Skip to content

Method and limits of the study

A source-level study of Kolla-Ansible 2024.1-eom (openstack/kolla-ansible, tag 2024.1-eom, 18.x series, commit 890396f). It covers how Kolla-Ansible orchestrates an OpenStack platform, what each tag really executes, how runs scale with the number of compute nodes, and what Kolla-Ansible commands do to a running production platform.

The pages were re-studied from the earlier study of 2023.1-eol: every 2023.1 citation was looked up in the 2024.1 sources and every statement was re-read there. Nothing is carried over from 2023.1, from newer releases or from documentation that the code does not confirm. Each conclusion is labelled with its evidence level: CODE or NOT AVAILABLE (see Evidence labels).

Note

Key answers

  • Sources: kolla-ansible, kolla and ansible-collection-kolla at 2024.1-eom, and ansible-core 2.16.20 (Sources).
  • How the 2024.1 pages were made: each 2023.1 page gave the structure and the questions; each of its path:line citations was looked up in the 2024.1 sources and every statement was re-read there, guided by a map of the changes between the two trees (Method).
  • Limits of the study: source code only and no timings; every statement about data-plane or VM impact is NOT AVAILABLE. The Podman paths new in 2024.1 were read in the code only, and ansible-core 2.15 was not read (Limits of this study).

Sources

Source Version
kolla-ansible tag 2024.1-eom (18.x series, commit 890396f); openstack_release: "2024.1" (group_vars/all.yml:679)
kolla (image repository) tag 2024.1-eom (commit ed7a36a)
ansible-collection-kolla tag 2024.1-eom (commit 82a0f5c); kolla-ansible's requirements.yml:5 pins the stable/2024.1 branch
ansible-core 2.16.20. tools/kolla-ansible:10-11 accepts 2.15 to 2.16; only 2.16.20 was read, so whether 2.15 behaves the same is NOT AVAILABLE

Changed from 2023.1: the 2023.1 study read the 2023.1-eol tags (kolla-ansible commit 4891a21, kolla de82941, collection pinned to unmaintained/2023.1) and ansible-core 2.14.18. The 2023.1 sources (kolla-ansible, kolla and ansible-collection-kolla 2023.1-eol) were read only to see what changed, mostly with diff -ru of the two trees. No statement about 2024.1 rests on them. ansible-core 2.14.18 was not on disk for this study, so no ansible-core diff between the two releases was made.

Method

Phase What was done
1. Source exploration (earlier studies) The Ussuri study read tools/kolla-ansible, site.yml, nova.yml, gather-facts.yml, group_vars/all.yml, the container and toolbox modules, merge_configs, filters, and the common, nova, nova-cell, neutron, openvswitch and service-* helper roles in full, plus the kolla image start path and one-shot containers; the 2023.1 study re-read them in 2023.1-eol. These phases set the structure and the questions of the pages, not their answers
2. Change map Before the pages were written, the two trees were read and diffed in the parts of the code that decide what Day-2 commands do. The result is a working map of the cross-cutting changes 2023.1 → 2024.1 (research/2024.1/change-map.md, items X1–X18, each with 2024.1 citations) and of the 2023.1 pages each change touches. It is used as a map only: every statement of a page is re-verified in the 2024.1 code
3. Re-study Each page starts from its 2023.1 page for its structure and its questions, never for its answers. Every 2023.1 path:line citation was looked up in the 2024.1 sources and every statement was re-read there, with its own 2024.1 citation; statements without a citation (commands, counts, lists of tags, containers and hosts) were re-read too, since the same lines do not prove the same behaviour. A statement that no longer holds says so, with the code; one that could not be re-verified was dropped or labelled NOT AVAILABLE. The main structural changes were each checked in the code, for example: kolla_docker → kolla_container, which loads a Docker or a Podman worker after container_engine (library/kolla_container.py:410-420), fed from kolla_container_engine (group_vars/all.yml:92,153), so the 2023.1 statement "Docker only" no longer holds; a new systemd unit template (module_utils/kolla_systemd_worker.py:20-39); max_fail_percentage on every play and no serial on the etcd play (site.yml:470-482); RabbitMQ quorum queues on by default (group_vars/all.yml:268); removed services such as murano, sahara and monasca (no such directories under ansible/roles/); ansible-core 2.14 → 2.16
4. Kolla image sources Re-read in kolla 2024.1-eom: the base start path and the start scripts of the images listed in Kolla images §1, with diff -ru of docker/base against 2023.1-eol
5. Generated material Regenerated from the 2024.1 tree, not edited by hand: Generated inventory (tools/extract_kolla_inventory.py; 71 plays, 159 containers), the Ansible listings (ansible-core 2.16.20) and the service cards of the component pages (tools/service_cards.py, run with the multinode inventory parsed by ansible-core 2.16.20: 22 cards in 20 pages). The 2023.1 listings, ansible-core 2.14.18 output, do not carry over
6. Dependency, performance and operational analysis Re-done on 2024.1 from the role sources. Risk IDs, finding IDs, question IDs and anchors are kept; an item that no longer holds in 2024.1 keeps its entry and says so, with the code
7. Validation ansible-core 2.16.20: --list-tags/--list-tasks listings of the unmodified playbooks (Ansible listings); play and batch semantics read in the ansible-core 2.16.20 source, for example that max_fail_percentage counts the failures of the current batch only (lib/ansible/executor/playbook_executor.py:184-190, lib/ansible/executor/task_queue_manager.py:330-338). Each page is checked by a citation checker that flags line ranges beyond the end of the cited file and citations whose text is identical to the 2023.1 page while the cited lines changed; each flagged citation was corrected or confirmed against the 2024.1 file

Limits of this study

  • Source code only. The study rests only on reading source code: kolla-ansible, kolla and ansible-collection-kolla 2024.1-eom, and the ansible-core 2.16.20 source. The only Ansible output used is its --list-tags/--list-tasks parse of the playbooks (Ansible listings), produced with the multinode example inventory, Kolla-Ansible 2024.1 installed in a virtual environment and ansible-collection-kolla 2024.1-eom on the collections path; it runs nothing against a host. The semantics of kolla_set_configs --check are read from the kolla source (Kolla images §3).
  • No runtime or lab evidence, from either release. The 2023.1 study used none either. Its Ansible listings were ansible-core 2.14.18 output and do not carry over: they were regenerated with ansible-core 2.16.20 from the 2024.1 tree. The subjects that only a run could settle stay code-only or NOT AVAILABLE in 2024.1:
    • signal delivery to services behind a shell wrapper that does not exec them: NOT AVAILABLE (R24, Kolla images §2.6);
    • kolla_set_configs and kolla_container behaviour (check exit codes, restart and recreate): code-only (F14, Kolla images §3, F3);
    • libvirt/QEMU and Open vSwitch behaviour across container restarts: NOT AVAILABLE (Q1, Q2, Q3);
    • task-result counts per compute, restart timings and wall time at any number of computes: NOT AVAILABLE; only work counted from the code is given (Compute scaling, Q7);
    • --limit fact gathering and run_once repeats with serial: code-only (F9, F10).
  • There are no timings. No run is used, so no wall time is known; work can only be counted from the code (Compute scaling).
  • No container or VM behaviour is used as evidence. Every statement about data-plane or VM impact is NOT AVAILABLE.
  • The Podman paths were read in the code only. New in 2024.1, kolla_container_engine: podman selects the Podman worker (library/kolla_container.py:415-420), which talks to the Podman socket through the Podman Python client (module_utils/kolla_podman_worker.py:21,73). That worker and the Podman branches of the roles were read; the Podman Python client, Podman itself and systemd were not, and no Podman host was used. What they do on a host is NOT AVAILABLE.
  • The depth of reading varies. The core services are traced in the pages of the study other than the fact sheets; the other roles are covered by the fact sheets only.
  • Only one ansible-core version was read. Kolla-Ansible 2024.1 also accepts ansible-core 2.15 (tools/kolla-ansible:10-11); its behaviour is NOT AVAILABLE. Nothing can be said from code about what changed in ansible-core itself between 2.14 and 2.16, since 2.14.18 was not on disk.
  • The galaxy collections were not on disk. Only ansible-collection-kolla (openstack.kolla) was read and put on the collections path. When the ansible package is not installed, kolla-ansible install-deps also installs requirements-core.yml (tools/kolla-ansible:131-138), which pins galaxy collections such as community.docker <4 and containers.podman <2 (requirements-core.yml:18-23). Their modules were not read, so what they do is NOT AVAILABLE.
  • The kolla image sources were read, not built or run. From the kolla 2024.1-eom repository: the base start path (docker/base: Dockerfile.j2, start.sh, set_configs.py, copy_cacerts.sh, httpd_setup.sh, sudoers, healthcheck scripts) and the start scripts of the images listed in Kolla images (Nova, Neutron, Open vSwitch, OVN, Keystone, Glance, Placement, Cinder, Heat, Horizon, Barbican, Octavia, Manila, Swift, MariaDB, RabbitMQ, memcached, HAProxy, keepalived incl. check_alive.sh, fluentd, cron, kolla-toolbox, iscsid, tgtd; there is still no chrony image). Of the other images only the ironic-api bootstrap branch and the bifrost-deploy ENTRYPOINT/CMD were read, and every Dockerfile was searched for HEALTHCHECK and for ENTRYPOINT/CMD overrides. What the scripts do against real services and data is NOT AVAILABLE.

To remember

Lumo: if you keep only 5 things from this page, keep these.

  • The study rests only on reading source code: no runtime or lab result is used, from 2023.1 or from 2024.1.
  • Every statement was re-checked against the 2024.1 sources; nothing is carried over from 2023.1, from newer releases or from documentation the code does not confirm.
  • There are no timings: no run is used, so no wall time is known; work can only be counted from the code.
  • The Podman paths that are new in 2024.1 were read in the code only; how Podman and systemd behave on a host is not available. NOT AVAILABLE
  • Only ansible-core 2.16.20 was read: Kolla-Ansible 2024.1 also accepts 2.15, whose behaviour is not available. NOT AVAILABLE

Check yourself

Lumo: 3 quick questions to check the main ideas of this page. Choose an answer to see the explanation and the section that gives the evidence. Your score counts your first choices and stays in this browser only.

1. Was any run against a host used as evidence for the 2024.1 pages?
Show the answer

Answer: No: the study rests only on reading source code. The only Ansible output used is the ansible-core 2.16.20 --list-tags/--list-tasks parse of the playbooks, which runs nothing against a host. No runtime or lab result is used, from 2023.1 or from 2024.1. Read the section.

2. What does the study tell you about how a deployment with kolla_container_engine: podman behaves on a host?
Show the answer

Answer: Only what the code of the Podman worker and the roles says; behaviour on a host is NOT AVAILABLE. The Podman paths were read in Kolla-Ansible's own code only. The Podman Python client and the containers.podman collection were not on disk, and nothing was run, so what Podman and systemd do on a host is NOT AVAILABLE. Read the section.

3. How were statements from the 2023.1 study treated for 2024.1?
Show the answer

Answer: Every 2023.1 citation was looked up in the 2024.1 sources and every statement re-read; nothing is carried over. The 2023.1 pages give only the structure and the questions. Each 2023.1 citation was looked up in the 2024.1 sources and every statement re-read there, since the same lines do not prove the same behaviour. Read the section.