Skip to content

docs(alauda-os): correct data retention on Bare Metal host reinstall - #1440

Merged
chinameok merged 1 commit into
masterfrom
docs/bm-reinstall-data-retention
Nov 18, 2026
Merged

chinameok merged 1 commit into
masterfrom
docs/bm-reinstall-data-retention

Conversation

@chinameok

Copy link
Copy Markdown
Contributor

Summary

Corrects what the Alauda OS documentation says about data on a Bare Metal host when the host is reinstalled (version upgrade or node replacement).

Expected behavior (given by the product owner, to be confirmed by R&D): When a Bare Metal host is reinstalled, the provider does not wipe its disks. Data on the host's disks is not cleared.

The pages said the opposite: node-local data (hostPath volumes, local PersistentVolumes) was "Not kept", and only pool-managed persistent disks, which exist on Huawei DCS, Huawei Cloud Stack, and VMware vSphere only, were listed as an exception. This PR scopes the "not kept" statements to those three providers and adds what Bare Metal keeps. Statements about the IaaS providers are unchanged.

Relationship to #1409: that PR plans Bare Metal disks on the same basis (one system disk, no extra disk, upgrade does not wipe disks). This PR aligns the other pages with it. It does not touch #1409's branch or files.

Changes (before → after)

# File Before After
1 workload_clusters/alauda_os/index.mdx, data table One row: data on the node's own disks (hostPath, local PV) → "Not kept", for all providers Row scoped to Huawei DCS, Huawei Cloud Stack, VMware vSphere. New row: Bare Metal /var/cpaas on the system disk and data on other disks, such as raw disks used by LVM Storage (TopoLVM) → "Kept. The provider does not wipe the disks of a host when it reinstalls the host." New row: other data on the Bare Metal system disk → "Not guaranteed", because the OS image is written again
2 upgrade/overview.mdx, data table Same "Not kept" row for all providers Same split as #1, plus a pointer to the NoHostPathWorkloads exemption procedure
3 upgrade/pre-upgrade.mdx, warning before the provider tabs "On Alauda OS, data on a node's own disks … is not kept when the node is replaced" Scoped to the three IaaS providers. New paragraph for Bare Metal: the provider does not wipe disks; data on the system disk other than /var/cpaas is not guaranteed; the preflight still reports hostPath and node-local PVs
4 upgrade/pre-upgrade.mdx, execution host paragraph Files on a replaced node (/root, /home, Core Package) "are lost" "are not guaranteed to be kept" (still correct for every provider; the advice not to use a node is unchanged)
5 upgrade/troubleshooting.mdx, Bare Metal NoHostPathWorkloads row "…where it is not kept when the host is reinstalled" "No Pod keeps data on the disks of its node: …" (reason clause removed; check behavior unchanged)
6 workload_clusters/alauda_os/bare-metal/overview.mdx "Each time a host is reinstalled, its Kubernetes state is cleared." Adds: the OS image is written to the system disk again, the provider does not wipe the disks of the host, link to the data table. The original sentence was about Kubernetes state and did not contradict the expectation, but readers can take it as "everything is cleared"
7 configure/nodes/alauda_os/bare-metal.mdx, "Inventory recycling" "The OS is cleaned later … it runs the reprovision plan." "The OS is reinstalled later … the reprovision plan … writes the OS image to the system disk again and does not wipe the disks of the host."
8 llms.txt, llmstxt-state.json Descriptions and hashes of the changed pages Synchronized for upgrade/overview.mdx, upgrade/pre-upgrade.mdx, upgrade/troubleshooting.mdx, bare-metal/overview.mdx, configure/nodes/alauda_os/bare-metal.mdx

Found but not changed

  • workload_clusters/alauda_os/bare-metal/register_hosts.mdx (disk preparation) and install/alauda_os/bare-metal/prerequisites.mdx: the first installation wipes the install disk and stale COS_* partitions on other disks. This is the one-time registration, not the reinstall, so it is unchanged. See question 4.
  • install/alauda_os/bare-metal/deleting.mdx, workload_clusters/alauda_os/bare-metal/deleting.mdx: the user wipes disks when reclaiming hosts. Consistent with the expectation.
  • upgrade/pre-upgrade.mdx Bare Metal tabs and upgrade/troubleshooting.mdx (exemption procedure): the NoHostPathWorkloads check still blocks on hostPath volumes and node-local PVs on Bare Metal. The procedure is real product behavior, so only the reason sentence was changed. See question 2.
  • storage/storagesystem_ceph/installation/immutable_infrastructure_prerequisites.mdx (Bare Metal section): asks the reader to confirm that the provider does not format the OSD device. Consistent with the expectation, but still points to the provider docs for confirmation. See question 5.
  • networking/functions/configure_node_local_dns.mdx: kubelet configuration is lost when nodes are re-provisioned. This is OS configuration, not data.
  • configure/scalability/disk_configuration.mdx: owned by docs(install): add typical installation scenarios and a small Ceph baseline #1409.

Please confirm (R&D)

  1. Does the Bare Metal reprovision plan leave every non-system disk untouched, and is /var/cpaas on the system disk preserved? The wording "kept" in rows 1, 2, 3, 6, 7 depends on this. What exactly does the reinstall rewrite on the system disk (OS partitions only)?
  2. The NoHostPathWorkloads preflight on Bare Metal blocks the upgrade for hostPath volumes and node-local PVs, and its original text says the data is "not kept when the host is reinstalled". Why does the check exist if disks are not wiped? Should it stay, be removed for Bare Metal, or have a different reason (for example, the node can be allocated to another host of the pool)?
  3. When a pool has spare hosts, can a replaced node come back on a different host? If yes, data on the old host's disks is kept on that host but is not available to the node, and the table needs a sentence on it.
  4. Is the stale COS_* partition wipe in the first installation also applied at reinstall? If so, a disk with COS_* labels can be wiped at reinstall.
  5. Which data does a reinstall actually clear besides the Kubernetes state: container images and the containerd store, kubelet data, etcd data on control plane hosts?
  6. Is /var/cpaas on the system disk the correct example of preserved platform data, and are LVM Storage (TopoLVM) raw disks the correct example for other disks?

The provider does not wipe the disks of a Bare Metal host when it reinstalls the host. Scope the "not kept" statements for node-local data to Huawei DCS, Huawei Cloud Stack, and VMware vSphere, and describe what Bare Metal keeps.
@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 18, 2026 •

Copy link
Copy Markdown

Deploying alauda-container-platform with  Cloudflare Pages  Cloudflare Pages

Latest commit: 19c60dd
Status: ✅  Deploy successful!
Preview URL: https://16137e19.alauda-container-platform.pages.dev
Branch Preview URL: https://docs-bm-reinstall-data-reten.alauda-container-platform.pages.dev

View logs

@chinameok
chinameok marked this pull request as ready for review November 18, 2026 10:34
@chinameok
chinameok enabled auto-merge (squash) November 18, 2026 10:44
@chinameok
chinameok merged commit 7558c3d into master Nov 18, 2026
2 checks passed
@chinameok
chinameok deleted the docs/bm-reinstall-data-retention branch November 18, 2026 12:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants