search

visit-dav / visit

VisIt - Visualization and Data Analysis for Mesh-based Scientific Data
https://visit.llnl.gov
BSD 3-Clause "New" or "Revised" License
438 stars 116 forks source link

Parallel Volume Rendering crash #19601

Open dpugmire opened 5 months ago

dpugmire commented 5 months ago

Describe the bug

I frequently get the following error when using the parallel compositing volume renderer in version 3.3.3. I've tried this in the latest version of visit and get the same results. I was able to reproduce this on a spatially decomposed BOV file created from noise.silo. The data I'm working with is 1k^3, so I've resampled noise onto the same sized grid. The behaviour seems to be random. Sometimes it crashes on the first render, othertimes it takes a few changes to the attributes to produce a crash. The DefaultRenderer works fine.

**The compute engine running on andes.olcf.ornl.gov has exited abnormally.

Shortly thereafter, the following occured...

Scalable Render Request Failed (VisItException) viewer: Obtained null data reader for rendered image for engine andes.olcf.ornl.gov. **

I'm attaching the session file and data below.

I'm running the client on a MacBook and the server on the Andes cluster at ORNL.

visitVRBug.zip

sambo57u commented 4 months ago

I can confirm this on Fedora Linux as well.

biagas commented 4 months ago

@dpugmire can you tell me the volume settings you used for 3.4.1 that showed the crash?

I'm trying to diagnose a crash I am seeing ... seems to be dependent on the number of processors for the parallel run. Just wondering if you are using the same settings: Parallel Renderer type, Parallel-per-rank resampling.

biagas commented 4 months ago

@sambo57u can you tell me what settings you are using in the Volume plot when it crashes? Can you reliably reproduce? If so, with sample data or only with your own?

sambo57u commented 4 months ago

Hi, this happened to me with the older version and Fedora39beta version but I cannot reproduce it with the version that was compiled on Fedora 40 with your assistance. I tried a number of different volume plots (time-dependent) and they all worked fine with parallel rendering.