Octavia¶
This page is the starting point for Octavia, the octavia role: the containers octavia_api,
octavia_health_manager, octavia_housekeeping and octavia_worker, plus octavia_driver_agent when
enable_octavia_driver_agent (default: only with neutron_plugin_agent: ovn, group_vars/all.yml:831)
(roles/octavia/defaults/main.yml:2-57), on the control
nodes of the example inventory (inventory/multinode:165-166), selected by the tag octavia (site.yml:981-996).
Still holds in 2024.1: with octavia_auto_configure (default: true when amphora is in octavia_provider_drivers,
which it is by default, group_vars/all.yml:1323, roles/octavia/defaults/main.yml:386), deploy and reconfigure
create the amphora flavor, SSH keypair, security groups, lb-mgmt network and subnet (and a router for IPv6), run_once
on octavia-api[0] (roles/octavia/tasks/deploy.yml:4-5, roles/octavia/tasks/prepare.yml:2-161) CODE. The amphora
image and the four certificate files are still the operator's (Component dependencies
§3). New in 2024.1: the amphora provider also turns on the Octavia
jobboard by default, which needs Redis and a second database (see What changed in 2024.1).
The lists below say where the study covers how to operate it, its impact, its dependencies and the evidence, what to
check around a run and what stays open; the service card follows them.
What changed in 2024.1¶
- The jobboard is on by default with the amphora provider CODE.
enable_octavia_jobboarddefaults toenable_octaviaand'amphora' in octavia_provider_drivers(group_vars/all.yml:832), so it is on in a default Octavia deployment.octavia.confthen gets a[task_flow]section (roles/octavia/templates/octavia.conf.j2:177-189) with:jobboard_backend_hosts: theapiaddress of every host of theredisgroup (roles/octavia/templates/octavia.conf.j2:180), which is the control group in the example inventory (inventory/multinode:183-184);- the Redis password
redis_master_password, the sentinel portredis_sentinel_port(26379,group_vars/all.yml:606) and the sentinel namekolla(roles/octavia/templates/octavia.conf.j2:181-186), the name the redis role gives its sentinel master (roles/redis/templates/redis-sentinel.conf.j2:8); persistence_connectionto a new database,octavia_persistence(roles/octavia/templates/octavia.conf.j2:188,roles/octavia/defaults/main.yml:80-82).
- A precheck requires Redis for the jobboard CODE.
prechecksfails with "Redis must be enabled when using octavia jobboard" whenenable_octavia_jobboardis true andenable_redisis not (roles/octavia/tasks/precheck.yml:67-73), andenable_redisdefaults tono(group_vars/all.yml:840). A default Octavia configuration therefore failsprechecksuntil the operator setsenable_redis: yesorenable_octavia_jobboard: no(both listed, commented, inetc/kolla/globals.yml:403,etc/kolla/globals.yml:414). The assert runs only in the precheck action: the role includes<kolla_action>.yml(roles/octavia/tasks/main.yml:2) andprecheckssetskolla_action=precheck(tools/kolla-ansible:444-446).deploy,reconfigureandupgradedo not assert it: they render the[task_flow]section from theredisgroup whether or not the redis play runs, and that play runs only on hosts whereenable_redisis true (site.yml:384-396). How Octavia behaves when the jobboard points at hosts with no Redis is NOT AVAILABLE (the Octavia source is not in these sources). - A second database,
octavia_persistence, is created CODE.deployandreconfigurecreate it and its user (octavia_persistence, all privileges on the schema) inbootstrap.yml, next to the main database (roles/octavia/tasks/bootstrap.yml:18-32,roles/octavia/tasks/bootstrap.yml:54-72). These tasks do not depend on the jobboard setting: their only condition isnot use_preconfigured_databases. With ProxySQL, the user and the schema are part of Octavia's shard rules (roles/octavia/defaults/main.yml:87-101). The passwordoctavia_persistence_database_passwordis a new key ofetc/kolla/passwords.yml:140; it has no default anywhere underansible/(grep), and it is used by those tasks, the shard users andoctavia.conf(roles/octavia/defaults/main.yml:96,roles/octavia/templates/octavia.conf.j2:188). Apasswords.ymlwritten for 2023.1 does not have it (2023.1etc/kolla/passwords.ymlhas no such key). upgradenow creates that database too CODE.upgrade.ymlruns config, check-containers, then the two persistence database tasks (with the comment "Run octavia persistence db tasks for upgrade"), then the bootstrap container and the handler flush (roles/octavia/tasks/upgrade.yml:1-46). The newoctavia.confwithpersistence_connectionis written before the database exists, but the containers restart only at the flush, after the bootstrap container (roles/octavia/tasks/upgrade.yml:43-46).- The bootstrap container also runs
upgrade_persistenceCODE. WithKOLLA_BOOTSTRAP, theoctavia-apiimage runsoctavia-db-manage upgrade headthenoctavia-db-manage upgrade_persistence(kolladocker/octavia/octavia-api/extend_start.sh:5-9), on every deploy, reconfigure and upgrade (roles/octavia/tasks/bootstrap_service.yml:2-20). Whatupgrade_persistencedoes when the jobboard is off, and sopersistence_connectionis not rendered, is NOT AVAILABLE. kolla-ansible octavia-certificates --check-expiry <days>CODE. The option is new (tools/kolla-ansible:207,tools/kolla-ansible:411-414); it adds-e octavia_certs_check_expiry=yes -e octavia_certs_expiry_limit=<days>(tools/kolla-ansible:550-555). Theoctavia-certificatesrole then only readsserver_ca.cert.pem,client_ca.cert.pemandclient.cert-and-key.pemin/etc/kolla/config/octavia/on the Ansible host and asserts that each one is still valid in<days>days (roles/octavia-certificates/tasks/main.yml:10-12,roles/octavia-certificates/tasks/check_expiry.yml:2-24); the generation and copy block is skipped (roles/octavia-certificates/tasks/main.yml:14-51). The CA private keyserver_ca.key.pemis not checked. Without the option the role behaves as before; itsopensslsteps carrycreates:, so existing files in the work directory are kept (roles/octavia-certificates/tasks/server_ca.yml:9,roles/octavia-certificates/tasks/server_ca.yml:29). The default validity is 365 days for the client certificate and 3650 days for both CAs (roles/octavia-certificates/defaults/main.yml:22,roles/octavia-certificates/defaults/main.yml:30,roles/octavia-certificates/defaults/main.yml:38).- Other changes to
octavia.confand the containers CODE:- notifications:
driver = messagingv2andtopics = notificationsonly with Ceilometer, otherwisedriver = noop(roles/octavia/templates/octavia.conf.j2:123-130,roles/octavia/defaults/main.yml:276-280); 2023.1 wrote nodriverline; - RabbitMQ:
rabbit_quorum_queue = truewith the default quorum queues (roles/octavia/templates/octavia.conf.j2:147-148, F17); withom_enable_queue_manager,use_queue_manager, an[oslo_concurrency]lock path and a/dev/shmmount on all five containers (roles/octavia/templates/octavia.conf.j2:133-138,roles/octavia/templates/octavia.conf.j2:191-194,roles/octavia/defaults/main.yml:192); - HAProxy: the external frontend takes
external_fqdnandoctavia_api_public_port(roles/octavia/defaults/main.yml:19-26), which follows the single external frontend option (group_vars/all.yml:535); get_resources_info.ymlandoctavia.conf.j2no longer fall back to theopenstack_*result keys of olderos_*_infomodules (roles/octavia/tasks/get_resources_info.yml:72-75,roles/octavia/templates/octavia.conf.j2:105-107).
- notifications:
- Still holds in 2024.1 CODE: the five handlers, in the same order, with the module renamed
kolla_containerand skipped whenkolla_actionisconfig(roles/octavia/handlers/main.yml:2-79); the play still takesserialfromkolla_serialand now also setsmax_fail_percentage, default 100 (site.yml:989-993), which never stops the play early (Controlled batches §1.3). The first 2024.1 check of a container whose unit 2023.1 wrote reports drift (F16).
Where to look¶
Operate¶
No operation page gives a command line for the octavia tag, and the tag has no row of its own in Tags
§3 or in the Tag index. What the
study says about running it is in fact sheet D: the plays the tag selects
(§2.1) and the task flow per action
(§2.3). The certificate expiry
check is the command kolla-ansible octavia-certificates --check-expiry <days> (above); it runs on localhost only
(ansible/octavia-certificates.yml:1-5).
Dependencies¶
- Dependencies: Component dependencies §3
- Redis (new in 2024.1) CODE: with the default jobboard, Octavia's config lists the
redishosts (roles/octavia/templates/octavia.conf.j2:180) andprechecksrequiresenable_redis(roles/octavia/tasks/precheck.yml:67-73). Insite.ymlthe redis play comes before the octavia play (site.yml:384-396,site.yml:981-996).
Evidence¶
- Fact sheet: fact sheet D §2
- Image side: Kolla images
§4 (row
octavia-api), Kolla images §5
Before and after a run¶
- Before
prechecks,deployorupgradewith the amphora provider, decide on the jobboard: setenable_redis: yes(Redis is then deployed by its own play) orenable_octavia_jobboard: no; and make surepasswords.ymlhasoctavia_persistence_database_password(see What changed in 2024.1) CODE. - fact sheet D §2.5, "OpenStack resources: Kolla creates none": still not true in 2024.1 with the
default
octavia_auto_configure. The role then creates the flavor (amphora, 1 vCPU, 1024 MB, 5 GB), the keypair, the security groups,lb-mgmt-net/lb-mgmt-subnet(10.1.0.0/24by default) and, for IPv6,lb-mgmt-router, andoctavia.conftakesamp_boot_network_list,amp_secgroup_listandamp_flavor_idfrom what it looked up (roles/octavia/tasks/prepare.yml:2-161,roles/octavia/tasks/get_resources_info.yml:2-75,roles/octavia/tasks/config.yml:2-3,roles/octavia/templates/octavia.conf.j2:91-109,roles/octavia/defaults/main.yml:319-324,roles/octavia/defaults/main.yml:369-382) CODE. It does not create the amphora image (amp_image_tag = amphora,roles/octavia/templates/octavia.conf.j2:89,roles/octavia/defaults/main.yml:295). The certificates come fromkolla-ansible octavia-certificatesor the operator; a precheck fails when one of the four files is missing from/etc/kolla/config/octavia/(roles/octavia/tasks/precheck.yml:40-54). Withoctavia_auto_configure: no, the fouramp_*options are written only if the operator defines them (roles/octavia/templates/octavia.conf.j2:91-103), with no precheck. octavia_auto_configureand older upgrades: the comment that says to set it tonowhen upgrading from Ussuri is still there (group_vars/all.yml:1320-1322), but 2024.1 upgrades fromopenstack_previous_release_name: "2023.2"(group_vars/all.yml:688). With it left at its default,upgraderunsconfig.yml, whose first step looks up the flavor, project, security groups and network by name (roles/octavia/tasks/upgrade.yml:2,roles/octavia/tasks/config.yml:2-3,roles/octavia/tasks/get_resources_info.yml:2-75) CODE; what happens when resources created by hand have other names is NOT AVAILABLE (theos_*modules are not in these sources).- fact sheet D §2.6, health-manager network requirements: every octavia host (api, worker and housekeeping
too) must have
octavia_network_interfacewith an address, or templating fails (roles/octavia/templates/octavia.conf.j2:80-82,group_vars/all.yml:292,group_vars/all.yml:317). With the defaultoctavia_network_type: provider(group_vars/all.yml:1331), the interface defaults toapi_interfaceand Kolla creates nothing on the controllers. Withoctavia_network_type: tenant(andoctavia_auto_configure), the interface defaults too-hm0, and on each health-manager host Kolla creates a Neutron port, adds it tobr-intwith<engine> exec openvswitch_vswitchd ovs-vsctl, and installs and starts anoctavia-interfacesystemd unit (roles/octavia/tasks/config-host.yml:2-6,roles/octavia/tasks/hm-interface.yml:2-100); a precheck rejectstenantunlessneutron_plugin_agentisopenvswitch(roles/octavia/tasks/precheck.yml:56-65) CODE. - fact sheet D §2.3:
reconfigure=deploy, so the database creation (now two databases), the DB sync, keystone registration and, withoctavia_auto_configure, the resource creation re-run;upgraderuns neitherregister.ymlnorprepare.ymlnorconfig-host.yml(roles/octavia/tasks/reconfigure.yml:2,roles/octavia/tasks/deploy.yml:2-19,roles/octavia/tasks/upgrade.yml:1-46) CODE. - fact sheet C §4.2: enabling or disabling a service that has a horizon plugin (designate, octavia,
magnum, …) changes the horizon container environment. For Octavia this is
ENABLE_OCTAVIA, fromenable_horizon_octavia, which followsenable_octavia(roles/horizon/defaults/main.yml:20,group_vars/all.yml:785); the horizon check compares the environment (roles/horizon/tasks/check-containers.yml:11,module_utils/kolla_container_worker.py:84), so the horizon container is recreated on the next deploy/reconfigure even if no horizon config changed CODE.
Known limits and open points¶
- Known bugs and limitations, row
octavia: both Ussuri items stay gone in 2024.1.upgrade.ymlhas no "Removing octavia user from admin project" task (roles/octavia/tasks/upgrade.yml:1-46), and amphora resources other than the image are created whenoctavia_auto_configureis on (above). Status: CODE. - New in 2024.1: the Redis requirement of the jobboard is checked only by
prechecks; anupgradeordeployrun withoutprechecksrenders a jobboard configuration whether or not Redis is deployed (roles/octavia/tasks/precheck.yml:67-73,roles/octavia/templates/octavia.conf.j2:177-189) CODE. The effect on the Octavia services is NOT AVAILABLE. - fact sheet D §2.6: adding or removing a health-manager host changes octavia.conf on all octavia hosts
and restarts them (
controller_ip_port_listloops overoctavia-health-manager,roles/octavia/templates/octavia.conf.j2:82) CODE. NOT AVAILABLE: whether the controller list is written into an amphora at boot, so that existing amphorae keep the old list until failover. Likewise, a change of theredisgroup changesjobboard_backend_hosts(roles/octavia/templates/octavia.conf.j2:180) and so octavia.conf on all octavia hosts CODE. - fact sheet D
§7, item
5, read against 2024.1:
heartbeat_key = insecureis still hard-coded (roles/octavia/templates/octavia.conf.j2:81); the amp network, secgroup and flavor are filled in by default (octavia_auto_configure) and stay unchecked only with it off; the admin-role removal on upgrade is gone; the HM precheck still tests the port onapi_interface_addresswhile the health manager bindsoctavia_network_interface_address(roles/octavia/tasks/precheck.yml:29-38,roles/octavia/templates/octavia.conf.j2:80), which differ only whenoctavia_network_interfaceis notapi_interface(e.g.tenant); the CA private key (server_ca.key.pem) is still copied to worker, housekeeping and health-manager (roles/octavia/tasks/config.yml:120-171) CODE. New in 2024.1: the jobboard section writesssl:Falsefor both the Redis backend and the sentinel (roles/octavia/templates/octavia.conf.j2:185,roles/octavia/templates/octavia.conf.j2:187) CODE. - fact sheet D §2.5: if
enable_barbican, the region and endpoint for the cert manager are set inoctavia.conf(roles/octavia/templates/octavia.conf.j2:21-25). NOT AVAILABLE: whether TLS-terminated listeners need barbican. - fact sheet D §8: NOT AVAILABLE: Octavia amphora behaviour after a change in the health-manager host list.
- fact sheet D
§7, item
15: Barbican and other bootstraps (octavia, manila, ceilometer, masakari, zun, cyborg) re-run their DB sync container
on every deploy/reconfigure. For Octavia this now includes
upgrade_persistence(kolladocker/octavia/octavia-api/extend_start.sh:5-9) CODE. - F5, "
reconfigure=deploy": a reconfigure re-runs Keystone registration, DB creation/grants and the DB-sync bootstrap containers of most services CODE. - What is not known: fact sheet D has its own NOT AVAILABLE list, and its §2.3, §2.5 and §2.6 are among the sections that contain NOT AVAILABLE marks.
- Open questions and Findings have no entry on
Octavia. What the amphora resources created by
prepare.ymldo when they already exist with other settings (theos_*modules run withstate: present,roles/octavia/tasks/prepare.yml:12) is NOT AVAILABLE: the modules are not in these sources.
Service card¶
The rows of the card are explained on the Components page.
| Item | Value |
|---|---|
| Play | site.yml:981 |
| Tags | octavia |
| Play host patterns | octavia-api, octavia-health-manager, octavia-housekeeping, octavia-worker |
| Node types in play (example inventory) | control |
| Node types running its containers | control |
| Containers | octavia_api, octavia_driver_agent, octavia_health_manager, octavia_housekeeping, octavia_worker |
deploy/reconfigure flow (tasks/) |
register → prepare (if) → config-host → config → check-containers → clone (if) → bootstrap → flush handlers |
| Task files | bootstrap.yml, bootstrap_service.yml, check-containers.yml, check.yml, clone.yml, config-host.yml, config.yml, config_validate.yml, copy-certs.yml, deploy-containers.yml, deploy.yml, get_resources_info.yml, hm-interface.yml, loadbalancer.yml, main.yml, openrc.yml, precheck.yml, prepare.yml, pull.yml, reconfigure.yml, register.yml, stop.yml, upgrade.yml |
| Templates | octavia-api.json.j2, octavia-driver-agent.json.j2, octavia-health-manager.json.j2, octavia-housekeeping.json.j2, octavia-interface.service.j2, octavia-openrc.sh.j2, octavia-ssh-key.j2, octavia-worker.json.j2, octavia-wsgi.conf.j2, octavia.conf.j2 |
| Handlers (definition order) | Restart octavia-api container; Restart octavia-driver-agent container; Restart octavia-health-manager container; Restart octavia-housekeeping container; Restart octavia-worker container |
| Key variables | enable_octavia, octavia_amp_*, octavia_network_interface |
| Role dependencies | none |
Related¶
- fact sheet D §2 — the per-role evidence: plays, services, task flow, config files and handlers, health-manager network requirements
- Horizon — octavia is one of the services with a horizon plugin: enabling or disabling it recreates the horizon container on the next deploy/reconfigure
- Barbican — if
enable_barbican,octavia.confgets the region and endpoint for the cert manager - Component dependencies §3 — Octavia's config references Nova, Neutron and Glance, and in 2024.1 Redis for the jobboard
- Execution flow §2.2 — Octavia is near the end of the deployment order
- Kolla images §4 — what the
bootstrap_octaviacontainer runs, includingupgrade_persistence - Components — the other component pages