You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug]: lume crashes in Virtualization's VNC server (_VZVNCServer → addAccessorObserver assertion) and the VM stops #8372
lume run crashed twice with EXC_BREAKPOINT (an assertion inside Virtualization.framework's private VNC server), and the running macOS VM stopped with it. Both crashes came within seconds of a VNC client connecting or disconnecting.
Reproduction
lume run <vm> --display none (VNC enabled, the default) for a macOS 27.2.1 guest, from a LaunchAgent and once with --detach.
Connect a scripted VNC client to the local URL from lume get --format json, take a screenshot or click, and disconnect. I used vncdo -s 127.0.0.1::<port> -p <password> move X Y pause 1 capture shot.png, a new connection each time, closed abruptly when the command ends.
Repeat. It is intermittent: in about a dozen connections over 30 minutes, lume crashed twice (both times in a VM session that had several connections already).
Expected behavior
The VNC session ends and the VM keeps running.
Actual behavior
lume exits with Trace/BPT trap: 5, asi: libsystem_c.dylib: assertion failed, and the VM stops (no "stopping VM" log line; lume ls shows it stopped). Faulting thread, queue com.apple.virtualization.vzvirtualmachine-accessor-manager:
The two crash reports are identical. Timeline of the first: VNC capture at 10:46:10, lume logged (Virtualization) VNC: Connection had error at 10:47:10, and crashed at 10:47:13. The second came right as a new vncdo connection opened (its capture never completed).
I don't know whether the trigger is a new connection, the earlier connection's teardown, or both. The stack shows the VNC server re-running _setupVirtualMachineAccessor and the accessor asserting on the observer it adds.
Environment
lume 0.7.2 (Developer ID build from the lume-v0.7.2 release)
Crash reports lume-2026-11-20-104713.ips and lume-2026-11-20-105431.ips (the faulting stack is above; I can attach the full reports if useful).
Workaround and additional context
lume run <vm> --display none --vnc disabled, driving the guest over SSH instead. No crash since.
Since the assertion is inside Apple's private _VZVNCServer, it may be a framework bug that only shows up under rapid connect/disconnect. Still, lume might avoid it, for example by keeping one VNC server for the VM's lifetime, or by not creating a new accessor per connection. Related: #3878 (an explicit bind address for VNC, since the listener is also reachable on the LAN).
Submission checks
I searched for duplicate issues and active pull requests.
This issue describes a single problem.
I removed credentials, private data, sensitive screenshots, and vulnerability details from this public report.
Thanks for the report and the crash details. We're looking into it. If you can, please share the lume version, the macOS version, and what the VM was doing when the VNC server asserted, so we can check whether it's tied to a specific VNC client or to display resizing.
Activity
f-trycua commentedon Nov 21, 2026
Thanks for the report and the crash details. We're looking into it. If you can, please share the lume version, the macOS version, and what the VM was doing when the VNC server asserted, so we can check whether it's tied to a specific VNC client or to display resizing.