{ default_container_dimensions }}"
 ovn_nb_db_dimensions: "{{ default_container_dimensions }}"
 ovn_sb_db_dimensions: "{{ default_container_dimensions }}"
+ovn_sb_db_relay_dimensions: "{{ default_container_dimensions }}"
 
 ovn_northd_default_volumes:
   - "{{ node_config_directory }}/ovn-northd/:{{ container_config_directory }}/:ro"
@@ -58,11 +72,16 @@ ovn_sb_db_default_volumes:
   - "/etc/localtime:/etc/localtime:ro"
   - "ovn_sb_db:/var/lib/openvswitch/ovn-sb/"
   - "kolla_logs:/var/log/kolla/"
+ovn_sb_db_relay_default_volumes:
+  - "{{ node_config_directory }}/ovn-sb-db-relay{{ ('-' + item | string) if item is defined }}/:{{ container_config_directory }}/:ro"
+  - "/etc/localtime:/etc/localtime:ro"
+  - "kolla_logs:/var/log/kolla/"
 
 ovn_db_extra_volumes: "{{ default_extra_volumes }}"
 ovn_northd_extra_volumes: "{{ ovn_db_extra_volumes }}"
 ovn_nb_db_extra_volumes: "{{ ovn_db_extra_volumes }}"
 ovn_sb_db_extra_volumes: "{{ ovn_db_extra_volumes }}"
+ovn_sb_db_relay_extra_volumes: "{{ ovn_db_extra_volumes }}"
 
 #####
 # OVN
@@ -75,6 +94,19 @@ ovn_openflow_probe_interval: "60"
 ovn_db_inactivity_probe: "60000"
 ovn_sb_db_inactivity_probe: "{{ ovn_db_inactivity_probe }}"
 ovn_nb_db_inactivity_probe: "{{ ovn_db_inactivity_probe }}"
+ovn_sb_db_relay_active_inactivity_probe: "{{ ovn_db_inactivity_probe | int * 2}}"
+ovn_sb_db_relay_passive_inactivity_probe: "{{ ovn_db_inactivity_probe }}"
+ovn_sb_db_relay_max_backoff: "{{ ovn_db_inactivity_probe }}"
+
+# TODO(mnnasiadka): remove that once Debian gets OVS 3.3+
+ovn_sb_db_relay_config_file_args: >-
+  {% if kolla_base_distro == 'debian' %}
+  --db-sb-relay-remote={{ ovn_sb_connection_no_relay }}
+  -- --remote=ptcp:{{ ovn_sb_db_relay_port_prefix ~ ovn_sb_db_relay_group_id }}:{{ api_interface_address | put_address_in_context('url') }}
+  {% else %}
+  --db-sb-relay-config-file=/etc/ovn/ovsdb-relay.json
+  {% endif %}
+
 # OVN startup commands
 ovn_nb_command: >-
   /usr/share/ovn/scripts/ovn-ctl run_nb_ovsdb
@@ -94,5 +126,13 @@ ovn_sb_command: >-
   --db-sb-pidfile=/run/ovn/ovnsb_db.pid
   --db-sb-file=/var/lib/openvswitch/ovn-sb/ovnsb.db
   --ovn-sb-logfile=/var/log/kolla/openvswitch/ovn-sb-db.log
+ovn_sb_relay_command: >-
+  /usr/share/ovn/scripts/ovn-ctl run_sb_relay_ovsdb
+  --db-sb-relay-ctrl-sock=/run/ovn/ovnsb_db_relay_{{ ovn_sb_db_relay_group_id }}.ctl
+  --db-sb-relay-pidfile=/run/ovn/ovnsb_db_relay_{{ ovn_sb_db_relay_group_id }}.pid
+  --db-sb-relay-sock=/run/ovn/ovnsb_db_{{ ovn_sb_db_relay_group_id }}.sock
+  --db-sb-relay-use-remote-in-db=no
+  --ovn-sb-relay-logfile=/var/log/kolla/openvswitch/ovn-sb-relay-{{ ovn_sb_db_relay_group_id }}.log
+  {{ ovn_sb_db_relay_config_file_args }}
 # Workaround: pause after restarting containers to allow for leader election.
 ovn_leader_election_pause: 5
diff --git a/ansible/roles/ovn-db/handlers/main.yml b/ansible/roles/ovn-db/handlers/main.yml
@@ -25,6 +25,20 @@
     volumes: "{{ service.volumes | reject('equalto', '') | list }}"
     dimensions: "{{ service.dimensions }}"
 
+- name: Restart ovn-sb-db-relay container
+  vars:
+    service_name: "ovn-sb-db-relay"
+    service: "{{ ovn_db_services[service_name] }}"
+  become: true
+  kolla_container:
+    action: "recreate_or_restart_container"
+    common_options: "{{ docker_common_options }}"
+    name: "{{ service.container_name }}_{{ item }}"
+    image: "{{ service.image }}"
+    volumes: "{{ service.volumes | reject('equalto', '') | list }}"
+    dimensions: "{{ service.dimensions }}"
+  loop: "{{ range(1, (ovn_sb_db_relay_count | int) + 1) | list }}"
+
 - name: Restart ovn-northd container
   vars:
     service_name: "ovn-northd"

diff --git a/ansible/roles/ovn-db/tasks/bootstrap-db.yml b/ansible/roles/ovn-db/tasks/bootstrap-db.yml
@@ -56,3 +56,20 @@
   until: check_ovn_sb_db_port is success
   retries: 10
   delay: 6
+
+- name: Wait for ovn-sb-db-relay
+  vars:
+    ovn_sb_db_relay_group_id: "{{ item }}"
+    ovn_sb_db_relay_port: "{{ ovn_sb_db_relay_port_prefix ~ ovn_sb_db_relay_group_id }}"
+  wait_for:
+    host: "{{ api_interface_address }}"
+    port: "{{ ovn_sb_db_relay_port }}"
+    connect_timeout: 1
+    timeout: 60
+  register: check_ovn_sb_db_relay_port
+  until: check_ovn_sb_db_relay_port is success
+  retries: 10
+  delay: 6
+  when:
+    - enable_ovn_sb_db_relay | bool
+  loop: "{{ range(1, (ovn_sb_db_relay_count | int) +1) }}"
diff --git a/ansible/roles/ovn-db/tasks/config-relay.yml b/ansible/roles/ovn-db/tasks/config-relay.yml
@@ -0,0 +1,35 @@
+---
+- name: Ensuring config directories exist for OVN relay containers
+  vars:
+    ovn_sb_db_relay_group_id: "{{ item }}"
+  file:
+    path: "{{ node_config_directory }}/ovn-sb-db-relay-{{ item }}"
+    state: "directory"
+    owner: "{{ config_owner_user }}"
+    group: "{{ config_owner_group }}"
+    mode: "0770"
+  become: true
+
+- name: Copying over config.json files for OVN relay services
+  vars:
+    # NOTE(mnasiadka): Used in the ovn_sb_relay_command template in defaults/main.yml
+    ovn_sb_db_relay_group_id: "{{ item }}"
+  template:
+    src: "ovn-sb-db-relay.json.j2"
+    dest: "{{ node_config_directory }}/ovn-sb-db-relay-{{ item }}/config.json"
+    mode: "0660"
+  become: true
+  notify:
+    - Restart ovn-sb-db-relay container
+
+- name: Generate config files for OVN relay services
+  vars:
+    # NOTE(mnasiadka): Used in the ovn_sb_relay_command template in defaults/main.yml
+    ovn_sb_db_relay_group_id: "{{ item }}"
+  template:
+    src: "ovsdb-relay.json.j2"
+    dest: "{{ node_config_directory }}/ovn-sb-db-relay-{{ item }}/ovsdb-relay.json"
+    mode: "0660"
+  become: true
+  notify:
+    - Restart ovn-sb-db-relay container
diff --git a/ansible/roles/ovn-db/tasks/config.yml b/ansible/roles/ovn-db/tasks/config.yml
@@ -8,6 +8,7 @@
     mode: "0770"
   become: true
   with_dict: "{{ ovn_db_services | select_services_enabled_and_mapped_to_host }}"
+  when: item.key != 'ovn-sb-db-relay'
 
 - name: Copying over config.json files for services
   template:
@@ -16,3 +17,11 @@
     mode: "0660"
   become: true
   with_dict: "{{ ovn_db_services | select_services_enabled_and_mapped_to_host }}"
+  notify:
+    - Restart {{ item.key }} container
+  when: item.key != 'ovn-sb-db-relay'
+
+- name: Ensure configuration for relays exists
+  include_tasks: config-relay.yml
+  loop: "{{ range(1, ovn_sb_db_relay_count | int + 1) | list }}"
+  when: enable_ovn_sb_db_relay | bool
diff --git a/ansible/roles/ovn-db/templates/ovn-northd.json.j2 b/ansible/roles/ovn-db/templates/ovn-northd.json.j2
@@ -1,5 +1,5 @@
 {
-    "command": "/usr/bin/ovn-northd -vconsole:emer -vsyslog:err -vfile:info --ovnnb-db={{ ovn_nb_connection }} --ovnsb-db={{ ovn_sb_connection }} --log-file=/var/log/kolla/openvswitch/ovn-northd.log --pidfile=/run/ovn/ovn-northd.pid --unixctl=/run/ovn/ovn-northd.ctl",
+    "command": "/usr/bin/ovn-northd -vconsole:emer -vsyslog:err -vfile:info --ovnnb-db={{ ovn_nb_connection }} --ovnsb-db={{ ovn_sb_connection_no_relay }} --log-file=/var/log/kolla/openvswitch/ovn-northd.log --pidfile=/run/ovn/ovn-northd.pid --unixctl=/run/ovn/ovn-northd.ctl",
     "permissions": [
         {
             "path": "/var/log/kolla/openvswitch",

diff --git a/ansible/roles/ovn-db/templates/ovn-sb-db-relay.json.j2 b/ansible/roles/ovn-db/templates/ovn-sb-db-relay.json.j2
@@ -0,0 +1,18 @@
+{
+    "command": "{{ ovn_sb_relay_command }}",
+    "config_files": [
+        {
+            "source": "{{ container_config_directory }}/ovsdb-relay.json",
+            "dest": "/etc/ovn/ovsdb-relay.json",
+            "owner": "openvswitch",
+            "perm": "0600"
+        }
+    ],
+    "permissions": [
+        {
+            "path": "/var/log/kolla/openvswitch",
+            "owner": "root:root",
+            "recurse": true
+        }
+    ]
+}
diff --git a/ansible/roles/ovn-db/templates/ovsdb-relay.json.j2 b/ansible/roles/ovn-db/templates/ovsdb-relay.json.j2
@@ -0,0 +1,20 @@
+
+{
+    "remotes": {
+        "ptcp:{{ ovn_sb_db_relay_port_prefix ~ ovn_sb_db_relay_group_id }}:{{ api_interface_address | put_address_in_context('url') }}": {
+            "inactivity-probe": {{ ovn_sb_db_relay_passive_inactivity_probe }}
+        }
+    },
+    "databases": {
+        "OVN_Southbound": {
+            "service-model": "relay",
+            "source": {
+                "{{ ovn_sb_connection_no_relay }}": {
+                    "inactivity-probe": {{ ovn_sb_db_relay_active_inactivity_probe }},
+                    "max-backoff": {{ ovn_sb_db_relay_max_backoff }}
+                }
+            }
+        }
+    }
+}
+
diff --git a/ansible/roles/service-check-containers/tasks/iterated.yml b/ansible/roles/service-check-containers/tasks/iterated.yml
@@ -0,0 +1,36 @@
+---
+- name: "{{ kolla_role_name | default(project_name) }} | Check containers with iteration"
+  become: true
+  vars:
+    service: "{{ outer_item.value }}"
+  kolla_container:
+    action: "compare_container"
+    common_options: "{{ docker_common_options }}"
+    name: "{{ service.container_name }}"
+    image: "{{ service.image | default(omit) }}"
+    volumes: "{{ service.volumes | default(omit) }}"
+    dimensions: "{{ service.dimensions | default(omit) }}"
+    tmpfs: "{{ service.tmpfs | default(omit) }}"
+    volumes_from: "{{ service.volumes_from | default(omit) }}"
+    privileged: "{{ service.privileged | default(omit) }}"
+    cap_add: "{{ service.cap_add | default(omit) }}"
+    environment: "{{ service.environment | default(omit) }}"
+    healthcheck: "{{ service.healthcheck | default(omit) }}"
+    ipc_mode: "{{ service.ipc_mode | default(omit) }}"
+    pid_mode: "{{ service.pid_mode | default(omit) }}"
+    security_opt: "{{ service.security_opt | default(omit) }}"
+    labels: "{{ service.labels | default(omit) }}"
+    command: "{{ service.command | default(omit) }}"
+    cgroupns_mode: "{{ service.cgroupns_mode | default(omit) }}"
+  loop:
+    - "{{ range(1,(service.iterate_var | int) + 1) | list }}"
+  register: container_check
+
+# NOTE(yoctozepto): Must be a separate task because one cannot see the whole
+# result in the previous task and Ansible has a quirk regarding notifiers.
+# For details see https://github.com/ansible/ansible/issues/22579
+- name: "{{ kolla_role_name | default(project_name) }} | Notify handlers to restart containers"
+  debug:
+    msg: Notifying handlers
+  changed_when: container_check is changed
+  notify: "Restart {{ outer_item.key }} container"
diff --git a/ansible/roles/service-check-containers/tasks/main.yml b/ansible/roles/service-check-containers/tasks/main.yml
@@ -25,13 +25,26 @@
     command: "{{ service.command | default(omit) }}"
     cgroupns_mode: "{{ service.cgroupns_mode | default(omit) }}"
   with_dict: "{{ lookup('vars', (kolla_role_name | default(project_name)) + '_services') | select_services_enabled_and_mapped_to_host }}"
+  when: not (service.iterate | default(False)) | bool
   register: container_check
 
 # NOTE(yoctozepto): Must be a separate task because one cannot see the whole
 # result in the previous task and Ansible has a quirk regarding notifiers.
 # For details see https://github.com/ansible/ansible/issues/22579
 - name: "{{ kolla_role_name | default(project_name) }} | Notify handlers to restart containers"
+  vars:
+    service: "{{ item.value }}"
   debug:
     msg: Notifying handlers
   changed_when: container_check is changed
+  when: not (service.iterate | default(False)) | bool
   notify: "{{ container_check.results | select('changed') | map(attribute='item.key') | map('regex_replace', '^(.*)$', 'Restart \\1 container') | list }}"
+
+- name: Include tasks
+  vars:
+    service: "{{ outer_item.value }}"
+  include_tasks: iterated.yml
+  loop: "{{ lookup('vars', (kolla_role_name | default(project_name)) + '_services') | select_services_enabled_and_mapped_to_host | dict2items }}"
+  loop_control:
+    loop_var: outer_item
+  when: (service.iterate | default(False)) | bool
diff --git a/doc/source/reference/networking/neutron.rst b/doc/source/reference/networking/neutron.rst
@@ -231,6 +231,33 @@ To change this behaviour you need to set the following:
 
    neutron_ovn_distributed_fip: "yes"
 
+By default, the number of relay groups (``ovn_sb_db_relay_count``) is computed
+by dividing the total number of ``ovn-controller`` hosts by the value in
+``ovn_sb_db_relay_compute_per_relay`` (which defaults to 50), and rounding up.
+For instance, if you have 120 hosts in the ``ovn-controller`` group, you would
+get ``ceil(120 / 50) = 3`` relay groups.
+You can override ``ovn_sb_db_relay_compute_per_relay`` to scale how many hosts
+each relay group handles, for example:
+
+.. code-block:: yaml
+
+   ovn_sb_db_relay_compute_per_relay: 25
+
+You can also bypass the automatic calculation and manually set a fixed number
+of relay groups with ``ovn_sb_db_relay_count``:
+
+.. code-block:: yaml
+
+   ovn_sb_db_relay_count: 10
+
+.. note::
+   If you set ``ovn_sb_db_relay_count`` explicitly, it effectively overrides
+   the calculated count based on ``ovn_sb_db_relay_compute_per_relay``.
+
+It is also possible to set a static mapping between a ``ovn-controller`` host
+(network node or hypervisor) and particular OVN relay using an Ansible host_var
+``ovn_sb_db_relay_client_group_id``.
+
 Similarly - in order to have Neutron DHCP agents deployed in OVN networking
 scenario, use: