Skip to content

checkpoint create fails with "sandbox does not belong to this organization" on exec-accessible, correctly-owned sandbox #609

Description

@phatblat

Summary

oc checkpoint create fails with "sandbox does not belong to this organization" on a sandbox that was just created under the same authenticated session, even though the sandbox is fully reachable via oc exec and its org_id matches oc whoami. The sandbox also never appears in oc sandbox list/oc ls, despite being live and exec-accessible.

Environment

  • oc version: 0.6.0.39 (macOS arm64)
  • oc whoami: confirmed org bd2db151-436c-48c5-961b-d85add39ec29

Repro

$ oc sandbox create --timeout 600 --json
{
  "sandboxID": "sb-e88df370",
  "status": "running",
  ...
  "token": "eyJhbGc..."   # decodes to org_id bd2db151-436c-48c5-961b-d85add39ec29, matching `oc whoami`
}

$ oc exec sb-e88df370 --wait -- echo hello
hello                      # exec works fine, immediately and repeatedly, over several minutes

$ oc sandbox list
ID           TEMPLATE  STATUS   CPU  MEM     AGE
                                                    # sb-e88df370 never appears, even after 30+ seconds

$ oc checkpoint create sb-e88df370 --name env-ready --kind disk_only
Error: sandbox does not belong to this organization

Retried the checkpoint create call after a 5s and again after a 20s delay (in case of index propagation lag) — same error both times, while oc exec against the same sandbox ID kept succeeding throughout.

Reproduction rate

Hit this on 2 of 2 sandboxes created and checkpointed in a single debugging session (a third sandbox, created earlier in the same session, did appear correctly in oc sandbox list — I didn't test checkpoint create on it before killing it, so I can't confirm whether it would have succeeded).

Impact

Not fully blocking for us — our workflow treats checkpoint create failure as non-fatal (prints a warning, continues) — but it means the "fault-tolerance baseline" checkpoint silently never gets created for a meaningful fraction of sandboxes, and a caller that does treat it as fatal would fail outright on a sandbox that's otherwise completely healthy.

Expected

oc checkpoint create should succeed on any sandbox that's currently exec-accessible under the authenticated org, or oc sandbox list should reliably list every live sandbox belonging to the org shortly after creation.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions