Elemental 3 in Rancher: parity status
As of Nov 18, 2026 04:42 UTC. Started Nov 12, 2026. Work is done by software agents; every result is checked against its evidence before it is recorded.
Day 7
since work started on Nov 12, 2026
17 of 52
plan tasks proven in the lab and in review
6
more tasks in progress
57
task results recorded, each checked against its evidence
What "parity" means here
Parity means a Rancher user can deploy, manage and upgrade an Elemental 3 cluster with an experience equal to or better than each solution it replaces. It is measured, not asserted.
- Baselines: Rancher provisioning v2, Elemental 2, SUSE Edge, and Harvester's own lifecycle. Each is measured in the same lab with the same harness.
- Unit of comparison: a user journey, for example "create a cluster" or "upgrade the OS and Kubernetes". There are 52 approved journeys across deploy, manage, upgrade, tenancy and migration.
- A journey passes when it completes, no node is left cordoned, and it takes no more user steps than the legacy path. Steps are the numbered actions in the official procedure. A tie counts as a pass.
- Time is checked but is not the deciding measure: up to 110% of the legacy time for the same cluster shape passes. The one-time image build for a first cluster is not counted; a prebuilt image path is.
- Elemental 3 must work with or without Rancher. No lifecycle function may depend on Rancher being present.
- Out of scope: K3s, bring-your-own OS, Windows workers, release downgrade, BIOS boot. ARM64 is in scope but not yet measured; every result so far is x86-64.
- Anything better than parity is recorded in a backlog and not built yet, so effort stays on closing gaps.
Milestones: plan against actual
Targets are the estimates made before work began. Later milestones will be re-estimated from the measured pace.
| Wave | Goal | Target | Status | Notes |
|---|---|---|---|---|
| W0 | Scorecard live; legacy baselines measured; Harvester migration spike started | Nov 18, 2026 | Met Nov 13 (5 days ahead of target) | Scorecard and both legacy baselines measured by day 2. |
| W1 | Create, scale and upgrade an Elemental 3 cluster on Harvester through the API | Dec 2, 2026 | Shown in the lab Nov 17 (15 days ahead of target) | Measured end to end by the harness on a 3 + 2 cluster. Still uses lab workarounds for the Harvester provider. |
| W2 | Same flow in the Rancher UI with tenancy | Dec 23, 2026 | Under way | Create wizard works for a tenant (day 6). Tenant can create, see, scale and delete their cluster with prototype roles (day 7). Upgrade UI and cluster detail are in progress. |
| W3 | Bare metal: registration, inventory, in-place upgrade | to be re-estimated | Not started | Not started. |
| W4 | AWS, Azure, GCP, vSphere; upgrade by replacement; air-gap; migration | to be re-estimated | Under way | Upgrade by replacement shown on Harvester (day 6). Other targets not started. |
| W5 | Harvester itself on Elemental 3, with in-place migration | to be re-estimated | Under way | Three-node rolling migration prototype with live guest and rollback ran on day 2. |
Activity by task
All 52 plan tasks and their status, including those not started.
Proven in the lab, in review
In progress
Each dot is one verified result committed with its evidence (57 shown; hover for details). 22 of 52 plan tasks have activity.
Measured so far
Same cluster shape (3 control plane + 2 workers, 2 vCPU), same harness, x86-64, lab conditions. Mostly single samples.
| Journey | Rancher provisioning v2 | Elemental 2 | Elemental 3 |
|---|---|---|---|
| Create a cluster: user steps | 18 | 8 | 6 |
| Create a cluster: time | 25 min 30 s | 42 min 15 s | 27 min 03 s |
| Create in the Rancher UI: interactions | 18 | not measured | 12 |
| Upgrade Kubernetes only | 24 min 29 s | 36 min 45 s | part of the combined upgrade |
| Upgrade OS and Kubernetes as one action | not offered | fails (node left cordoned) | 90 min in place; 55 min by replacement |
| Works with no Rancher at all | no | no | yes: create, scale, upgrade, delete |
Known gaps
- Cluster creation still needs four lab workarounds for the Harvester Cluster API provider. Fixes are in progress on a fork.
- With default Rancher roles a tenant cannot create or see an Elemental 3 cluster. A fix is proven in the lab (day 7) and needs a small change in Rancher Turtles to ship.
- The in-place upgrade takes 90 minutes; the legacy Kubernetes-only upgrade takes 24.
- Only Harvester virtual machines are covered so far. Bare metal, vSphere and the public clouds are not started.
- Everything is proven in a lab on fork branches. No upstream pull request has been opened yet, by decision.