Skip to content

feat: --odf trims resources but never disables features (floor 24 GiB) - #20

Merged
raghavendra-talur merged 1 commit into
TheEasyShift:mainfrom
raghavendra-talur:rtalur-odf-full-features
Aug 25, 2026
Merged

feat: --odf trims resources but never disables features (floor 24 GiB)#20
raghavendra-talur merged 1 commit into
TheEasyShift:mainfrom
raghavendra-talur:rtalur-odf-full-features

Conversation

@raghavendra-talur

Copy link
Copy Markdown
Collaborator

Summary

Policy change for --odf, following the maintainer's direction: trim, never disable. easyshift's goal is a production-like setup for developers — they give up production infrastructure (one node, one disk), not production features.

  • The StorageCluster no longer carries any reconcileStrategy: ignore: NooBaa (Multicloud Gateway), Ceph RGW object storage, and ODF monitoring all reconcile normally, floored by the recipe's existing 125m/128Mi component requests (which already covered noobaa-core/db/endpoint and rgw).
  • The single-node fit remains exactly what it was: floored resource requests, emptied placements, the SINGLE_NODE switch, the CephCSI Driver container floors, and the platform monitoring request-trim. Nothing else changed.
  • RAM floor: 19456 → 24576 MiB (8 vCPUs unchanged). The enabled components add ~1.3 GiB of requests to a node that was at 98%, and NooBaa's real usage runs 1.5–2 GiB above its floors. Consequence, stated plainly in the docs: --odf no longer fits a 24 GB host; a 36 GB macOS host or a 32 GB+ Linux host is the floor.
  • Docs and the design spec updated with the policy and numbers; Insights remains enabled (the disable during the earlier odfy debugging session was a one-off, never product behavior).

Test plan

  • make check green. Renderer tests now assert the absence of any reconcileStrategy: ignore and the presence of the object-storage component floors; the floor test asserts 24576.
  • --simulate traces all four ODF phases unchanged.
  • Not yet hardware-validated: the dev machine (24 GB) cannot host the 24 GiB VM. First real run on a 36 GB macOS host:
    ./easyshift create -n odffull --odf --master-disk 60 (the floor auto-raises RAM/CPUs; on Linux hosts no flags needed). Acceptance: StorageCluster Ready, Ceph HEALTH_OK, RBD/CephFS StorageClasses, noobaa/rgw pods Running, an ObjectBucketClaim binds, PVC write test passes.

🤖 Generated with Claude Code

Drop all three reconcileStrategy: ignore blocks from the StorageCluster:
NooBaa (Multicloud Gateway), Ceph RGW object storage, and ODF monitoring
now reconcile normally, floored by the recipe's existing 125m/128Mi
component requests. The single-node fit comes exclusively from resource
trims and emptied placements — developers lose production infrastructure
(one node, one disk), never production features.

Consequence: the RAM floor rises 19456 -> 24576 MiB (requests grow
~1.3 GiB and NooBaa's real usage runs well above its floors), so --odf
no longer fits a 24 GB host. First hardware validation of this mode is
deferred to a 36 GB macOS / large Linux host; covered here by renderer
tests (assert no reconcileStrategy: ignore, object components present),
the floor test, make check, and --simulate.

Assisted-by: Claude Code/claude-fable-5
Signed-off-by: Raghavendra Talur <raghavendra.talur@gmail.com>
@raghavendra-talur
raghavendra-talur merged commit 0521bc5 into TheEasyShift:main Aug 25, 2026
2 checks passed
@raghavendra-talur
raghavendra-talur deleted the rtalur-odf-full-features branch August 25, 2026 18:57
raghavendra-talur added a commit that referenced this pull request Aug 25, 2026
Tick completed items (Rosetta-from-first-boot, bake follow-ups merged in
paper #18), drop stale branch references, add the two open ODF items:
full-feature validation on a big host and the --odf-profile knob with the
measured request math.

Assisted-by: Claude Code/claude-fable-5
Signed-off-by: Raghavendra Talur <raghavendra.talur@gmail.com>
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