[Bug Report] Sandbox stuck in unknown state cannot be paused or deleted

[nullptr-error]

Summary

create sandbox with a unknown state. v0.7.0 cube.
if i delete the box, it. display error message : 操作失败: cubemaster error 130593: delete paused tombstone b2c270b5eca64a98a2fdb2fe555e0c42 ret=TaskStateInvalid msg=sandbox lifecycle operation is in progress; retry DELETE after 2 seconds
if i paused the box, it display error message: 操作失败: cubemaster error 130400: begin pause snapshot: sandbox b2c270b5eca64a98a2fdb2fe555e0c42 already has pause snapshot snap-dc6df6d5366345678c664d0c

Environment

• CubeSandbox version / commit:
• Host OS and kernel version:
• KVM info (modinfo kvm):
• Deployment mode:
• Relevant component:

Steps to Reproduce

1. ...
2. ...
3. ...

Expected Behavior

Actual Behavior

paste logs here

Additional Context

[cubesandboxbot]

This issue appears to overlap with existing reports about paused sandboxes getting stuck in an inconsistent state where delete/pause operations are rejected. Your symptoms (DELETE failing with 130593 / TaskStateInvalid / "sandbox lifecycle operation is in progress", and pause failing with 130400 "already has pause snapshot") closely match these existing issues:

• [Bug Report] DELETE paused 沙箱返回 130593 MasterInternalError,必须先 connect(resume)再删 #932: DELETE on a paused sandbox returns 130593 MasterInternalError and requires a connect(resume) first — same delete-paused 130593 error.
• [Bug Report] Pressure-test cleanup cannot batch-delete paused sandboxes: 130593 and OpenTap resume failure #987: Cleanup cannot batch-delete paused sandboxes — 130593 plus resume failures.
• [Bug Report] Intermittent failure to resume/delete sandboxes after long-running pause (VM state deadlock) #1274: Paused sandboxes intermittently cannot be resumed/deleted after a long-running pause (VM state deadlock where state is not rolled back) — this one was itself marked as a duplicate.
• [Bug Report] CubeMaster 状态不一致导致沙箱暂停/恢复死循环、僵尸沙箱及清理程序失败 #985: CubeMaster state inconsistency causing pause/resume deadlock, zombie sandboxes, and failed cleanup of paused sandboxes.

Please check whether your scenario is already covered by any of these (several are closed as fixed — if you can still reproduce on v0.7.0, adding a comment with your reproduction details to the relevant open issue, or clarifying what differs here, would help).

[luzhixing12345]

@nullptr-error Hi, Could you explain how this sandbox ended up in the "unknown" state? Did an issue occur during its creation (such as insufficient disk resources)?

Are there any key error messages or logs available, and could you provide a simple, reproducible example?

[cubesandboxbot]

This issue appears to overlap with existing reports about paused sandboxes getting stuck in an inconsistent state where delete/pause/resume operations are rejected. Your symptoms (DELETE failing with 130593 / TaskStateInvalid / "sandbox lifecycle operation is in progress" on a tombstone, and pause failing with 130400 "already has pause snapshot") closely match:

• [Bug Report] DELETE paused 沙箱返回 130593 MasterInternalError,必须先 connect(resume)再删 #932: DELETE on a paused sandbox returns 130593 MasterInternalError and requires a connect(resume) first — same delete-paused 130593 error.
• [Bug Report] Pressure-test cleanup cannot batch-delete paused sandboxes: 130593 and OpenTap resume failure #987: Cleanup cannot batch-delete paused sandboxes — 130593 plus resume failures.
• [Bug Report] Paused sandboxes permanently undeletable: leaked TAP devices cause resume to fail with ResourceBusy (435/500 orphaned TAPs) #1166: Paused sandboxes permanently undeletable (leaked TAP devices cause resume to fail) — stuck paused-sandbox cleanup.
• [Bug Report] Intermittent failure to resume/delete sandboxes after long-running pause (VM state deadlock) #1274: Intermittent failure to resume/delete sandboxes after a long-running pause (VM state deadlock, state not rolled back) — this one was itself marked as a duplicate.
• [Bug Report] CubeMaster 状态不一致导致沙箱暂停/恢复死循环、僵尸沙箱及清理程序失败 #985: CubeMaster state inconsistency causing pause/resume deadlock, zombie sandboxes, and failed cleanup of paused sandboxes.

Most of these are closed as fixed. If you can still reproduce on v0.7.0, it would help to clarify whether this is the same root cause or a distinct one — in particular, the pause failing with 130400 "already has pause snapshot" while the sandbox reports an unknown/tombstone state may indicate leftover snapshot state on an already-deleted sandbox.

As @luzhixing12345 noted, a simple reproducible example (especially how the sandbox first entered the "unknown" state) would help confirm whether this is a duplicate or a new regression.

[lkml-likexu]

@nullptr-error Pls feel free to reopen this issue if you have more detailed log.