The live Concert transaction appears to reconstruct or re-register runtime-generated nDisplay/ICVFX actor components.
The transaction references generated component names that no longer resolve on recipient nodes.
The duplicate root-component registration warning suggests that component teardown/reconstruction is not fully balanced.
Stale or prematurely destroyed state may remain referenced by:
a pending Concert replication query future,
`FDisplayClusterViewportProxy` render-thread data, or
the UObject hash/Outer tables until incremental GC.
The eventual crash location depends on which subsystem touches the invalid state first.
The delay between the original transaction and the fatal error is consistent with a lifetime/use-after-free issue, but memory diagnostics or a source/debug build are needed to confirm that diagnosis.
Steps to Reproduce
1. Launch an nDisplay cluster from Switchboard using Unreal Engine 5.8.1.
2. Connect the control editor, camera nodes, and render nodes to the same Multi-User/Concert session.
3. Keep the nDisplay cluster running.
4. From the control editor, select the active nDisplay root actor/configuration.
5. Disable the inner frustum for `ICVFXCamera_A`.
6. Freeze the outer viewports.
7. Continue running the cluster.
8. Observe synchronized Concert transactions for `ICVFXCamera_A` and `DisplayClusterConfigurationData_0`.
9. Observe missing `DrawFrustumComponent` warnings and, in several runs, `Component already registered`.
10. After a variable delay, one cluster node crashes. Other nodes then exit because failover is disabled.
The first failing node and delay are not deterministic. Captured delays from the final relevant ICVFX/config transaction to the fatal error range from approximately 3 minutes to 34 minutes.
Hey Shaun, thank you for your reply. We are currently not on the partner build of the engine. For simplicity sake we’ve been using the launcher version of the engine. We took this route to allow us to use some pre-built plugins for the vanilla version of the engine. I’ll try testing the CL sometime this month. Thanks, Shaun.
Thank you for pushing through, Shaun. I’ll compile the nDisplay Plugin separately and test it. I think that is a more realistic move for us in this show. I’ll let you know by the end of the day how it goes.
I can confirm that compiling just the nDisplay plugin and moving it back to Unreal Engine 5.8.1 has fixed this issue. I verified the fix on our second Cluster for close to 2 hours without a crash! Thank you so much, Shaun.
Thanks, Shaun! Sorry for the delay in responding. This could be very helpful as we move forward. After this show wraps up in 2 weeks, I believe the best course of action is to return to the Partner build.