[1853/Engine Stability/Performance] Udon Exception Isolation and Recovery Observed (Log Included)
closed
Riria Nee-chan
Hello VRChat team,
As a long-time user, I would like to share a positive observation regarding the recent improvements in client[1853] stability, specifically concerning how the engine handles Udon exceptions.
I recently observed a scenario in
[［JP］ USiOPORT ウシヲポート]
where a Udon script encountered a NullReferenceException
and became Halted
. While NullReferenceExceptions
occur in various worlds, the client's behavior in this instance was particularly notable. While in the past such errors would often lead to instance-wide issues or a full client hang, I noticed that the engine now correctly isolates this exception.Despite the script failure, the client maintained stability, allowing me to continue essential tasks such as navigating the world and exiting via portals without any interruptions. Although there was a brief latency (client freezing) during the process, it is clear that the client's current error-handling mechanism successfully isolated the failing script and prevented the entire instance from crashing.
I have attached logs capturing this event. I am sharing this because it demonstrates that the client's current error isolation and resilience mechanisms are working exactly as intended, providing a robust user experience even when world-specific scripts fail. I appreciate the hard work put into these underlying improvements and hope this log data helps you further refine the client's error-handling capabilities, particularly regarding the latency during the recovery process.
Best regards,
Riria Nee-chan
---
I have attached the full log for the session. The 'Udon Halted' event occurred around
[2026.05.30 00:00:09 KST]
during my session in [［JP］ USiOPORT ウシヲポート]
. You can check the logs around that time to see the engine's recovery process and the subsequent latency.
Log In
White Lotuz
updated the status to
closed
Hey there, thanks for the positive observation. Regarding your micro-stuttering comment, after reviewing we found the information you shared to not be related to any micro-stuttering issues but we'll continue to monitor and test to keep improving client stability and performance. Thanks!
Riria Nee-chan
White Lotuz You're very welcome! I truly appreciate this update. With this refactoring, I’m genuinely looking forward to a much brighter and more robust future for VRChat!
Riria Nee-chan
I would like to provide an update regarding the micro-stuttering(aka freeze, brief latency) issue during world transitions.
After further investigation, I have confirmed that this issue persists on a clean, stock Windows environment, which indicates that this is an engine-level issue rather than a system-specific conflict.
Observation: The stuttering is significantly mitigated—and in some cases entirely eliminated—when disabling SysMain (Superfetch).
Analysis: This suggests the root cause is I/O contention.
During world transitions, the engine makes large, sudden data requests.
When the OS performs background I/O tasks (such as SysMain indexing or caching), it creates a bottleneck, resulting in the observed frame-time spikes and freezing.
Suggestion: Since this is an I/O scheduling conflict, I recommend implementing engine-level optimizations, such as asynchronous I/O processing or prioritizing critical asset streaming over background tasks.
Improving how the engine requests and buffers data would likely make the loading process much smoother, even in standard Windows environments where these background services are active.
I hope this helps the team isolate the cause.
Please let me know if you need any additional logs or system specifications.