Swift¶
This page is the starting point for Swift, the swift role: swift_proxy_server, swift_rsyncd and the account,
container and object containers (18 in all; the three *_replication_server containers start only when
swift_replication_interface differs from swift_storage_interface, which it does not by default, so 15 by default:
roles/swift/tasks/start.yml:97,176,256, roles/swift/defaults/main.yml:66, group_vars/all.yml:288-289), on the
control nodes (proxy, inventory/multinode:122-123, inventory/multinode:327-328) and storage nodes (account,
container and object servers, inventory/multinode:330-337) of the example inventory, selected by the tag swift
(site.yml:511-526). The rings are operator-built
(Component dependencies §3),
and deploy does not apply swift config or ring changes to running containers: no task of the role notifies a handler,
and its only handler, Restart swift-proxy-server container, is a debug no-op (roles/swift/handlers/main.yml:2-4)
(fact sheet D §3.4). Still
holds in 2024.1 CODE. 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 reconfigure and upgrade tasks now pass
docker_common_optionsCODE. In 2023.1, the swift tasks that read the container state and environment, remove and restart containers (reconfigure), and stop containers (rolling and legacy upgrade) passed nocommon_options. In 2024.1 they pass them (roles/swift/tasks/reconfigure.yml:35,roles/swift/tasks/reconfigure.yml:62,roles/swift/tasks/reconfigure.yml:73,roles/swift/tasks/reconfigure.yml:94,roles/swift/tasks/rolling_upgrade.yml:34,roles/swift/tasks/rolling_upgrade.yml:52,roles/swift/tasks/legacy_upgrade.yml:10). What this changes in the 2024.1 code:- the container engine is
kolla_container_engine(group_vars/all.yml:153), not the module fallbackdocker(library/kolla_container.py:372); - the restart of the reconfigure goes through the container's systemd unit, which it rewrites when its text differs
(
module_utils/kolla_docker_worker.py:439-459,module_utils/kolla_systemd_worker.py:172-183). The unit is now built fromdocker_common_options:-t <docker_graceful_timeout>(60 s) andRestart=alwaysfrom the defaultunless-stoppedpolicy (group_vars/all.yml:117,group_vars/all.yml:139,module_utils/kolla_systemd_worker.py:56-58,module_utils/kolla_systemd_worker.py:69-79). Withoutcommon_options, the module falls back to a 10 s graceful timeout (library/kolla_container.py:370) and passes no restart policy, which the unit writes asRestart=no(module_utils/kolla_systemd_worker.py:56). The same fallbacks existed in the 2023.1 module (2023.1 tree:library/kolla_docker.py, lines 357 and 360;module_utils/kolla_systemd_worker.py, line 51), so a swift container restarted by the 2023.1 reconfigure may carry such a unit; what a given host holds is NOT AVAILABLE.
- the container engine is
- External frontend CODE: the external HAProxy entry now carries
external_fqdn(roles/swift/defaults/main.yml:16), used by the single external frontend map (roles/haproxy-config/tasks/main.yml:25-53). Unlike most services, Swift has no public port variable: the public endpoint keepsswift_proxy_server_port, 8080 (group_vars/all.yml:623,group_vars/all.yml:1315). - The play CODE: it still takes
serialfromkolla_serial, default0(site.yml:519), and now also setsmax_fail_percentage, default 100 (site.yml:520-523), which never stops the play early (Controlled batches §1.3). - Not affected by F16 CODE: the swift role has no
check-containers.ymland never callscompare_container(grep ofroles/swift/).deployrunsstart_container(roles/swift/tasks/deploy.yml:6-8), which recreates a container only when its parameters differ (image, volumes, environment and the othercheck_container_differsitems) and writes the unit only when the container is not running (module_utils/kolla_docker_worker.py:365-389,module_utils/kolla_container_worker.py:67-89). So the changed unit template alone does not restart a running swift container. - Everything else in the role is unchanged apart from the module rename
kolla_docker→kolla_container: templates, prechecks,start.ymlandstop.ymlhave the same tasks and line numbers. In kolla, only the base image changed: it copiesswift_api_audit_map.confinto/etc/swift/(kolladocker/swift/swift-base/Dockerfile.j2:40).
Where to look¶
Operate¶
No operation page gives a command line for the swift 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
(§3.1) and the task flow per action
(§3.4). Its reconfigure is
not deploy: swift and bifrost have their own reconfigure logic (Repository architecture
§4.2). The 2024.1 deploy flow is registration,
config, start, check (roles/swift/tasks/deploy.yml:2-10) CODE.
Dependencies¶
- Dependencies: Component dependencies §3
Evidence¶
- Fact sheet: fact sheet D §3
- Image side: Kolla images §5 (rows
swift-*)
Before and after a run¶
- fact sheet D §3.3, ring
files: Kolla does not build rings; the precheck fails unless
account|container|object.builderand.ring.gzexist under{{node_custom_config}}/swift/(roles/swift/tasks/precheck.yml:108-131) CODE. - fact sheet D §3.4,
deploy: it does not apply swift config or ring changes to running containers (see the introduction and What changed in 2024.1). - fact sheet D §3.4,
reconfigure: it fails if any listed container is not running (roles/swift/tasks/reconfigure.yml:32-42), so it cannot repair a crashed container; it runskolla_set_configs --checkin each container through<kolla_container_engine> exec -u root(roles/swift/tasks/reconfigure.yml:46-54); under COPY_ALWAYS (the default,group_vars/all.yml:43) it restarts a container when that check returns rc 1 (roles/swift/tasks/reconfigure.yml:91-105), and with the defaultkolla_serial=0every changed proxy, on every proxy host, restarts in that one task (site.yml:519) CODE. NOT AVAILABLE: whether this opens an outage window. - fact sheet D §3.4,
upgrade: by default it stops all storage containers on all storage hosts at once, restarts them, then stops and starts the proxies (swift_enable_rolling_upgrade: "yes",roles/swift/defaults/main.yml:73;roles/swift/tasks/upgrade.yml:2-3;roles/swift/tasks/rolling_upgrade.yml:31-43,roles/swift/tasks/rolling_upgrade.yml:46-68) CODE. Still holds in 2024.1, including the TODO comment that these tasks should run host by host (roles/swift/tasks/rolling_upgrade.yml:28-30). - R21, precautions for enabling Swift (or
glance_backend_ceph) on a running cloud: setglance_backend_*explicitly before enabling Swift. - R24, container stop timeout: with
common_optionsnow passed, the stops and restarts of the swift reconfigure and upgrade usedocker_graceful_timeout(60 s,group_vars/all.yml:139) through the container's unit (module_utils/kolla_systemd_worker.py:32) CODE.
Known limits and open points¶
- Known bugs and limitations, row
swift: no handlers are notified (deploy does not apply config/ring changes);stop.ymlwrong container name; precheck probes rsync port 873 vs 10873. Status: CODE, all three still in 2024.1 (roles/swift/tasks/stop.yml:23-31,roles/swift/tasks/precheck.yml:89,group_vars/all.yml:628); runtime effects NOT AVAILABLE. - What is not known: the sources do not establish the runtime effects of the swift limitations.
- Q12: "Glance backend switch on enabling Swift". Status: NOT AVAILABLE.
- F5, "
reconfigure=deploy". In 2024.1, 54 roles have atasks/reconfigure.yml: 52 rundeploy.yml(skyline throughinclude_tasks), and only swift and bifrost have their own logic (Repository architecture §4.2) CODE. - fact sheet D §3.8, "Oddities":
enable_swift_reconis tested without| bool(NOT AVAILABLE: whether the template treats that string as true and so always enables the recon middleware); theproxy-loggingandversioned_writesfilters are defined but not in the proxy pipeline. Both still in 2024.1:enable_swift_recon: "no"(group_vars/all.yml:844) is tested as{% if enable_swift_recon %}(roles/swift/templates/account.conf.j2:15,17,container.conf.j2:15,17,object.conf.j2:19,22);roles/swift/templates/proxy-server.conf.j2:13,33,92CODE. - fact sheet D
§3.5:
NOT AVAILABLE: what changing
swift_hash_path_suffix/prefixafter data is written does. Both come frompasswords.yml(etc/kolla/passwords.yml:103-104) and are rendered intoswift.conf(roles/swift/templates/swift.conf.j2:2-3) CODE. - fact sheet D
§7, items
1 to 4: the account-auditor guard of
stop.ymlchecksswift_object_auditor; the precheck checks rsync port 873, but rsyncd uses 10873;{% if enable_swift_recon %}without| boolagainst the string"no"; deploy never restarts on config or ring change, reconfigure fails if any swift container is down, and the "rolling" upgrade stops all storage at once. All four still hold in 2024.1 (citations above) CODE. - fact sheet D §8:
NOT AVAILABLE: whether the swift containers serve the new rings once
reconfigurehas restarted them; the effective value ofenable_swift_reconin the renderedobject-server.conf. - What is not known: fact sheet D has its own NOT AVAILABLE list, and its §3.4, §3.5 and §3.8 are among the sections that contain NOT AVAILABLE marks.
Service card¶
The rows of the card are explained on the Components page.
| Item | Value |
|---|---|
| Play | site.yml:511 |
| Tags | swift |
| Play host patterns | swift-account-server, swift-container-server, swift-object-server, swift-proxy-server |
| Node types in play (example inventory) | control, storage |
| Node types running its containers | control, storage |
| Containers | swift_proxy_server, swift_rsyncd, swift_account_server, swift_account_auditor, swift_account_replication_server, swift_account_replicator, swift_account_reaper, swift_container_server, swift_container_auditor, swift_container_replication_server, swift_container_replicator, swift_container_updater, swift_object_server, swift_object_auditor, swift_object_replication_server, swift_object_replicator, swift_object_updater, swift_object_expirer |
deploy/reconfigure flow (tasks/) |
register → config → start → check |
| Task files | check.yml, config.yml, config_validate.yml, copy-certs.yml, deploy-containers.yml, deploy.yml, legacy_upgrade.yml, loadbalancer.yml, main.yml, precheck.yml, pull.yml, reconfigure.yml, register.yml, rolling_upgrade.yml, start.yml, stop.yml, upgrade.yml |
| Templates | account.conf.j2, container.conf.j2, object.conf.j2, proxy-server.conf.j2, rsyncd.conf.j2, swift-account-auditor.json.j2, swift-account-reaper.json.j2, swift-account-replication-server.json.j2, swift-account-replicator.json.j2, swift-account-server.json.j2, swift-container-auditor.json.j2, swift-container-replication-server.json.j2, swift-container-replicator.json.j2, swift-container-server.json.j2, swift-container-updater.json.j2, swift-object-auditor.json.j2, swift-object-expirer.json.j2, swift-object-replication-server.json.j2, swift-object-replicator.json.j2, swift-object-server.json.j2, swift-object-updater.json.j2, swift-proxy-server.json.j2, swift-rsyncd.json.j2, swift.conf.j2 |
| Handlers (definition order) | Restart swift-proxy-server container |
| Key variables | enable_swift, swift_devices_*, swift_*_server ports |
| Role dependencies | none |
Related¶
- fact sheet D §3 — the per-role evidence: plays, services, ring files, task flow, config files, networks and ports
- Glance — enabling Swift changes Glance's default backend
- Repository architecture §4.2 — swift is one of the two roles with their own reconfigure logic
- Kolla images §5 — swift config and rings are copied only at container start
- Predict and inspect a run —
kolla_set_configs --check, which the swift reconfigure runs - Components — the other component pages