Cinder¶
This page is the starting point for Cinder, the cinder role: the containers cinder_api, cinder_scheduler,
cinder_volume and cinder_backup, on the control and storage nodes of the example inventory, selected by the tag
cinder. iscsid and tgtd are in the iscsi play, not under -t cinder (Tags
§3.1). 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.
Where to look¶
Operate¶
No operation page gives a command line for the cinder tag. The scenario above describes
kolla-ansible -t <svc> reconfigure for an API service such as keystone, glance, cinder, heat or placement; its command
example is keystone (Reconfigure a single service
§1.1).
Impact and risk¶
Dependencies¶
- Dependencies: Component dependencies §3
Evidence¶
- Fact sheet: fact sheet C §3
- Image side: Kolla images
§4 (row
cinder-api), Kolla images §5
Before and after a run¶
- Pre-execution checklist, step 6 (workload gates): no live migrations,
resizes, image uploads or volume migrations in flight on the targeted hosts; one of its two checks is
openstack volume list --status *ing. - Tags §3.1, row
<svc>API service: safe on its own "yes, it is the intended unit; it still runs Keystone registration and DB sync"; batch withANSIBLE_SERIAL. Prerequisites: MariaDB, RabbitMQ, memcached, Keystone (registration), HAProxy VIP. Glance and cinder transfers in progress are SIGKILLed after 60 s (docker_graceful_timeout,group_vars/all.yml:140,152;module_utils/kolla_systemd_worker.py:33; R24). - fact sheet C §3.8, item 7:
the cinder role itself does not target computes; enabling
enable_cinder_backend_iscsi/LVM also flips glance-api to privileged with/dev, so glance-api is recreated. In 2023.1cinder_backend_cephalso makes glance-api privileged, without/dev(roles/glance/defaults/main.yml:10,192-193) CODE. - R8, precautions for a HAProxy config change:
ANSIBLE_SERIAL=1for haproxy changes;--skip-tags haproxywhen you don't intend to change frontends. - R24, precautions for the container stop timeout (60 s in 2023.1: the systemd unit
stops the container with
docker stop -t <docker_graceful_timeout>): drain long operations before restarts; raisedocker_graceful_timeoutif needed (it reaches the unit on the next restart). - Reconfigure a single service
§3,
verification of
-t <svc> reconfigure: before,kolla-ansible -t <svc> genconfig, thenkolla_set_configs --checkin each<svc>container and a diff of/etc/kolla/haproxy/services.d/<svc>*.cfgagainst the previous version; after,openstack endpoint list --service <name>is unchanged for each Keystone service the role registers, e.g.cinderv3for cinder (the only service incinder_ks_servicesin 2023.1,roles/cinder/defaults/main.yml:319-325), the API answers through the VIP, the container start times show which hosts restarted, anddocker ps --filter name=haproxyshows whether HAProxy restarted.
Known limits and open points¶
- Known bugs and limitations, row
cinder: nohost/backend_host/cluster; etcd coordination uses only the first etcd host; Ceph backup driver default without a Ceph config. Status: CODE. Checked against 2023.1:cinder.conf.j2still sets nohost,backend_hostorcluster; the etcd item no longer holds as written: withcinder_coordination_backend: etcd(the default only when Redis is disabled and etcd enabled,group_vars/all.yml:951) thebackend_urlisetcd3+<proto>://kolla_internal_fqdn:etcd_client_port, i.e. the internal VIP, where the etcd role has a HAProxy frontend (roles/cinder/templates/cinder.conf.j2:245-253,roles/etcd/defaults/main.yml:26-32); the backup driver still defaults toceph(group_vars/all.yml:954,roles/cinder/templates/cinder.conf.j2:27-30) whileceph.confis copied only withcinder_backend_ceph(roles/cinder/tasks/config.yml:15-19) CODE. - F3, "
docker_graceful_timeoutanddocker_client_timeoutare ignored": an Ussuri finding that no longer holds as written in 2023.1. The role handlers passdocker_common_options, which carriesgraceful_timeout: "{{ docker_graceful_timeout }}"(60 s) andclient_timeout(120 s) (group_vars/all.yml:105,140, 152-153), and the container's systemd unit stops it withdocker stop <name> -t <graceful_timeout>(module_utils/kolla_systemd_worker.py:33). On every restart, a service still running 60 s after SIGTERM is killed with its in-flight work (heat-engine, cinder copies, glance uploads); the 10 s value is only the module default when nograceful_timeoutis passed (module_utils/kolla_docker_worker.py:438-440,462-464) CODE (R24). - F5, "
reconfigure=deploy": a reconfigure re-runs Keystone registration, DB creation/grants and the DB-sync bootstrap containers of most services CODE. - fact sheet C §3.8,
"Operational notes", nine items: no
host/backend_host/clusterset in cinder.conf; cinder-volume restart impact; cinder-backup restart; upgrade restarts every cinder-volume and cinder-backup at once; oddity: backup driver vs. ceph files; multipath is never configured; compute-host touchpoints; external Ceph needs cinder-volume and cinder-backup hosts in thestoragegroup; the sanity check (if enabled) creates a 1 GB volume on every deploy/reconfigure. Checked against 2023.1 CODE:- the upgrade still restarts every cinder container of the play at once (handlers,
kolla_serial0 by default,site.yml:519-530), and then a second time: after the flush,reload.ymlrestarts every enabled cinder container "to remove RPC version pin", after acinder_rpc_version_startup_delaypause (30 s) that only runs onupgrade; the online data migration runs last (roles/cinder/tasks/upgrade.yml:21-44,roles/cinder/tasks/reload.yml:2-9,roles/cinder/handlers/main.yml:78-88,roles/cinder/defaults/main.yml:229); - the cinder role still sets no multipath option (no
multipathinroles/cinder/); 2023.1 has a separatemultipathdplay, off by default (site.yml:428,group_vars/all.yml:714); external_ceph.ymlcopiesceph.confand keyrings to the hosts of each service's own group (cinder-volume,cinder-backup), with no reference to thestoragegroup (roles/cinder/tasks/external_ceph.yml:13-15,30);- the sanity check does not exist in 2023.1: the role has no
check.ymlandtasks/deploy.ymlhas no check step (roles/cinder/tasks/deploy.yml:1-14).
- the upgrade still restarts every cinder container of the play at once (handlers,
- fact sheet C §8,
"NOT AVAILABLE list", items 5 to 7: NOT AVAILABLE: whether cinder-backup works with default
cinder_backup_driver=cephandcinder_backend_ceph=no; whether the volumes owned by one of several RBD cinder-volume nodes can still be managed while that node is down, and what state an in-flight create-from-image is in after a cinder-volume restart; Cinder RPC pinning afterkolla-ansible upgrade. In 2023.1 the role restarts all cinder services after an upgrade to remove the RPC version pin (roles/cinder/tasks/reload.yml:2); whether the pin is in fact lifted is service-side behaviour: NOT AVAILABLE. - fact sheet C §6: a reconfigure with a bumped image tag runs cinder's
db syncwithout its online data migration step (the bootstrap container setsKOLLA_BOOTSTRAP, which runscinder-manage db sync, notKOLLA_OSM;KOLLA_OSMis set only byupgrade.yml:roles/cinder/tasks/bootstrap_service.yml:11, kolladocker/cinder/cinder-api/extend_start.sh:6-14,roles/cinder/tasks/upgrade.yml:26-44); whether this leaves cinder in a partially upgraded state is NOT AVAILABLE. - Kolla images §6, item 10: recursive chown on data volumes at every
start:
kolla_set_configsmakes onechownper file of thecindervolume, twice for cinder-volume, on everyCOPY_ALWAYSstart (roles/cinder/templates/cinder-volume.json.j2:38-43, kolladocker/cinder/cinder-volume/extend_start.sh:3-5) CODE. - What is not known: fact sheet C has its own NOT AVAILABLE list, and its §3.8 is among the sections that contain NOT AVAILABLE marks.
- Open questions has no entry on Cinder.
Service card¶
The rows of the card are explained on the Components page.
| Item | Value |
|---|---|
| Play | site.yml:519 |
| Tags | cinder |
| Play host patterns | cinder-api, cinder-backup, cinder-scheduler, cinder-volume |
| Node types in play (example inventory) | control, storage |
| Node types running its containers | control, storage |
| Containers | cinder_api, cinder_scheduler, cinder_volume, cinder_backup |
deploy/reconfigure flow (tasks/) |
register → config → check-containers → clone (if) → bootstrap → flush handlers |
| Task files | bootstrap.yml, bootstrap_service.yml, check-containers.yml, clone.yml, config.yml, config_validate.yml, copy-certs.yml, deploy-containers.yml, deploy.yml, external_ceph.yml, loadbalancer.yml, main.yml, precheck.yml, pull.yml, reconfigure.yml, register.yml, reload.yml, stop.yml, upgrade.yml |
| Templates | cinder-api.json.j2, cinder-backup.json.j2, cinder-scheduler.json.j2, cinder-volume.json.j2, cinder-wsgi.conf.j2, cinder.conf.j2 |
| Handlers (definition order) | Restart cinder-api container; Restart cinder-scheduler container; Restart cinder-volume container; Restart cinder-backup container; Wait for cinder services to update service versions |
| Key variables | enable_cinder, enable_cinder_backend_*, cinder_backend_ceph, cinder_backup_driver, cinder_coordination_backend |
| Role dependencies | none |
Related¶
- Reconfigure a single service — what
-t <svc> reconfigureruns for an API service such as cinder, with the risks and the verification - iscsid / tgtd — the
iscsirole, on which cinder-volume (LVM) depends - Glance — enabling the cinder iSCSI/LVM or Ceph backend recreates glance-api
- Nova — Nova needs Cinder for volume-backed boots
- R24 — cinder-volume copies are among the in-flight work killed 60 s
(
docker_graceful_timeout) after SIGTERM - Pre-execution checklist — the eight checks before every production run
- Components — the other component pages