Skip to content

docs(JDWLABS-433): complete truenas-csi PoC inventory, live status and fsck finding - #395

Open
jdwillmsen wants to merge 1 commit into
mainfrom
docs/JDWLABS-433-truenas-csi-poc-prep
Open

docs(JDWLABS-433): complete truenas-csi PoC inventory, live status and fsck finding#395
jdwillmsen wants to merge 1 commit into
mainfrom
docs/JDWLABS-433-truenas-csi-poc-prep

Conversation

@jdwillmsen

Copy link
Copy Markdown
Member

Summary

Read-only preparation for the truenas-csi migration PoC (ADR-0029 follow-up), recorded in the existing runbook docs/truenas-csi-migration-poc.md. No cluster state was changed; no PV/PVC/Secret was created.

  • Live PVC inventory on truenas-nfs/truenas-iscsi: six PVCs, not the three ADR-0024 counted. The "third" PVC ADR-0029 could not place is the Minecraft backup archive volume in the jdwillmsen tenant's external deploymentRepo; that chart has since also added two AFK-bot token-cache PVCs and a restore scratch PVC. Table includes consumer, data, criticality and PV driver.
  • Rollout status: csi.truenas.io CSIDriver, both -poc StorageClasses, controller and all five node pods healthy with 0 restarts; ExternalSecret SecretSynced (credential already seeded). ArgoCD Application is OutOfSync/Failed on the controller Deployment Recreate+rollingUpdate conflict, being fixed in a separate PR — noted as a pre-check.
  • Credential path: kv/truenas-csi confirmed as democratic-csi's; kv/truenas-csi-driver is the live distinct path — documented, no new name introduced.
  • fsck-on-stage: truenas-csi v1.2.0 runs no fsck of its own but stages iSCSI through k8s.io/mount-utils v0.34.1 SafeFormatAndMount, which runs fsck -a before every rw mount of a formatted device; the binaries are present in the pinned image (verified by exec into a node pod). Unforced and non-failing on "corrected", unlike democratic-csi's fsck -f -p, so the scratch probe still runs but is narrowed.
  • PoC target: Loki storage-platform-loki-0 reaffirmed against the full inventory; corrected the StatefulSet/PVC names in the cutover and rollback steps.

Test plan

  • Cluster reads only (get on PVC/PV/SC/CSIDriver/pods/ExternalSecret/Application; one which fsck exec)
  • Upstream source read at tag v1.2.0 (pkg/driver/iscsi.go, pkg/driver/node.go, go.mod, Dockerfile) and mount-utils v0.34.1 mount_linux.go
  • CI green

🤖 Generated with Claude Code

https://claude.ai/code/session_01B6FWEmAs2beQvKnX4LwWN2

…d fsck finding

Fill in the read-only preparation the migration runbook deferred: the full
live PVC inventory on both democratic-csi classes (six PVCs, not the three
ADR-0024 counted — the tenant's Minecraft chart has grown a backup archive
volume, two bot token caches and a restore scratch volume since), the live
state of the truenas-csi rollout (driver, classes and credential all up;
ArgoCD sync blocked on the controller Deployment strategy conflict being
fixed separately), the confirmed credential path, and a source-level answer
to the fsck-on-stage question: truenas-csi inherits an unforced `fsck -a`
from k8s.io/mount-utils on every rw stage, so the probe narrows rather than
disappears. Reaffirm Loki as the PoC target against the full inventory and
correct the Loki StatefulSet/PVC names in the cutover steps.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01B6FWEmAs2beQvKnX4LwWN2
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