Our game Combat Arena is experiencing an unusually high crash rate (subplaces Arena and Arena Touchscreen). This originally started on March 24th, and I filed a bug report here. After investigation by Roblox staff, it was determined to be an issue present in client versions 665 and 666, which was reported to be patched.
In spite of this, our crash rates have refused to drop and I have new reports for versions 668 and 669. Crash dumps are included in the attached private message. There is a different error in the analysis now, though it is still related to invalid pointers:
There’s an interesting behavior here that I’ve noticed - whenever I publish and migrate servers to the latest update, there’s a period of somewhere around 15 minutes where the crash rates return to pre-bug levels. Then suddenly they all spike together. This is consistent for every update I’ve published since the issues started.
Here’s a graph showing this effect. Included the session times graph so it’s easy to see when I migrated (there was a non-urgent followup update that I did not restart servers for, just ignore that)
I’ve recorded more crash logs from version 672 (current as of right now), that continue to report invalid memory access attempts.
(Moved upload to private message to protect personal info)
Based on current crash rates and average daily session counts, we’re seeing nearly 9,000 crashes every day due to this issue. I have no clues on my side on what the cause is so that I can work around it. Our player counts are suffering. I’ve been met with complete silence for nearly a month now. Is anybody out there?
Over the weekend, one of our developers got this error in Studio. I think it could be related given that the error message appears to be referencing a “bad weak pointer.” Maybe it’s a networking issue?
I’ve attached new logs for version 676 to the private message.
In our latest update I tried reducing any connections given to Maids for objects that were destroying at the same time (specifically the local character). It might have helped a little, but the problem is still definitely there. The chance of a crash on respawn seems to be reduced based on our own testing, but it still exists and happens eventually.
@ProgramDude@LightBeamRays I’ve been told by another Roblox employee that our crash data on the Roblox side points to rendering, specifically occlusion culling, as being the cause of the problem. Is there any chance I could get someone on your team to look into this?
I’ve also tried to repro the problem by setting the camera to a CFrame with NaN/max number components, but it seems like there are protections against doing that now anyways. Anything else I can try or investigate would be helpful. I can privately provide our camera code if that’s helpful to your teams as well.
We have play-tested your experience for many hours over multiple days with additional internal checks and validation, but unfortunately, we haven’t been able to find the conditions which cause the crash yet.
We do receive new logs and reports on crashes in occlusion culling with most (but not all) coming from your experience. Hopefully having more of them will let us find the root cause.
Sorry for the problems this is causing for your experience in particular.
It would be helpful to know more about the exact circumstances when the crash happens (map/location on the map/weapons/camera setup).
As for particular setups, in the last week I’ve had some willing members of the community share with me their current weapon / map / mode at the time of the crash. So far, I haven’t found any patterns. The only thing that appears to be 100% consistent is the timing, in that it seems to be a random chance at any point within the first few seconds of respawning.
