Skip to content

feat(deploy): use native XFS for Aliyun nodes - #48

Merged
tianyuzhou95 merged 1 commit into
inclusionAI:mainfrom
tianyuzhou95:albert/aliyun-xfs-filestore
Aug 31, 2026
Merged

tianyuzhou95 merged 1 commit into
inclusionAI:mainfrom
tianyuzhou95:albert/aliyun-xfs-filestore

Conversation

@tianyuzhou95

Copy link
Copy Markdown
Collaborator

Summary

  • attach a dedicated 300 GiB ESSD to Terraform-managed AKernel nodes and have ACK mount it as XFS at /home/akernel
  • use the native filesystem for sandboxd writable layers and checkpoints, avoiding a loop-backed filestore
  • apply the layout to regular extra node pools, reject conflicting CSI storage, and document migration and opt-out behavior

Validation

  • terraform fmt -check -recursive
  • terraform validate
  • make deploy-script-check

Scope

The current sandboxd release has no independent allocatable-cap setting, so reserving part of the filesystem exclusively for image and checkpoint storage remains a follow-up.

Attach a dedicated data disk to Terraform-managed AKernel node pools and
have ACK format and mount it as XFS at /home/akernel. Configure sandboxd to
consume that filesystem directly so writable layers and checkpoints share a
reflink-capable filesystem without a loop-backed filestore.

Apply the storage layout to regular extra node pools, keep Dragonfly pools
independent, reject conflicting CSI ephemeral storage, and document the
capacity, migration, and opt-out behavior.

Signed-off-by: Tianyu Zhou <albert.zty@antgroup.com>
@tianyuzhou95
tianyuzhou95 merged commit c9657c4 into inclusionAI:main Aug 31, 2026
8 checks passed
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.

1 participant