search

vsg-dev / VulkanSceneGraph

Vulkan & C++17 based Scene Graph Project
http://www.vulkanscenegraph.org
MIT License
1.32k stars 212 forks source link

Threading issue with OperationThreadQueue::Latch #160

Closed mp3butcher closed 4 years ago

mp3butcher commented 4 years ago

Describe the bug MT Freeze (Latch stays locked after count reach 0)

To Reproduce video The bug emerged when i used Latch in vsg-dev:subpass_threading branch... Hard to reproduce, but launching lot of instances of the vgssubpass example yield application freeze in some windows (launch 6 instances and swap between windows seems to be a good way to provoke the bug)

Desktop

Additional context The issue might affect DataPager but harder to reproduce I fix it locally with https://github.com/vsg-dev/VulkanSceneGraph/commit/b4429c414c0fa10bc81e92e89dcbdec53dabda14 using a mutex instead of a condition_variable

robertosfield commented 4 years ago

The way you test multi-threading is my running multiple threads within a single application, not by running lots of separate programs. See a problem behavior when you are thrashing your window manager is a stress test of completely different kind. It could be that you are hitting a window manager issue, or a driver issue, or CPU bug or a memory bug.

The way to create a stress test for the VSG threads is to write a single application with multiple threads that contend over the same resources.

I'm closing this "Issue" as it's a dubious test, on an experimental branch that has no relevance to master. Any of the work that may make in from the experimental branch will be thoroughly vetted. If then we find problems we need to establish it with a single application. multi-threaded test.

mp3butcher commented 4 years ago

If it was a driver bug or a window manager issue my fix wouldn't work. But it does... You're bad faith

robertosfield commented 4 years ago

Threading problems are often a very difficult to pin down and fix. Some "fixes" can just avoid a race condition but aren't fixes at all. In this instance you "fixes" are sledge hammer hacks that change the conditions without actual pinpointing and fixing the actual bug if an actual bug actually exists.

What you are reading as bad faith is years of experience of dealing with threading problems and how it's most effective to deal with them. Me saying you aren't helping progress things forward on this is not some random statement. You are doing stuff that it's clear you don't yet have the experience and insight to tackle.

On Sat, 11 Apr 2020 at 09:46, Julien Valentin notifications@github.com wrote:

If it was a driver bug or a window manager issue my fix wouldn't work. But it does... You're bad faith

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/vsg-dev/VulkanSceneGraph/issues/160#issuecomment-612369153, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAKEGUGEYTQY5KVPMYDPKR3RMAU6NANCNFSM4MFOKD3A .